Under the golden arches
DOES IT ALL FIT UNDER THE GOLDEN ARCHES?: DY NAMICS OF
DISRUPTIVE HORIZONTAL PRODUCT DIFFERENTIATION WITH
CONSUMER HETEROGENEITY AND NETWORK EFFECTS
Hannah Rakoff, Jeroen Struben
Desautels Faculty of Management, McGill University, Montreal QC H3A 1G5 Canada
hannah.rakoff@ mail.mcgill.ca; jeroen.struben@ mcgill.ca
Abstract
This paper examines dynamics following introduction of nutritious food by a company well
known for high-motivational and low-nutritious products. Employing a system dynamics model,
we investigate how consumer dynamics affect uptake of the disruptive product. Our example is
the burger chain McDonald’s, which introduced salads, fruit and other healthier options in the
early 2000s. Focusing on consumer choice, we analyze the process of newcomers trying
McDonald’s and either becoming “core” customers, or not. The paper distinguishes overall
commercial success from that of the new product per se. We examine conditions that separate
commercial success (by drawing in new types of customers, whether these are the profitable ones
or whether they simply accompany more burger-eaters), neutrality (in which existing customers
simply change over), or failure (by alienating existing customers so that they abandon the
company). We focus on the role of heterogeneity in products and consumers, and on interactions
with social exposure-mediated network effects. We consider in detail the large and inertial
installed base of pre-existing burger eaters, and the degree to which its dominance is hard to
unseat, drawing parallels with reactions to other disruptive and ‘progressive’ products in
industries ranging from consumer products to electric vehicles to utilities.
Summary of results
When one looks at whether a new product launch will be successful, one needs to consider
overall profitability, rather than just sales of the new product per se. In this paper we look at two
factors that can affect this: first, the psychological ‘distance’ between the new product and the
original product, and second, network effects. Considering ‘distance,’ the model shows that
having a product that is very distant is in fact likely to offer some advantages over one that is
close. This makes sense according to standard consumer choice theory--by satisfying an unmet
need, a firm can draw in new customers and minimise cannibalization. It is important to
Under the golden arches
consider distance here in particular because this is the most ‘distant’ product McDonald’s has
ever introduced. In the simplified world of the base case, the model suggests success.
However, we need to look at network effects because customers do not make their purchasing
choices in a vacuum. Additionally, this may give us some insight into the perils of ‘blurring’ the
initial brand. Indeed, we see that if more- and less-health-conscious customers are allowed to
influence each other, this increases the important of the large pre-existing stock of burger eaters
and makes it harder to unseat their dominance. Take-up of salads is retarded and overall sales
reduced, although McDonald’s does still come out ahead compared to not introducing salads at
all. Interestingly, initial burger customers actually are retained a little bit longer than in the base
case. The dominance of their initial stock seems to protect them from the potential alienating
effect of a changed brand.
We postulate that such fine effects depend on the degree of homophily — that is, the relative
extent to which customers experience network effects with regard to other customers who are
more like them, rather than with regard to those less like them. Indeed, when we consider a
network effect scenario in which each customer group (More Health Conscious (MHC) and Less
Health Conscious (LHC)) looks exclusively or predominantly at what others in their group are
buying, McDonald’s has the best overall outcome in terms of total sales. Health campaigners
might also be happy; McDonald’s sells the most salads to MHCs in this scenario. However,
LHCs are left out of the health gains.
Introduction
This paper examines the dynamics of when a company, well known for a particular type of
product, introduces something new, which appears highly disruptive compared to its existing
brand. The example we examine in this paper is taken from the well-known burger chain
McDonald’s, which introduced salads, fruit and other healthier options around 2002. The paper
is ‘set’ in the years just before this launch, and uses this case as an example to consider all the
scenarios that might occur, rather than the particular one, which did. While in one sense the
paper sheds light in particular on how firms can profitably improve the mass accessibility of
healthier foods, the larger goal is to determine what the key factors are which a company
Under the golden arches
preparing to launch a ‘disruptive innovation’ should consider in order to affect the best overall
change in corporate profitability.
While at first it may seem that success or failure is understood from the intrinsic efficacy of the
disruptive product on its own, empirical evidence suggests that the study of success requires
more careful scrutiny. Increased overall profit might be achieved in many different ways. The
product in question could be a big hit, like those of the world’s biggest carmaker, which began
with Toyoda Automatic Loom Works’ 1933 foray into automobile production. But success could
also be achieved by the sheer novelty of the new item improving brand distinction and
recognition. Consider A pple, which, having put itself on the map with the introduction of the
graphical user interface in the 1980s, can now command substantial brand-based premiums for
their entire line of laptops. More subtly, the new product could provide arriving customers with a
‘feel-good’ factor regardless of what they actually choose to order (witness the advertisement of
optional higher-priced ‘green’ electricity by certain utility companies), or, in a related
phenomenon, the new product could make the vendor’s overall product variety more socially
acceptable, thus attracting larger, more diverse groups of customers. McDonald’s itself notes
that its Canadian Lighter Choices menu, introduced around the same times salads were
introduced to the U.S. market, seemed to appeal to moms, who ate them while bringing their
kids for Happy Meals (2002 McDonald’s Corporation Summary Annual Report (henceforth
referred to as the ‘2002 Report’)).
However, just as success can be multifaceted, there are also many ways to fail. Most simply, the
new product could sell poorly, and perhaps even tu off customers to the existing brand. For
instance, Coca Cola’s disastrous experience replacing traditional Coca Cola with New Coke in
1985 followed testing in focus groups in which 10 to 12 per cent of respondents reported feeling
angry and alienated simply at the thought of replacement of the sacrosanct beverage. They stated
they might cease drinking Coke altogether (Prendergast, 1994) While that case is clearly unique
in that the original product was completely withdrawn, the example does show not only the
potential for alienation, but also for strong responses tied to cultural solidarity. Once New Coke
was launched, the company initially saw sales rise, all to face growing resistance in the
American Southeast. Many former Coke drinkers who objected to the new formula were
Southerners who ‘viewed the company's decision to change the formula through the prism of the
Under the golden arches
Civil War’, seeing Atlanta-based Coca Cola’s choice to change formula as a second surrender to
the North. Although vocal opponents were a small minority nationwide, peer pressure kept
proponents of the change quiet in many areas, even stifling bottlers who were embarrassed to
tout the new product.’ In a striking example of the power of network effects to subvert a product
introduction, even when resistance apparently stemmed from a fairly small percentage, Coke
executives reintroduced Coke Classic just three months after it had been withdrawn.
This case also suggests that social exposure mediated the network effects that were in play: the
ability of a seemingly small percentage of strong objectors on a nationwide scale (just 10 to 12
percent) to nonetheless spawn a strong regional revolt, lends evidence to the hypothesis that
people are more affected by the buying behaviour of those more similar to them or with whom
they have more social interaction. This can lead to striking (and easily underestimated) ‘power’
stemming from the opinions and actions of a few crucially-connected customers (or ex-
customers). Furthermore, the ability of this strong regional revolt to lead to the national downfall
of a new product introduced by one of America’s leaders in the casual consumables sector,
suggests that McDonald’s would be wise not to underestimate this effect.
An example of more successful navigation of the challenges of updating and broadening a brand
without turning off loyal customers comes from famous Maine-based L.L. Bean. While it has
become more fashion-conscious over the years, particularly after reaching hard times in the mid-
1990s, the company is careful to stick to its roots of high levels of customer service and to pitch
its expansions as a wider implementation of outdoor pursuits throughout ones’ activities, rather
than to become a designer brand. For instance, a newly-launched skincare line included products
chosen to protect skin during outdoor pursuits, while a co-branded car marketed by Bean was the
Subaru Outback Limited Special L.L. Bean Edition. According to Bean’s entry in Gale
Contemporary Fashion, a scholarly compendium of entries on more than 450 fashion houses,
designers, and the like, Bean is conscious that a decisive move into the ‘designer’ world would
alienate its existing, loyal customer base. (Gale Contemporary Fashion) We see here a clear
example of active management of ‘product distance;’ that is, the degree of difference (in a
1 (Oliver, 1986, pp. 149-151, as cited in Wikipedia (http://en.wikipedia.org/wiki/New_coke, accessed 18 March
2011. Quotation from Wikipedia.).
Under the golden arches
consumer’s mind) between two products or groups of products, in this case, between existing and
new offerings
With McDonald’s brand profile in mind, a sobering story is provided by another company which
tried to take a ‘value’ brand into higher-end products. Bic, long synonymous in the public’s eyes
with disposable pens, single-use razors, and cheap lighters, acquired an existing watersports
company in 1979 and re-introduced the products under the Bic name in the mid-1980s. Bic may
insist on its website that an item in its surfboard line has ‘gained a reputation [as a]... versatil[e],
long-lasting and affordable surfboard’.” But the company’s entry in Wikipedia, which by its
collaborative nature is a unique window into collective impressions, leads off with the
unfortunate statement that Bic is ‘known for making disposable products including lighters,
magnets, ballpoint pens, shaving razors and watersports products’.’ And indeed, we have been
told that indeed it took a while for Bic Surf to catch a swell, because initially nobody wanted
what they perceived, from their prior image of the brand, would be a disposable surfboard.
When considered in tandem with the Bean’s example, the Bic case makes clear the care required
with product distance. While Bean’s wanted to minimise the apparent distance between old and
new products, Bic would have liked the new product to be more distinct in consumers’ minds
than it was. McDonald’s may be caught in a dilemma here: like Bean’s, they don’t want to
alienate existing customers, but like Bic, they want to attract new diners. Salad customers are
probably both drawn by the idea of consistency, quick service and affordability promised by
McDonald’s (Ray Kroc’s guiding mantra was “quality, service, cleanliness and value’), but are
also desiring something with a bit more ‘cachet’ —i.e., a slightly different brand image- than a
fast-food burger.
In this paper we examine product diversification dynamics, focusing on the role of network
effects, jointly with heterogeneous consumers in relation to the altemative and existing product.
In what follow we describe, first, our motivating example, McDonald’s Introduction of Salads,
and develop from that initial hypotheses. Then we describe the method and scope of the model.
2 (°7°9” Natural Surf 2,’ http://www.bicsportsurfboards,com/products/acs,3,61/7-9-natural-surf-2,480.html, accessed
16 March 2011).
> Wikipedia entry for Société Bic, http://en.wikipedia.ora/wiki/Soci%C3%A 9t%C3%A9 Bic, accessed 16 March
2011).
‘<The Ray Kroc Story,” http://www.mcdonalds.com/us/en/our_story/our_history/the_ray_kroc_story.html, (accessed
16 March 2011).
Under the golden arches
Next, we provide an exposition of the model and analyze this. We do this by first focusing on the
product distance, after which we also include network effects. This is work in progress, we end
with a discussion on furthering our hypotheses and next steps.
McDonalds and the Introduction of Salads
At the turn of the millennium, McDonald’s was in trouble. Y ears of fast expansion abroad had
led to an overextended network and disappointing shareholder return. In McDonald’s 2002
Summary Annual Report (‘2002 Report’), then Chairman and CEO Jim Cantalupo acknowledges
in his letter to shareholders (dated 21 March 2003) that ‘McDonald’s has lost momentum...and
lost what it takes to make customers feel special’ [ellipses in original]. He summarises his plan
to make the company great again by noting that ‘McDonald’s is in transition from a company
that emphasizes “adding restaurants to customers” to one that emphasizes “adding customers to
restaurants.” He presents a determination to ‘offer more products that provide the wholesome
choices and variety people are seeking’ as part of a ‘plan to attract new customers and encourage
existing customers to visit us more often.’ (2002 Report, pp. 1-2)
For the U.S., ‘premium salads,” including ‘fresh Caesar, California Cobb and Bacon Ranch
salads, topped with slices of warm, juicy grilled or crispy chicken breast meat,’ are presented as
permanent additions to keep the menu ‘contemporary’. In Canada, the Lighter Choices menu,
including salads as well as veggie burgers, Whole Wheat Chicken McGrill sandwiches, and
yogurt parfaits are being introduced to give customers ‘more choices’ and to respond to
‘changing eating habits’ (2002 Report, p. 25-26).
McDonald’s used to be known the world over for a hot, quick, consistent meal of burger, fries
and Coke — ‘billions and billions served!’ However, the fast food market isn’t what it was in the
days Ray Kroc discovered the McDonald brothers’ carry-to-car service in Califomia surfland: for
one thing, customers are beginning to worry about the health consequences of those burger meals
— films such as Super Size Me certainly haven’t helped — and other chains such as Subway are
now offering meals that are almost as quick, almost as low-priced, and (at least as advertised) a
good deal healthier. Government is pressuring for people to eat more healthily, and for
restaurants to offer healthier options.
Under the golden arches
The potential gains of a diversified menu are clear: McDonald’s could steal market share back
from Subway. They could beat more conventional rivals such as Burger King to the punch in
introducing a new product line. They could make McDonald’s, for the first time, a socially
acceptable option for those looking for a quick meal for a large and diverse group that includes
some health-conscious people, some vegetarians, and some people still looking for a good old
Happy Meal. More health-conscious eaters might be less price elastic, so that they could reap
better margins from those sales (Renaghan, M., Pers. comm., 31 May 2010).° Introduction of a
higher-margin product reduces the impact of any possible cannibalisation.® And McDonald’s
might be able to delay government intervention: the Economist, looking back from an (all-
knowing) 2010 vantage point, notes that, ‘‘[s]o far, fast-food firms have nimbly avoided
government regulation. By providing healthy options, like salads and low-calorie sandwiches,
they have at least given the impression of doing something about helping to fight obesity’ (The
changes facing fast food : good and hungry, The Economist, 17 June 2010).
But — there are risks as well. McDonald’s is known for burgers and McNuggets. They aren’t
known for crisp greens, broiled chicken, and fancy dressings with unpronounceable exotic
names. There is a risk that the diehard burger fans will begin to feel that their old standby has
forsaken them. This is not an unknown concem; as explained above, other companies have had
to tread very carefully in expansion and re-positioning of their lines to avoid alienation of
existing customers.
Hypotheses: Product distance and social exposure-mediated network effects
In this paper we develop hypotheses on underlying explanations of the various patterns observed.
Focusing on the point of customer decision and using a system dynamics model of the process of
newcomers trying McDonald’s and either becoming regular (‘Core’) customers, or not, this
paper investigates what factors can make introduction of an innovative or disruptive product go
5 Mark Renaghan is a principal of consultancy RMS, which has done detailed store-by-store pricing work for
McDonalds.
5 This stands in sharp strategic contrast to Burger King’s recent decision to emphasise its value menu. Success of
these offerings has unfortunately come at the expense of higher-profit items, with value options accounting for 20
per cent of all sales in June 2010, up from 12 per cent in October 2009 (The changes facing fast food : good and
hungry, The Economist, 17 June 2010).
Under the golden arches
in totally different ways. The product could become a commercial success (drawing in new
types of customers, whether these are the profitable ones or whether they simply accompany
more burger-eaters), neutral (in which existing customers simply change over), or a failure (by
alienating existing customers so that they abandon the company). The concept of the large and
inertial ‘installed base’ of existing burger eaters which would exist at the point of salad
introduction, perhaps exerting a dominance which would be hard to unseat, is considered in
detail.
Of particular note, while this paper looks at what is in some ways a new product, in other ways,
there is absolutely nothing new about salads — they’ve been around much longer than thin
mincemeat patties between sandwich buns have. W hat is new is the way and the setting in which
they are being sold (as a standardised, relatively processed product) and who is selling them (an
established fast-food restaurant not previously known for particularly healthy options). The fact
that most of what is new is just the positioning and business model for salads mean that
substitution is likely to be an important effect. Whether in terms of cannibalisation of
McDonald’s original sales, or in terms of sales won from (or lost to) other quick-serve chains
that have emphasised their health benefits for longer (for instance, Subway), substitution is a
phenomenon particularly liable to social exposure-mediated network effects.
We center our analysis, first, on the distance between the conventional and alternative products,
and, with that, on the effect of heterogeneity of potential customers. That is, we examine
dynamics where the alternative product may attract a different customer base. We focus on social
exposure mediated network effects as the central mechanism by which different customer types
are being attracted and get habituated in the consumption of, predominantly, one or the other
product.
Network effects have long been studied as demand-side drivers of adoption (see, for instance,
Katz and Shapiro 1986; Arthur 1989;). Network effects can refer to network externalities, a term
often used to represent the increased attractiveness of certain goods or services once more people
are using that product. See, for example, Anderson, de Palma and Thisse, 1992, who illustrate
the ‘importance of positive externalities’ (Anderson, DePalma and Thisse, 1992, p. 258) by
noting that the French government gave initial free access to the Minitel interactive system in
order to boost user base. As a subclass of network effects falls the effect of social influence. If
Under the golden arches
everyone around you is using e-mail, you will be more likely to do so. This is not just because of
the convenience, but because you may feel increasingly silly continuing to rely on postal mail or
the telephone. Similarly, if everyone around you at McDonald’s is ordering burgers, it becomes
harder to choose a salad.
Not everybody has the same influence on others. People tend to be more influenced by people
“like them”, whether status or value (Lazarsfeld and Merton 1954). That is, under social
influence, people are subject to forces of homophily. Furthermore, Christakis and Fowler applied
a social network analysis approach to longitudinal data from the famous Framingham Heart
Study cohort and reported in the New England Journal of Medicine that an adult’s chances of
becoming obese in a given time period increased if a friend, sibling or spouse became obese.
They conclude that ‘[n]Jetwork phenomena appear to be relevant to the biologic and behavioural
trait of obesity, and obesity appears to spread through social ties.’ (Christakis and Fowler, 2007,
p. 370). Following on from this, we hypothesise here — as we consider healthy and less-healthy
food choices that may be tied to obesity - that people are more likely to be socially influenced by
those whom they perceive as being more like them. So, if you are health-conscious and you see
another health-conscious person order a burger, that is likely to have a greater effect on you than
if you see a person who does not appear to be healthy-conscious choose a burger.
If we envision network effects as an example of social interactions layering themselves over the
purely technical and practical concepts of network externalities, then what we might term
‘relevance-selective’ social influence — social exposure-mediated network effects - is the next
logical step in uniting sociology with the ‘harder’ systems beneath. There is a good deal of
literature available on basic network externalities (one example is Anderson, DePalma and
Thisse, 1992, cited above) as well as on strategy in the face of network effects. For example,
Katz and Shapiro (1985) are concemed with the compatibility between different products, and
incentives to standardise, while Sun, Xie and Cao (2004) look at business model strategy for
innovating firms in the face of the network effect. However, our key focus here is different: to
capture the effect of homophily on consumer choice.
The network effects formulation presented here is structurally based on a related, but different,
formulation presented in Struben and Sterman (2008). In that paper, which deals with the
challenges of achieving ‘sustained adoption’ of alternative fuel vehicles (AF Vs), the authors use
Under the golden arches
weightings to represent willingness to consider a different type of vehicle from the one a driver
currently owns. The ‘willingness to consider’ factor ‘captures the cognitive, emotional, and
social processes through which drivers gain enough information about, understanding of, and
emotional attachment to a platform for it to enter their consideration set’. This is exactly parallel
to the network effect desired in the McDonald’s model, except that the AFV model captures the
relative distance between vehicle platforms based on their adoption patterns, while this paper
primarily focuses on the relative distance between two sets of consumers. While in the AFV
model network effect heterogeneity arose from different channels (e.g. non-drivers — drivers),
here we allow for variable network effects between the different types of consumers. Further,
and more importantly, the model developed here allows the distance between consumers, not just
between products, to be adjusted and the effects considered.
Scope and Method
We examine the decision to launch salads and the design of the launch as deriving from the
interactions within the system — running the gamut from salads being a way to drive sales of
conventional products, to the spectre of traditional customers decamping if they perceive their
restaurant changing too much. Through the standard method process (Hines 2001), we have
developed a model and a number of ‘thought’ experiments.
Of course, salads present a number of practical, logistical challenges to McDonald’s, as well as
the consumer-reaction issues mentioned so far. There are significant supply-chain challenges
involved in stocking fresh food in a restaurant that has heretofore gained significant advantage
from ease of handling processed food. This advantage is a key part of McDonald’s traditional
business model: according to Mark Renaghan, the consultant at RMS who has offered pricing
advice to McDonald’s, handling processed food makes national distribution easy and safe, and
therefore makes it easier to protect the McDonald’s brand (Renaghan, Pers. comm., 31 May
2010).While recognising the significance of operational and supply chain challenges, especially
for fresh food, this is not the focus here.
This model, rather, is about consumer choice, and about discrete consumer choice at that
(McFadden 1978). Therefore, we rely extensively on multinomial logit models. While powerful,
these can only offer an idealised representation of market shares. Consumers’ choosing food is
10
Under the golden arches
inherently a human activity, and furthermore a dynamic one — for many people, food is chosen
when they walk into the Golden Arches, based on factors around them at that moment. We
hypothesise, therefore, that taking network effects into account within consumer choice (Struben
and Sterman 2008) makes a significant difference in what factors can lead to profitability or
financial disaster after launch of a new product (or a product new for its setting).
We presume first and foremost that firms which launch new products are most interested in
improving their overall profits. Granted, there could be entirely different reasons for launching
new products — for instance, to stay a step ahead of government regulation, or to diversify
customer base to try to mitigate risk’— but as far as profit is at all concemed, we think it is
defensible to say that overall profit is — or at least will be over time - the measure which matters.
After an overview of the stocks and flows within the model, this paper covers briefly the main
components of the model, then tus in detail to network effects. A fter deriving the network
effect itself, based on total consumption of burgers and salads, we derive the network-adjusted
appeals and network- adjusted market shares, then close the network effects loop by explaining
how this influences development of loyalties of customers to one food or the other. We then
present our hypotheses in more detail and present the results of model runs testing them.
In this paper, we first consider the effect of (1) on its own, in a base model without network
effects. We then expand the model to take network effects into account, and examine the impact
of varying degrees of cross-group influence.
Model i: Core Customers and Product Distance
This section introduces the basics of how people and foods are handled within the model, and
then provides an overview of the main paths within the model.
7 With 20/20 hindsight, it has become clear that the decision to introduce salads may well have helped McDonald’s
weather the 2008-2010 recession. Burger King, which focused on young men (presumably, through a more
traditional fast-food menu), suffered as this demographic was hardest hit financially. McDonald’s, on the other hand,
with a more diverse menu and more varied clientele, including many women and older people, fared better. (Burger
King: Whopper to go, The Economist, v. 396 (4 September 2010), p. 72).
11
Under the golden arches
The basics
The model groups people into two customer types: Less-health conscious (LHC), and More-
health conscious (MHC). In addition, while McDonald’s offers a range of and less- and more-
healthy items, we consider the food products within two archetypal groups, denoted
‘foodchoices:’ BURGERS, which represents McDonald’s classic menu of burgers, McNuggets,
fries, etc.; and SALADS, which represents their newer, supposedly healthier products, such as
salads, fresh fruit, yogurt parfaits, and the like.
A number of variables in the model are indexed over customer types and/or foodchoices. An
array structure is used within Vensim to allow us to draw the model structure once, while
actually representing a number of indexed stocks and flows with any one model feature.
Customer type is always indexed using the letter c, while foodchoice is indexed with f.
The stocks and flows give the model its ‘inertia,’ making it difficult to move away from the
starting dominance of burgers. Figure 1 shows just the stocks and flows in the model:
Non ma sea — i — 9 Product Core
~ Rate of trying Tilers Core conversion Customers cf
MeD ec rate cf
4
Fast food Trier
exposure rate 4
Core Loss
Rate cf
=
fl] abandonment
A rate ¢
Abandonment
time constant c
Lees je
Figure 1: Stocks and flows in the model
McDonald’s can gain from introducing healthy products as new Triers are attracted, some of
whom become core customers (either using predominantly burgers or salads). However,
McDonald’s may lose core customers to competition as the product mix changes.To study the
underlying dynamics, we focus on endogenous factors including habitual and peer pressure
affecting dynamics of substitution of products (between healthy and less-healthy products) and
12
Under the golden arches
firms (McDonald’s vs. competitors). Index c refers to consumer type {More-health-conscious,
Less-health conscious}; f refers to product type {healthier; less-healthy} (i.e., {salads, burgers}).
Non-customers are, first and foremost, people who meet two conditions: (1) they are not current
Product Triers or Core Customers; and (2) they have either never tried McDonald’s, or if they
have and have then abandoned it, they have done so sufficiently long ago that they would
consider trying it again if the right product were offered. In fact, since the model presented here
leaves aside explicit theft of market share from a close competitor (focusing instead on implicit
substitution via arrivals and departures of customers), we need to put a third condition on Non-
customers as used here: they must be people who are not currently customers of any fast-food
restaurant.
When non-customers decide to try McDonald’s, they become Product triers. This occurs at the
Rate of trying. Product triers are indexed over customer type c, but not over foodchoice BURG
or SAL. The rationale is that people decide to try the restaurant without deciding necessarily
what they are going to eat when they get there. Triers can eat either foodchoice depending only
on the attractiveness, price, etc of the food and the interaction of those variables with their
customer type.
Product Triers become Core customers at the Core conversion rate appropriate to their customer
type, and at this point they also choose which foodchoice will be their core interest. Some
Product Triers instead give up on McDonald’s and become Lost Customers at the Trier
abandonment rate.
Product Core Customers do not necessarily stay that way forever. Every month a certain fraction
of them decide that McDonald’s no longer meets their needs and they also become Lost
Customers. This occurs at the Core Loss Rate. Alternatively, core customers can convert to the
other core product (Figure 2).
13
Under the golden arches
Product Core
et Customers ef I—x
Indicated Within Core
Core ef Conversion
See
Average Within Core
c Th Share ef
Pac / “ore
c sigpumption
Cum Core
Consumption Fractfon
Rate cf
Within Core
© Consumption Share
Within Core
Adjustment ef Consumption Share-ea = Within
oc “— off
Core Experience ike Raea
Adjustment Time pore
Figure 2. Core customer co-flows
Modelling discrete choice among restaurants and food options
We use a multinomial logit model based on the utilities of the options to represent the discrete
choice among them. ‘Pseudo-market share’ is so called because it does not represent
McDonald’s overall share of fast-food industry sales, but rather describes the fraction of non-
fast-food customers, who are exposed to fast food, who choose to eat at McDonald’s.® The
pseudo-market share for each customer type is calculated by taking the ratio of the anticipated
appeal of McDonald’s by newcomers of that customer type, to the sum of the appeals anticipated
from McDonald’s as well as from all other fast-food and non-fast-food eating options. °
5 Pseudo-market share is distinguished from unqualified market share (as the term is most commonly applied) in
several respects: (a) it is restricted to the share of a small subset of total possible customers; (b) it represents share of
people, not share of dollar or unit sales; and (c) it expresses McDonald’s share with regard to all eating options, not
Just fast-food choices.
Theoretical justification for this approach is provided by Cramer (2003); while he presents the multinomial logit
formulation slightly differently, the form used in this paper can be derived from Cramer’s formulation and his
references to McFadden on use of the utility function within logit models.
14
Under the golden arches
Four McDonald’s anticipated appeals (one for each foodchoice, for each customer type) are
calculated based on multi-attribute utilities (not shown in this schematic). The utilities include an
importance-weighted contribution from the price and quality newcomers expect McDonald’s will
provide, relative to other eating options, before they try it. Utilities are tempered by a ‘proximity’
factor, where we assume that burgers are fully proximate to LHCs and salads fully proximate to
MHCs, while the ‘crossover’ foodchoices have their utility multiplied by a fractional factor,
reducing the appeal.
“‘Pseudo-market share’ determines what fraction of non-customers will, once exposed to fast
food, become McDonald’s Triers.
Once customers are Triers, the model calculates the actual appeal, based on the real discovered
values of the price and quality attributes. Nice surprises or regrettable disappointments with
McDonald’s eating environment are also taken into account: the model scales the discovered
actual appeal by the relative pleasantness of environment, compared to that which Triers were
expecting. The frequency of riers’ visits to McDonald’s increases if this consolidated
discovered actual appeal is higher than the anticipated appeal, and falls if it is worse.
Additionally, for a given foodchoice f, the higher the discovered actual appeal, the higher the
fraction of Triers’ consumed units which are of that foodchoice (the Foodchoice fractional
market share, which is simply the ratio of that foodchoice’s appeal to the appeal of both
foodchoices).
Modelling consumption
Total consumption of each foodchoice by each customer type is the sum of consumption by
Triers and consumption by customers who are already Core. For Triers, the model calculates an
average personal consumption rate (for each customer type and each foodchoice) by multiplying
a standard Trier visit frequency by an average number of food units ordered per visit. This is
tempered, as explained above, by the relationship of the discovered appeal to that anticipated, as
well by the relevant Foodchoice fractional market share. The result is then multiplied by the
number of people currently in the Product Trier stock to yield the relevant trier consumption
(food units/month).
15
Under the golden arches
To calculate consumption by Core Customers, recall that each stock of Core Customers also
generates two consumption ‘streams:’ consumption of their Core food, and ‘crossover’
consumption, because even they occasionally choose the other kind of food (at a rate influenced
by the same ‘proximity’ factor mentioned above). Thus, for instance, calculation of the
consumption of salads by MHCs requires summing the core-foodchoice consumption by MHCs
who are Core for salads, as well as the peripheral-foodchoice consumption of MHCs who are
Core for burgers.
Conversion to a Core Customer
The real fundamental difference between Triers and Core Customers lies not in different visit
frequencies and consumption rates, but rather in the fact that Triers are not assigned to a
foodchoice, while Core customers are. In other words, while Non-customers make the choice to
try McDonald’s without specifying which product interests them most, the conversion to Core
carries inherent within it a sort of pledge of primary allegiance to one foodchoice or the other.
We postulate that in reality this decision is made as a single choice, not as a two-step process as
is more the case for Triers.!° People decide to go to McDonald’s more regularly because they
like a particular offering there, not just because they love the golden-arched shops. Therefore, the
rate of conversion to the stock of Core Customers needs to incorporate a double-choice within
one step (albeit represented by different terms): a decision to ‘go Core,’ and for a particular
foodchoice.
For each customer type, the overall rate of conversion to Core Customers (for all foodchoices
together) is given by an exponential function rising asymptotically towards the Customer
survival fraction, with a time constant of Trier conversion time constant. This is scaled by
multiplying it by the total number of Product Triers of that customer type. The share between
those going to the BURGER core and those to the SALAD core is then determined by
multiplying that overall conversion rate by the appropriate Network-adjusted market shares!'.
1° While there is a crossover consumption rate for Core Customers, this is secondary to the Core allegiance. It is
unlike FFMS which was the sole factor governing Triers’ product choices.
- Note that since the conversion rate is dependent on the number of Product triers remaining, the stream of
conversion to Core will dry up if no one new is coming to try the products (which is realistic).
16
Under the golden arches
The flows
Finally, three flows in the middle of the model, Core conversion rate, Trier abandonment rate,
and Core loss rate, are shown. Looking first at dynamics surrounding Triers, note (see Figure 1)
that there are two routes out from each Trier stock: Triers can convert to become Core customers
of either burgers or salads, or they can abandon McDonald’s entirely and go to the stock of Lost
Customers. However, a certain percentage simply remain as Triers, continuing to go to
McDonald’s now and then, consuming at the Trier rate and choosing, on average, whichever
product the Foodchoice fractional market share for Triers directs them to.
The shares of Triers in these three groups as the model approaches steady state are governed by
several fractions. The fraction of those converting to Core Customers (of both foodchoices taken
together) rises toward the Customer survival fraction (indexed over customer type c,
endogenous, and explained in more detail below when we return to the Core conversion rate).
The fraction of those remaining as Triers indefinitely is governed by the Perennial trier fraction
(also indexed, but exogenous); the fraction of Triers remaining falls over time according to a
time constant which is not defined explicitly in the model but which is clearly a function of the
time constants for the two routes out of Trying. Finally, the fraction of those lost entirely to
McDonald’s simply approaches whatever the remainder is: the Less fraction. jg given by
(1 — Customer survival fraction, — Perennial trier fraction.) The model assumes that all three
fractions are approached asymptotically according to a single-order process.
The Core conversion ratef (hereafter abbreviated as cor ), which gives the rate of conversion of
Triers of customer type c to Core Customers of foodchoice f, in units of people/month, is defined
as
° This appears to be an goal-seeking formulation, where the fraction of Triers on each route ‘seeks’ the appropriate
fraction — i.e., converges on it. The presentation here is somewhat different from the simple examples of goal-
seeking behaviour shown in Chapter 8 of Sterman (2000) (see, for example, p. 277), in that we discuss the rates in
terms of outflow from the stock of Triers rather than in terms of inflow into the Product Core Customers or Lost
Customers stock, and the discrepancy between the current state and ‘goal’ state is not explicitly shown. However,
the underlying concept is the same.
7
Under the golden arches
CSfrace ie
Core conversion rate! =——-~——* _.
© ~ TCTimeConst
(1]
where
e CSfrace is the Customer survival fraction for customer type c (see explanation below)
e TCTimeConst is the Trier conversion time constant. In the current version of the model,
the details of how this constant is arrived at are not our main focus, so for simplicity we
assume this to be an exogenous constant and one which is the same for all customer
types and all foodchoices. (The section on ‘Further avenues for exploration’ towards the
end of this paper lays out some ideas for how this constant could be represented in a
more sophisticated manner, particularly with respect to differences between how good
McDonald’s is compared to what expectations were.)
5f
e Finally, % is the network-adjusted market share for foodchoice f among converting
customers of customer type c
The CSfrac. is not exogenous. Rather, logically it depends on how much Triers like
McDonald’s. To represent this, the model uses, as a first approximation, the product of the
“pseudo-market share’ of McDonald’s for customer type c, and Trier visits relative to expected
for c:
CSfrac, = of? «TVrelE, 121
This makes sense since it represents scaling the fraction of people who try McDonald’s, by the
factor representing their relative pleasure or displeasure once they get there, and hence, their
likeliness to continue to ‘embed’ McDonald’s further into their lifestyle. Additionally, since by
using TVrelEc, which incorporates Relative pleasantness of environment, rather than just the sum
of the discovered actual appeals of the individual foodchoices, we take the effect of McDonald’s
environment into account.
It could be argued that the CSfrac should be higher, since at the point of conversion to Core, we
are dealing not with the general set of Non-customers but rather with people who have already
shown themselves to have some affinity for what McDonald’s offers. However, a counter-
18
Under the golden arches
argument would hold that it is one thing to use to temper Trier consumption, and quite another to
say that, for instance, discovering that McDonald’s is better than expected is enough to increase
by the same proportion a Trier’s probability to convert to Core, rather than just to increase the
frequency of his Trier visits.
Therefore, for this version of the model, Equation [2] is dimensionally consistent (all terms are
dimensionless) and makes an adequate starting point.
Next, for each customer type c, the Trier abandonment rate is given by an exponential function
rising asymptotically towards the Loss fraction, with a time constant of Abandonment time
constant. This is multiplied by the total number of Product triers of that customer type:
Lfrac,
Trier abandonment rate, = ATimeConst.*
‘imeConst,
PT,
[3]
where
e Lfrac, is the Loss fraction for customer type c, as described above, and
e ATimeConst, is the Abandonment time constant for customer type c, which is here
exogenous
e PT, is the number of Product triers of customer type c
Tuming to the last flow in the model, the Core loss rate (indexed over customer type and over
Core foodchoice) is modelled as a simple first-order process. The number of Core customers lost
per month is given by multiplying the number of people currently in the relevant Core stock by
an exogenous Monthly core loss fraction (which is also double-indexed).
Analysis of the Base Model - before network effects
We begin with the analysis with a base case with moderate product proximity (p=0.5) (Figure 3).
The model begins in equilibrium. After 12 months (time=0) salads are introduced. The trier stock
increases rapidly, due to a growth of health-conscious Triers. Most of the Triers become core
19
Under the golden arches
salad eaters. In addition, Less- Health-Conscious consumers convert from burger eaters to salad
eaters.
Product Trier ¢ Product Core Customers cf
400 = 6,000
=e x
it Less Health Conscious (LHC)
Se 4,500 PSS eye
5 ee | at
g jti ttt tg 5.000 SS
fs TT More Health Conscious (MHC) * _MIHC, Salad
250 1,500 == i
200 0 ATT L TL TM farer|
120 12 24 36 48 60 72 84 96 108 120 «12 24 «36 48 60 72 «84 «96 «108
Time (Month) Time (Month)
Relative Consumption
Total Consumption
~__|_ LHC, Burger,
0375 a a |
jaae= MC, Salad TAG, Salad
0 pS +++ MC, Burger|
120 «12 2 36 48 60 72 84 96 108
Time (Month)
Figure 3. Base Run
We now look at the effect of product distance on consumption (Figure 4). Introducing a product
with high proximity means that the products are full substitutes (Figure 4, left). The new product
is hardly different from the old product, and sales converge so that each product receives about
50 per cent of the sales from each customer type. Low proximity between the product choices, on
the other hand, means that consumers are less likely to switch. In this case, consumers are less
likely to cannibalize sales of the other product, since each consumer category is mainly
motivated by the product choice which is ‘closer’ to their customer type (Figure 4, right). In
other words, when the food choices are further apart from each other, the distance between the
customer types becomes more significant. The large stock of latent health conscious eaters
results in a rapid growth of the salad eaters. While we observe a small overshoot of health
conscious salad eaters, eventually the model equilibrates at the equal consumption level where
both consumption groups are fully satisfied. Overall sales equilibrate well above initial sales.
20
Under the golden arches
Relative Consumption Relative Consumption
1s a
1.125 | Total Consumption [Total vaae
tHe, Bee |
B o7s
LHC, Burger ee a a
, Barge _-— MIG, Salad
pe Berge
att} tT
LHC, Salad-=> 7 LHC, Salad
6 MHC, Salad . [mie, = =F
12 0 12 24 36 48 60 2 84 96 108-12 0 iz 24 36 48 60 72 84 6 108
Time (Month) Time (Month)
Figure 4. Relative consumption for high (left) and low (right) proximity.
These results are consistent with those from standard consumer choice theory. If additional
products are being introduced that do not satisfy additional, unobserved, preferences, demand is
not increased. Rather, each products receive an equal share of the market. In contrast, if we
expand an heterogeneous choice set, preferences are increasingly satisfied across the population
and sales increase. In more specific terms here, since salads have a lower proximity to burgers
(in terms of perception, niche, market positioning, etc) than did, for instance, the Chicken
McNugget, we can see that — at least before network effects are considered - McDonald’s is
likely to have a different experience when they introduce the ‘healthy’ menu than they may have
had in past menu expansions.
Model Expansion: Network Effects
This section first introduces network effects, and then explains in detail how they have been
applied to this model. We base our formulation on that applied for social exposure in Struben
and Sterman 2008, despite it being a somewhat different process (some adjustments are needed,
as explained below). Additionally, a hidden benefit has come from the fact that the algebraic
expressions used here are not simple to implement in Vensim: implementation has required
thinking very carefully about the meanings of the terms in order to see them as composites of
simpler building blocks. While conversion to a Core customer is the only part of the model
21
Under the golden arches
which explicitly and directly taking into account network effects, the influence is felt more
broadly because of the interacting nature of the feedback loops.
This a key difference with the simple model is that peer pressure network effects may lead
consumers to eat products that are distant from them. At the same time, network effects may
deter core customers. Here we modify the consumer choice equations to include network effects.
The Network effects loop
The Network effects loop is shown in Figure 5. Trier consumption (not shown in the figure) and
consumption by Core customers feed into Total monthly consumption. From there, we have the
Network effects loop, a reinforcing loop which shows the effects of social influence on choice of
food: it acts on the customer-type- and foodchoice-specific Core conversion rate, encouraging
more Triers who are in the process of converting to Core customers, to pick that foodchoice as
their Core.
Weighted sum of ee
all consumption bia! monthiy
observed ef coangeloa ag C
consumption ‘onsumpti
Network Triers _ observed of chaarvation
effects ef Exponent
riers
i
Total monthly Sg og
a consumption ef 2
Triers Network
adjusted appeal cf
nA : Core
aie eae consumption ef
adjusted market share Peripher
of r eriphera
Fr consumptio
‘ nef
INPT ival F IN PCC ee IN
ke > Product ps Product Core wa >
Trier ¢ Core Customers cf be-J
Fuying convSrgjon rate : Core Loss
Figure 5. Network effects on triers. Similar dynamics act on core customers. Core customers can
also defect if ‘alienated’ by the presence of too many consumers from their non-core products.
22
Under the golden arches
Note that network effects do not necessarily increase the overall conversion rate; rather, for a
given foodchoice, if everything else is held constant, a stronger network effect for that
foodchoice will increase the fraction of conversion which goes to that foodchoice."?
Of course, we do expect this network effect to increase absolute consumption of a given
foodchoice: once there are more Core customers of a particular foodchoice, the consumption of
that product rises. More Core customers mean more Core consumption; while it also means more
Peripheral consumption, that effect is less, because the Core consumption fraction will usually
mean that the large majority of consumption by Core customers is of their Core food.
We now explore the network effects in more detail.
Applying network effects to the McDonald’s model
In comparison with Struben and Sterman 2008, we need to make some adjustments to the use of
weightings within the calculation of network effects, since we are representing the perceived
distance between two sets of consumers, rather than a product from a consumer. So, we need to
use weightings to temper the effect of observed consumption by other consumers, rather than
using them directly on the appeal of a product. Only after taking the ratio yielded by the quotient
of the weighted sum of all observed consumption of the foodchoice in question, divided by the
sum of the weighted sums of all of the available foodchoices, will we have the network effect in
a form parallel to that of an appeal function. Then we will we be able to multiply that whole
factor - the network effect - by the raw appeal in order to create our network-adjusted appeal.
From that point, deriving network-adjusted market share is straightforward.
Calculating the network effect itself
We begin with the example of considering the network effect on the appeal of burgers to
customers of the More Healthy Customer (MHC) type. They are influenced more heavily by
burger consumption by other MHCs, although seeing LHCs eat burgers also makes them more
likely to choose that option. The more people they see eating salads, the less likely they are to
choose burgers, although again, consumption by their own customer type has a greater influence.
The numerator of the network effect is given by summing, over all the customer types, the
13 Because the network effects do not increase the overall conversion rate, there is not even the weak balancing loop
which would arise if greater conversion pulled people out of the Trier stock and thus decreased Trier consumption.
23
Under the golden arches
products of the weighting representing perceived proximity of that customer type to MHCs,
times the consumption of burgers by that customer type. For the denominator, where Struben and
Sterman 2008 summed over all of the possible vehicle platform options, we need to have the
weighted consumption of burgers plus the weighted consumption of the other option(s) — here,
salads. Thus we have
We have
NEwuc,purc
_ (Wuncmuc * CONSMme + Wancnc * CONStHC”)
(Waucmuc = CONSHe! + Wuc.uc * CONSEYRG) + (Wac.muc * CONS Mic + Wuuc.uc * CONS HAO)
[4]
where
e¢ = NEvwuc,surg is the network effect on consumption of burgers by MHCs
¢ = Wwucmuc is the weighting representing relevance of consumption by MHCs as
perceived by other MHCs
© = Wwuc.nc is the weighting representing relevance of consumption by LHCs as perceived
by MHCs
© CONSEYRS, CONSBYRS, CONSSAL., and CONS}A and are total consumption rates
(units: food units/month) where the subscript is the customer type consuming and the
‘gfood consumed yy
superscript is the food consumed (i.e. CONSg-oup consuming
Generalising this to give the network effect on consumption of foodchoice f by customer type c
gives us:
24
Under the golden arches
Yo(Weo * CONS)
NE cr > Ty (Was * CONS!)I
(5]
where 0 is the subscript for the customer type being observed consuming, and c is the subscript
for the customer type that is observing that consumption and being influenced. (In the Vensim
model, these are denoted CusTypObserved and CusTyp, respectively.) Therefore, W,,,is the
weighting representing the relevance of consumption by customers of customer type 0 as
perceived by those of customer type c, and CONS/ is the total consumption rate of food f by
customer type o.
We assume in this model that people find the consumption habits of other customers of the same
customer type to be entirely relevant (i.e., to have a weighting of 1) while they find the relevance
of consumption of customers of the other type to be somewhat less. Therefore, we set W... = 1 if
c=o, and otherwise W., = a, where 0 < a, < 1. Note that this is exactly parallel to the
definition of r/ used earlier. While we used that to define the proximity of foodchoices to people
of a particular customer type, we now use the same structure to define the relevance of customer
types to each other.
In matrix terms, we have
cus. type observed
LHC LHC MHC
W.o = cus.type 1 Qh
a 1
myc o MHC
{6]
Equation [5] is clearly in a form parallel to that of an appeal function, since it takes a calculation
relating to the foodchoice in question, and divides it by the sum of the same calculations for all
the possible foodchoices. So, we are now ready to multiply the network effect by the raw appeals
to create our network-adjusted appeals, and then our network- adjusted market shares.
Creating network-adjusted appeals and market shares
Since we are dealing with people who have already tried McDonald’s, it makes sense to use the
discovered actual appeals, rather than the anticipated appeals, as the raw appeals for these
25
Under the golden arches
calculations. Taking again as an example the case of the appeal of burgers to customers in the
More Healthy Customer (MHC) type, we have network-adjusted appeal given simply by
=BURG — + BURG
G@uuc’ = NEmuc,eurc * @uuc
7]
(Recall that is the discovered actual appeal of foodchoice f to customer type c; we denote the
network-adjusted appeal by putting a tilde mark above the a.)
Generalising this yields the network adjusted appeal of foodchoice f to customer type c as
af = NE, * ab
[8]
Finally, we can calculate the Network-adjusted market shares. Like the ‘pseudo-market share’
used for calculating the Rate of Trying, this market share also has a very specific meaning: it is
the share of Triers of customer type c who convert to Core customers, who choose to become
Core for a particular foodchoice f. (Triers who do not convert to Core customers are already
removed before this calculation.)
We have the Network-adjusted market share of foodchoice f for customer type c as
[9]
In parallel to the market shares calculated based on raw appeals, the network-adjusted market
share of foodchoice f among customers of customer type c is given by the ratio of the network-
adjusted appeal of foodchoice f for those customers, to the sum of the network-adjusted appeals
of all of the foodchoices available (here, only two: burgers and salads).
We saw within this section how Trier consumption and consumption by Core Customers are
calculated, in order to yield Total monthly consumption. We have just explained how this
consumption tempers the discovered actual appeal via network effects to yield the network-
adjusted appeals and market shares. In the next section we close the Network effects loop by
26
Under the golden arches
explaining how the network-adjusted market shares influence the rate of conversion of Triers to
Core Customers.
Simulation results: Network effects
Figure 6 shows the role of Core Customer network effects on consumption by consumer and
product category, with a proximity of 0.25. In the left-hand graph, network effects remain off
and results as before, for the low-proximity case. In Figure 6 (right), we switch network effects
on for both Triers and Core customers..
Relative Consumption Relative Consumption
1s a
a | leet
1125 fetal eksompton rots eons
* LHC, Burger [PRG Bier
2 07s ~— —
= =a
itt | tit} tt |
Cee === Sas
9.375 “NHC, Salad
eet hala _-— MG, Salad
LG, Salad = LHC, Salad
|tne, ye Mic, burger eq
“120 «12 24 36 48 60 72 84 96 108-12 0 12 2 36 48 60 72 84 9% 108
‘Time (Month) ‘Time (Month)
Figure 6. Network effects for moderately proximate products (proximity is 0.25, otherwise
equal to Base run values). Network effects are switched on for triers and core customers on
the right graph
When network effects are considered, we see that salad eaters tend to be selected out. Burgers
dominate the scene and few people receive a signal to go to McDonalds for the salad. This
difference holds in particular for health-conscious salad eaters, who reach steady state at about
47 per cent of total initial consumption levels, as opposed to almost 55 per cent without network
effects. In addition, while total McDonald’s consumption still rises, it rises less.
Homophily effects
The graphs below show the results for variation in the strength of network effects across
consumer groups. The parameter o. [0,1] captures the degree to which cross-consumer type social
pressure is smaller than within-consumer type. We observe that reducing the impact of network
effects across social groups increases sales (as well as total consumption, see NE, RCCI=0). In
equilibrium, sales even increase beyond the case of zero network effects (and full information).
27
Under the golden arches
Total Relative Consumption Relative Consumption of Salads by the MHC category
2 0.6 a
NE,Recleo—7] |
NE ReC0s |
LT 0.45 eee
sie
by a
Ne, Rect = o A
ous I 3 Retwork Effects (NE),
A ee No Network / Relative Cross-Category Interaction (RCC!) 41
<=Metwork Effects (NE), effects: | //)
1 Lio Relative Cross-Category interaction (RCCI) = 1 |0.15
No Network
08 [effects 0
12 0 12 24 36 48 60 72 84 96 108 -I2 0 12 24 36 48 60 72 84 96 108
‘Time (Month) Time (Month)
Figure 7: Consumption with varying relative cross-category interaction (RCCI)
Why do we get increased adoption for the asymmetric case? The new product (salads) gets the
benefit of social exposure through peers, while existing burger customers (who were
predominantly of the LHC customer type) are not ‘frightened off by this change. Hence,
homophily helps by introducing the differentiated product to new customers.
Network Effects: Planned tests and sensitivity
We have performed sensitivity analysis on a range of parameters. The model allows assessing
the parameters to which the model behaviour is most sensitive to. We performed initial analysis
on the effect of social pressure on defection (not shown in graphs). We found that, absent other
network effects, but allowing for defection, critically reduces McDonald’s sales. However,
including sufficiently strong network effects protects the dominant core product, and thus
customer, making McDonald’s and the core product robust against such consumer responses. In
this case, salads become a small niche product. In future analysis we will be we will expand on
this.
Only the factors which are very important are those on which it is worth spending significant
time and money to collect more accurate data. A main question is whether the large ‘installed
base’ of burger-eaters overwhelms the attempts of others foodchoices to ‘muscle in’ on its
dominance. That said, we would like to know how sensitive this is to certain factors. Sterman et
al (2007) reports on the underestimated difficulty of changing an inertial system in which
anthropogenic has been building up over the past 160 years or so since industrialisation began
.Similar, in the automotive industry major challenges in changing vehicle-propulsion types (from
internal-combustion to electric) derive from its inertia; this is a system in which vehicles last
28
Under the golden arches
perhaps a decade or more. Those are, respectively, systems with a really, really long time
constant and merely a long time constant. Changing of eating habits may not involve time
constants that are of equal length, however, social pressures and internal inertia (Dube 2010)
make shifting markets a big challenge. Not only would we like to know if burgers overwhelm
other foodchoices, but also whether varying the time constants affect this.
Tests of the sensitivity of the two parameters representing, respectively, the distance between the
customer types and the distance between the two foodchoices, would be worthwhile because of
the unfortunate difficulty in estimating them. Not only are these parameters very difficult to
reason ‘out of the blue,’ but collecting empirical data on them would require a sophisticated and
expensive market research study. Before devoting undue worry or money to this problem, we
would like to know how important the two parameters really are. The model dynamics are
mainly driven by competing positive feedback between food choices. Understanding in tipping
points requires gaining a deeper understanding of at what densities networks have particular
strength. That is, we may not assume linearity of these effects, and here with competing effects,
understanding how and to what extend gains of the positive feedbacks shift seems critical.
Representing these in more detail may reveal much stronger tipping points than has been done
here.
Although the rate of becoming a product trier (Rate of trying McD) is not foodchoice-specific,
trying does nonetheless occur at different rates for different customer types. Currently this is
because of the different weightings of the attributes leading to the appeals. However, it would be
interesting to include product variety and rate of product innovation as influencing factors, as
those might affect LHC and MHC people differently. For example, a lot of quick innovation on
McDonald’s part to change the menu rapidly from burger-heavy to salad-heavy would be
expected to result in more MHC people trying the restaurant. Later on in the model, therefore,
we would expect to see more salads being consumed. Quicker innovation might also ‘shock’
more of the initial ‘installed base’ of burger eaters to abandon McDonald’s. Struben and Sterman
2008 find that efforts to convert to alternative fuel vehicles need to be supported for a long time
— self-sustainability won’t come for several decades. This is primarily because of the length of
time that people own cars, before having the opportunity to make a choice regarding a new one.
While decisions to begin frequenting a new fast-food restaurant can be made much more quickly,
29
Under the golden arches
it might still be interesting to see the effects of varying the time constants. We might represent
more explicitly the effects of advertising and word of mouth, and we would be interested to
know how long we needed to keep up the advertising, or how significant the fallout from a
strategic mistake causing a brief spurt of bad word of mouth might be expected to be. When
considering these questions, we would be interested in knowing how sensitive the model is to the
time constant, so that we could decide how much money and effort it is worth to collect data on
it.
One of the most intriguing suggestions made for possible outcomes to the introduction of salads
is that it will, in fact, increase burger sales. It now feels socially acceptable to invite large groups
(of colleagues, family, etc.) to McDonald’s because there will be ‘something for everyone.’ In
fact, though, few people actually choose the salads once there, and burger sales go up. It would
be very interesting to construct an extension to the model which would allow us to test this
dynamic. That would require, at a minimum, creation of a variable for ‘variety of food offered’
or ‘social acceptability of inviting diverse parties to McDonald’s.’ We might also speculate that
people within large parties are more influenced, in terms of foodchoice selected, by the people in
their party, than even by others within their customer type of LHC or MHC. This would imply
that a third customer type, ‘same party,’ should be created, and the parameters (for the network
effects) adjusting accordingly.
On the other hand, the homophily findings suggest that McDonald’s might do well to attempt to
separate the two customer groups, in order to reduce cross-group network effects. For instance,
they could market each product type through specialised channels with more specific audiences,
in order to reduce awareness of customers of the ‘other’ product and customer type. While the
model suggests that this approach might lead to higher total unit sales, there is also a good
argument to be made that the ‘inclusive’ feel of the restaurant mentioned in the previous
paragraph is, in fact, a more important brand characteristic to cultivate for the long term — to say
nothing of the fact that ‘hiding’ salads reduces their power to fend off government intervention
in fast food. In a webcast announcing the company’s recent decision (described by the Wall
Street Journal as a ‘pre-emptive strike against the food police’) to replace half of the portion of
fries in Happy Meals with sliced apples, McDonald’s USA President Jan Fields said, ‘From a
business standpoint, [making meals healthier] is something we need to do to protect that
30
Under the golden arches
business. Clearly, this is the way people are moving — we have a health-conscious society. We’re
in business for the long term, and we have to evolve’ (Jargon, J., ‘Under pressure, McDonald’s
adds apples to kids meals,’ The Wall Street J ournal, 27 July 2011, p. B1).
Conclusion
In this paper we examined dynamics related to when firms seek to introduce new products in
order to move away from their existing brand recognition, or, alternatively, to broaden their
brand. For McDonald’s, will existing burger devotees find the presence of those green beds of
lettuce so shocking that they do not consider them? Will they find them exciting and attractive?
Perhaps salad availability will attract an entirely new clientele to the golden arches. Will this
scare away the old-timers? Altematively, salads will be, taken as an individual product line, a
flop, but their presence will draw bigger and more diverse crowds to McDonald’s, leading to a
commercially (though not health-wise) compensative rise in burger sales. Such questions are
critical, first, for companies, under pressure by public scrutiny and yet uncertain markets that
seek to transfer to their product line towards higher nutritious categories. Beyond the companies,
addressing such questions is important for public health Policy. For example, this may help
explain why health-campaigners’ push for more convenient availability of healthier food seems
“fruitless”. If businesses offering fast food cannot improve profits from introducing such options,
they are not going to do so. Our research, partially discussed in this paper, addresses many of
these questions.
The analyses, while preliminary, point to interesting findings. We developed hypotheses on
drivers of diverse scenarios involving dynamics of alternative and dominant product sales and
customers. We used the role of product distance and network effects as major explanations.
Indeed these factors explain multiple reference modes we discussed. Finally, referring to the loop
structure of the model and the ensuing dynamics, we identify the key factors which lead to
relative loop strength and therefore to one scenario or another. While our results demonstrated
that the final equilibrium depends on the adoption mediating parameters (in particular, network
effect strength, product distance), we believe that subsequent work can help explain more of the
dynamics we examined. The first factor we have begun focusing is on customer retention. The
31
Under the golden arches
inflow of alternative customer base (families) as well as consumption practices (salads), may
deter existing customers. On the other hand, upon limited take-off, healthy food triers may be
deterred; in addition, we plan including endogenous product portfolio dynamics. Finally,
regarding strategies, we plan to focus on rollout strategies targeted to socio-economically
differentiated areas.
This paper focused on the well-studied concept of ‘disruptive’ innovations, whether that be a
new product, or a familiar one appearing in an unusual setting. However, this work takes an
angle that seems relevant for other industries under pressure to shift to alternative practices either
because of unsustainable markets or threat of government intervention.
For instance, similar challenges (and opportunities) exist for alternative energy transitions.
Water-wheels and windmills powered industry long before electric utilities were on the scene,
just as salads existed long before McDonald’s. Now, however, the electric utility industry is
increasingly bringing ‘green’ tariffs online, offering customers the choice to pay a little more per
kilowatt-hour in exchange for the power distributor’s promise to buy a compensatory amount of
power from a sustainable generator. How can the firm, if at all, transition to a more socially
acceptable entity? Our point is not that fast food and power are identical, but rather that the type
of model we put forward here could be employed to help executives in such industries explore
how a marketing strategy could be integrated with product diversification and consumer
targeting.
The analysis shows how the existing stock of customers is on the one hand the vehicle to
transition, while on the other hand providing inertia and resistance. This is studied in other
industries (e.g. Struben and Sterman 2008), but perhaps surprising, seems also critical for a fast-
tumover retail product. For vehicles, the product turnover time is on the order of a decade, and
while the time constants related to changing food preferences may be shorter, decision-makers
may underestimate how the stock of burger eaters slows the rate at which salad eaters become an
important factor. The research also demonstrates how product replacement may be either through
core customer conversion or through core customer change. Because of this link to the customer
base, as well as the potential knowledge that consumers have with the “locally” disruptive
product, social-exposure-mediated network effects are critical in shaping the pathway of both the
customer and product base.
32
Under the golden arches
While network effects have been studied in many contexts, we believe that the coupling of
consumer heterogeneity, as mapped to the products available and network effects is a novel
contribution. We incorporated two factors, each built in the same fashion, accounting for the
difference between the perceived relevance of the ‘similar’ and the ‘alternative’, with the distant
between them being a parameter. The Foodtype ‘proximity’ represents the limited relevance of
the ‘further’ foodchoice to a customer of a given customer type, while the Network effects
weighting represents the limited influence of people of the other customer type, which leads to
the Foodtype ‘proximity,’. This then affected the network effects weighting, are both given a
value of one (full relevance) for the ‘close’ case and a lower, variable value, between 0 and 1, for
the ‘distant’ alternative.
Finally, in developing our model, we demonstrated the agility of the time-tested model logit-
model (McFadden 1973). Our model demonstrates how this structure can be replicated and
transformed for different decision-points. For example, we used three types of “appeals”
(anticipated appeal, trier discovered actual appeal, and network adjusted appeal) are derived,
using on the same logit model formulation. In addition, several parameters are ‘recycled’ to good
use at totally different points within the model. The first two types of appeal share the same beta
values which are used to weight the attributes upon which they are built. And Foodtype
‘proximity’ [a measure of how relevant a particular foodchoice is to someone of a particular
customer type] is used not only for both of these two types of appeals for Non-customers and
Triers, but also to temper the Core consumption fraction for Core customers. In addition, the
same equation structure is used for diverse calculations within the model, such as the Rate of
trying and Trier consumption.
References
2002 McDonald's Corporation Summary Annual Report, Oak Brook, Illinois: McDonald’s
Corporation, 2003. Available at
http://www.aboutmcdonalds.com/mcd/investors/publications/annual_report_archives/2002_annu
al_report.html (accessed 8 June 2011) (‘2002 Report’).
33
Under the golden arches
“79” Natural Surf 2,’ http://www.bicsportsurfboards.com/products/acs,3,61/7-9-natural-surf-
2,480.html, accessed 16 March 2011.
Anderson, S. P., de Palma, A., and Thisse, J. (1992), Discrete choice theory of product
differentiation, Cambridge, MA: The MIT Press.
Burger King: Whopper to go, The Economist, v. 396 (4 September 2010), p. 72.
Cramer, J.S., Logit models from economics and other fields, Cambridge, UK: Cambridge
University Press, 2003.
Christakis, N. A. and Fowler, J. H. (2007), ‘The spread of obesity in a large social network over
32 years,’ New England Journal of Medicine 357, pp. 370-379.
Dube 2010 tha
Hines 2001 tba
Jargon, J., ‘Under pressure, McDonald’s adds apples to kids meals,’ The Wall Street Journal, 27
July 2011, p. B1.
Katz, M. L. and Shapiro, C. (1985), ‘Network externalities, competition and compatibility,’
American Economic Review 75(3), pp. 424-440.
Lazarsfeld and Merton 1954. tba
L.L. Bean’s entries in Gale Directory of Company Histories and Gale Contemporary Fashion,
both available at http://www.answers.com/topic/l-l-bean-inc (accessed 16 March 2011).
McFadden 1973, 8 tha
Oliver, T. (1986), The Real Coke, The Real Story, Penguin, p. 149-151, as cited in Wikipedia
(http://en.wikipedia.org/wiki/New_coke, accessed 18 March 2011.
Prendergast, M (1994), For God, Country and Coca-Cola: The Definitive History of the Great
American Soft Drink and the Company that Makes It, New Y ork: Basic Books, p. 355, as cited in
Wikipedia (http://en.wikipedia.org/wiki/New_coke, accessed 18 March 2011).
Renaghan, M., Pers. comm., 31 May 2010.
34
Under the golden arches
Sterman, J. D. (2000), Business dynamics: systems thinking and modelling for a complex world,
Boston: Irwin McGraw-Hill.
Sterman, J. D. and Sweeney, L. B. (2007), ‘Understanding public complacency about climate
change: adults’ mental models of climate change violate conservation of matter,’ Climatic
Change 80, pp. 213-238.
Struben, J. R. (2006) Essays on transition challenges for alternative propulsion vehicles and
transportation systems, Ph.D. thesis, MIT. Used simply as a reference list for more sources on
network effects.
Struben, J. and Sterman, J. D., et al (2008), ‘Transition challenges for alternative fuel vehicle
and transportation systems,’ Environment and Planning B: Planning and Design 35(6), pp.
1070-1097.
Sun, B., Xie, J. and Cao, H. H. (2004), ‘Product strategy for innovators in markets with network
effects,’ Marketing Science 23(2), pp. 243-254.
The changes facing fast food : good and hungry, The Economist, 17 June 2010.
‘The Ray Kroc Story,’
http://www.mcdonalds.com/us/en/our_story/our_history/the ray_kroc_story.html, (accessed 16
March 2011).
Wikipedia entry for Société Bic, http://en.wikipedia.org/wiki/Soci%C3%A 9t%C3%A9 Bic,
accessed 16 March 2011.
35
der the Golden Arch
Hannah Rakoff Jeroen Struben
hannah.rakoff@mail.mcgill.ca jeroen.struben@mcgill.ca
Desautels Faculty of Management, McGill University
How do firms transition their
portfolios successfully?
Our approach: welookat Gm »
how the products interact with each = ;
other, and with the customers. RG
Our focus: customers and their
decisions among products. Firms
can ‘work back’ from these insights
to plan their transitions.
This is distinct from most research on portfolio transition, which focuses
instead on cognitive and capability aspects at the company.
lf healthy food is better for you,
why don’t people eat it?
It’s not
convenient Our
friends
will
laugh at
It’s
Le
It's expensive
Slow progress of
healthy food,
particularly in
affordable segment
lf we are supposed to eat healthier,
why don’t existing quick-serve
restaurants change their menus?
Will we
lose our
existing
customers?
How ‘close’
should the new
product be to
the ‘old’ one?
Dynamics of disruptive horizontal product differentiation
with consumer heterogeneity and network effects
This paper examines dynamics following introduction of
nutritious food by a company well known for _ high-
motivational and low-nutritious products. Employing a
system dynamics model, we investigate how consumer
dynamics affect uptake of the disruptive product. Our
example is the burger chain McDonald’s, which introduced
salads, fruit and other healthier options in the early 2000s.
Focusing on consumer choice, we analyze the process of
newcomers trying McDonald’s and either becoming “core”
customers, or not. The paper distinguishes overall
commercial success from that of the new product per se.
We examine conditions that separate commercial success
Our Question:
How do the dynamics of
social-exposure-mediated network effects
affect appeal to and choices by
existing and new customers,
and through this,
success after introduction of a
new, distinct product?
New, distinct product
Increased overall profit can come many ways,
not just by the new product selling well:
Active brand
‘enrichment’
* Novelty improves brand distinction (Apple’s GUI, 1980s)
¢‘Feel-good’ factor contributes to overall sales (corporate social
responsibility; ‘green’ electricity)
Positive ‘side
effects’ boost
existing *Canadian ‘Lighter Choices’ menu at McDonald’s attracts
product mothers - whose children order more Happy Meals. [2002
McDonald’s Corporation Summary Annual Report ]
Brand
carefully
managed to
retain existing
*L.L. Bean expands outdoor lifestyle products, but takes care not to
customers
become a ‘designer’ brand to avoid alienating existing customers
[Gale Contemporary Fashion]
interacts with brand in multiple ways
But there are also many ways to fail:
Social-exposure-
mediated
network effects
in play
*Strong culturally-based opposition in the South,
though a small minority nationwide, brings down
New Coke in 1985 [Prendergast, 1994; Oliver, 1986]
*Bic rumoured to have had initial trouble expanding into surfboards, since
Bic brand was associated with disposable product
Difficulty
Product in moving
distance
in play
a brand
‘up-
market’
(by drawing in new types of customers, whether these are
the profitable ones or whether they simply accompany
more burger-eaters), neutrality (in which existing
customers simply change over), or failure (by alienating
existing customers so that they abandon the company).
We focus on the role of heterogeneity in products and
consumers, and on interactions with social exposure-
mediated network effects. We consider in detail the large
and inertial installed base of pre-existing burger eaters,
and the degree to which its dominance is hard to unseat,
drawing parallels with reactions to other disruptive and
‘progressive’ products in industries ranging from consumer
products to electric vehicles to utilities.
Our case: 2002, and McDonald’s is in trouble.
They look to broaden into healthier foods for revitalisation.
Note: Success and failure: defined as change in total profits, rather
than just whether the new product sells.
Commercial Hi (Approximate) m™ Commercial
(- >) ‘a s
Salad salvation: Cannibalisation:* Not my MickyD’s anymore:
Salad t + ae Existing burger-eaters are
ara profitable. are Existing burger- alienated; salad sales do not
eaters change over make up the difference
‘ OR: - to salads L “OR: J
ii *) ( >
*There are substantial set-up Victim of own success:
Variety show: costs involved in launching salads, Salads are a big hit; in fact,
Variety attracts larger but margins are higher apne Hani Gina wennilet
groups, and in fact more (Renaghan, M., 2010), thus restaurants (like Subway) for
burgers are sold. J cannibalisation could result in more ‘healthy’ variety
\
some profit decrease or increase.
Adoption Pathways of Differentiated Products are Shaped by
Social-Exposure-Mediated Network Effects
Social proximity between consumers and between products and consumers is important
Scenario 1: Disassociation Scenario 2: Network Expansion®
Size of area indicates number of adopters
» »
Product location Product location
*) Several alternative scenarios may be hypothesized
Initial
positioning
Positioning of consumers
vs.product
Positioning of consumers
vs.product
Core constructs
- Network effets (Katz and Shapiro 1986
- Network effects for differentiated products (Anderson, DePalma and Thisse 1992)
- Homophily and social influence (Lazarsfeld and Merton, 1954; Christakis and Fowler, 2007)
Substitution matters.
Customers of
‘healthy’
competitors
(e.g. Subway)
Salad-eaters
(McDonald’s)
McDonald’s
Forms of substitution: burger-eaters
(large existing
stock)
*Cannibalisation
*Sales won from (or lost to)
competitors that have emphasised
‘healthiness’ for longer
Base model
Core customers and product distance
\
Non customers!
c
Rate of trying
McD ¢
yf
Fast food
exposure rate
¢
Abandonment
time constant c
> ae
Trier ¢
Core conversion
Two customer types (c)
Less Health Conscious (LHC)
More Health Conscious (MHC)
Two types of food-choices (f)
BURGERS
SALADS
> Product Core
Customers cf >
rate cf
f
Trier Core Los:
abandonment Rate cf
rate ¢
, Lost
SP Cusioners <—
S
When product distance is higher,
difference in customer types becomes more important.
Relative consumption
Total Consumption
Low product distance
LHC, Burger
MHC, Burger
LHC, Sala
MHC, Salad
_, distance
Total Consumption
HC, Salad
LHC, Salad
12 0 12 24
36 48 = 60's 72s 84H
Time (Month)
108-12 0 12
24
36 48 6072 84H
Time (Month)
108
Heterogeneous choice sets leads to preferences
increasingly satisfied across the population.
Sales increase.
Adding in network effects
div VIL UE
Fi ‘onstit nption Weighted sum of <+—_ x_
all consumption Total monthly
—_ observed cf a <
Network Triers consumpuon — <g_ Consumpti
erWORK ETHERS ~«d _ NE observed of Observation *
effects cf Exponent
riers
Triers Intrinise / ;
acinaal ol Total monthly ~ na
a consumption ct
Triers Network
adjusted appeal cf '
'
CX Core
Triers Network consumption cf
adjusted market share ,
ef Normal Core Peripheral
Survival Fraction consumptio *
nef
Normal Trier +
INPT Survival Fraction> IN PCC IN
Product
Trier c
x pst Priiduct'Care x —
Core Customers cf
conversion rate Core Loss
ar AL Data af
f trying
Can
But network effects reinforce the dominance of large
stocks of customers of existing products
Relative consumption*
,
1.125 dace TTT Total Consurfiption
LHC, Bulger LHC, Burger
0.75
MHC, Salad fie Salad
LHC, Salad LHC, Salad
MHC, Burger
J”
12 0 12 24 36 48 60 72 8 9% 4108-12 0 12 24 36 48 60 72 84 6 108
Time (Month) Time (Month)
*product proximity fixed at p = 0.25 — relatively distant products
Sales of new products, particularly to new customers, suffer.
Total sales also increase less.
Homophily
However, relative strength of within-group network
effects actually helps the ‘encroaching’ group.
17 Total relative consumption (
NE, RCCI = 0}
Zo14 NE, RCE OS |
letwork Effects (NE),
1.1 Relative Cross-Category Interaction (RCCI) =1 |(
No Network : :
0.8 Leffects Relative consumption of
-12 0 12 24 36 48 60 72 84 96 108 salads by MHCs
Time (Month) 06 1
NE, RCCL 5-0]
The lower the strength of social 0.45 Act
pressure across consumer types, as
contrasted with those within a 0.3 Network Effects (NE), |
No Network Relative Cross-Category Interaction (RCCI) 51
consumer type, the more that new effects
products can benefit from peer-to- — 0.15
peer exposure among susceptible
customers without ‘frightening off’ 0
existing customers of the old product* “7 9 OS sO es
Planned extensions
¢ Detailed examination in and calibration to
the context of MacDonald's case
— The role of the installed customer base
— And its Interaction with portfolio expansion rate
¢ Homophily in network interactions
— Further examine, empirically, the role of
“oroximity” between customer types and network
interactions
— Mechanisms and factors present in netowork
homophily
Conclusion and Implications and
conclusion
We examined the role of product, consumer distance, and
asymmetry in network effects
— Network effects suppress salad consumption among MHC customers,
through existing base of LHC customers
— However, once network effects are considered, high homophily leads not only
to more consumption of salads by MHC, but to higher consumption overall.
Implications for Practice
— Overly simplistic consideration of network effects while attempting to
transform their portfolios can lead firms to underestimate the difficulty both
of attracting new types of customers, and furthermore, of doing so while
retaining the original ones
Contributions for Research
— Our model combines network treatments and disruptive innovations to allow
network pressure and distance between consumers in the introduction of new
products
Selected References
2002 McDonald’s Corporation Summary Annual Report, Oak Brook, Illinois: McDonald’s Corporation, 2003. Available at
http://www.aboutmcdonalds.com/mcd/investors/publications/annual_report_archives/2002_annual_report.html (accessed 8 June
2011)
Anderson, S. P., de Palma, A., and Thisse, J. (1992), Discrete choice theory of product differentiation, Cambridge, MA: The MIT Press.
Christakis, N. A. and Fowler, J. H. (2007), ‘The spread of obesity in a large social network over 32 years,’ New England Journal of
Medicine 357,pp. 370-379.
Katz, ML, and C Shapiro. 1986. “Technology adoption in the presence of network externalities.” The journal of political economy
94(4):822-841.
Lazarsfeld, P.L. and R.K. Merton. 1954. “Friendship as a social process: A substantive and methodological analysis.” 18:18-66.
L.L. Bean’s entries in Gale Directory of Company Histories and Gale Contemporary Fashion, both available at
http://www.answers.com/topic/I-l-bean-inc (accessed 16 March 2011).
Oliver, T. (1986), The Real Coke, The Real Story, Penguin, p. 149-151, as cited in Wikipedia (http://en.wikipedia.org/wiki/New_coke,
accessed 18 March 2011.
Prendergast, M (1994), For God, Country and Coca-Cola: The Definitive History of the Great American Soft Drink and the Company
that Makes It, New York: Basic Books, p. 355, as cited in Wikipedia (http://en.wikipedia.org/wiki/New_coke, accessed 18 March
2011).
Renaghan, M., Pers. comm., 31 May 2010.
Struben, J. and Sterman, J. D., et al (2008), ‘Transition challenges for alternative fuel vehicle and transportation systems,
Environment and Planning B: Planning and Design 35(6), pp. 1070-1097.
