1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
Functional Economic Analysis of Purchasing at MITRE
Thomas Gulledge Henry Neimeier
The Institute of Public Policy The MITRE Corporation
George Mason University 7525 Colshire Drive
Fairfax, Virginia, 22030, USA Mc Lean, Virginia, 22102, USA
Abstract
Functional ic analysis is a mod pp that provides a uniform basis for analysis and
comparison of alternative i and The approach takes into account
the costs, benefits, and risks associated with new “ways of doing business and managing
or The entire ing process from initial request to final delivery, payment and
accounting is being re engineered at MITRE. A complete resourced process flow chart was
developed for both the present and proposed systems. An "i think" system dynamics model of both
the present and proposed process was developed. The model projects the seasonal workload over the
proposed system lifetime. Dynamic normal, overtime, and temporary staffing requirements were
calculated. The new system reduced total requisition delay by a factor of ten. This will greatly
reduce expediting actions and costs. Multiple data bases and computer systems along the process
were combined into a single system. This greatly reduces data entry and reconciliation effort. The
new process groups purchase requisitions by type that provides the opportunity for bulk discounts.
All these will result in 37 million dollars of savings over a ten year system life.
‘System Dynamics : Methodological and Technical Issues, page 33
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
Functional Economic Analysis of Purchasing at MITRE
Objective
The purchasing system at ‘MITRE spas several departments. Over the years manua
have been dently by different departments. These systems are or
different computers and do not directly communicate. Purchasing data must be keyed in on eact
system. This greatly increases cost and delay. Finally all the separate data bases must be
reconciled to meet Department of Defense auditing requirements. To reduce both purchasing cos
and delay, MITRE included purchasing in its financial re-engineering process.
Fi ee | A ysis proced
A functional economic analysis starts with the definition of what is to be included in the systen
studied. This should include all sub-systems that are under control of the organization, and have i
if impact on per Then the ‘kload measure is defined. In our case this wa:
completed purchase requisitions. A resourced process flow chart is then developed for thi
existing process, with all costs, times, and delays related to the workload measure. Flow chartinj
the existing process often reveals p les include elimination of non
value added activities such as redundant acti ities, storage and transport activities, inspection
activities, and- expediting activities. If Activity Based Costing ABC is employed in th
organization, determining the baseline and alternative activity costs is considerably simplified
The resourced process flow charts form the basis for model development. A model is required ti
project performance and costs over the system lifetime. Since the alternative process is new, :
model must be developed to project its cost and per In our case we developed a systen
dynamics model in “i think". The simulation model is executed over the system lifetime. /
comparison is made of performance and cost of the baseline and proposed alternatives. There i
considerable uncertainty in making future cost and performance Projections. There is als:
uncertainty in estimates of the proposed alternative process, since in most cases it is not i
and ilable for field Given limited study resources, even the baselin
process parameters are calculated from limited sample data and thus are uncertain. In functiona
ic analysis the uncertainties are all d d and the i
in fi inal measure of effectiveness is calculated. We used the analytic uncertainty modelin;
d in another conf paper to do this. In our case the DEMOS packag
was used to perform sensitivity and uncertainty analysis on the purchasing model. The analyti
uncertainty analysis calculations were included in DEMOS library subroutines. Our measure o
effectiveness was the discounted present value distribution of savings. Management decides if th
uncertainty in the final distribution of savings is acceptable for a decision. If there is too muc
uncertainty, then more detailed parametric data is required. For greater accuracy the proces
might be dis-aggregated to a finer level of detail and a new model developed. Of course both thes
actions take time and money and must thus be evaluated relative to the lower uncertainty in fine
measure of effectiveness that they may provide.
System Dynamics : Methodological and Technical Issues, page 34
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
1 :
met ee
Specify Compare
Functional Fo) Workload I TThink/DEMOS ) } Baseline & 2} Sensitivity }p} Risk
\ a x = /
Processes cof Alternative
Figure 1. Functional Economic Analysis Procedure
Present and proposed purchasing procedure
Figure 2 gives the staff time required per purchase requisition activity and activity delay. Data
was obtained from time-stamped purchase requisition forms and interviews of purchasing
personnel. Estimates were calibrated based on past staffing and actual numbers of purchase
requisitions processed or rejected. Note delay is far longer than activity staff minutes and includes
mail time, time to contact people for approvals, wait time for typing. Both delay and staff time
are modeled for both the present baseline and proposed alternative system. The purchase
requisition (PR) originates at the department level. It is typed up based on a purchasers submitted
form. In the new system the purchaser directly enters the information from his desktop
to the pi local area network. The input program does considerable
checking as the form is being filled out to insure each entry is valid. In the old system this
checking was done later as the process p ded. This greatly ii d delay because many
questions required contact with the purchaser for an answer (telephone tag, at meetings etc.).
The old system had many rejected and reworked PRs. This rework was included in the table staff
minutes and delay numbers. Given that the requisition is on line from the start and all approval
authorities in department and division also are on the local area network, a new approval
d can be impl d. The department and division approval authorities are
sequentially notified as soon as the PR is input. If no action is taken in a specified time period the
PR is routed to a designated alternate. This considerably reduces approval delays. In the old
system PR ey was ie a to several independent computer systems in division, configuration
ys g g and iving. These many data bases
required conshiceable staff time for redundant data input. In addition because of normal human
input errors, effort was required to reconcile the data base differences and prepare files for
potential DOD audits. Multiple purchase orders can be combined into a single purchase order to a
vendor. Similarly a single purcha can have orders to multiple
vendors. Manually keeping track of the purchase requisition number to purchase order number
led to considerable reconciliation effort, In the new system with its single common on line data
base this linking is automatic. With an on line data base the purchaser can find the status of his
order at any time without disturbing personnel performing purchasing activities. Given the large
variable delay of the baseline system, considerable purchasing staff time was taken in finding and
reporting status to frustrated purch Confi i is required for computer and
System Dynamics : Methodological and Technical Issues, page 35
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
peripheral purchases. These make up 40% of purchase requisitions. The new system has a menu
of configured systems the purchaser can select. If one of these configurations is selected no
configuration management activity is required. In the new system only 5% of purchase
quisitions should be tandard. A property number is automatically assigned in the new
system so this step was deleted. In the baseline system a separate budget approval was required by
accounting. Given the PR delays encountered, a call back to the division administrator was made
to see if there was still money in the budget. With the greater speed of the new system this was
not deemed necessary. In the old system there was no official direct link between accounting and
purchasi A ing could be withholding vendor payment while purchasing was issuing new
purchase orders to that vendor. Similarly vendor performance records were not kept. The new
common data base system links all parties and keeps vendor performance-data. Purchases can
now be guided to vendors with superior performance. With a more limited set of preferred
vendors, standard agreements and forms become a possibility. Electronic data interchange and
FAX links can be used rather than the mail. This considerably reduces the effort and delay in
obtaining and evaluating bids. With a preferred group of graded vendors, and a lower delay time,
the effort required for vendor expediting should also be reduced. In the process of gathering
baseline process data the purchasing manager mentioned that she could get significant discounts if
bulk orders were made. The new system supports this with a draft PR. Purchasers project their
next year needs with this device. The projected unapproved PRs are entered with a draft
categorization. These are categorized and summed and used as the basis for bulk purchase
agreements. The 15 to 20% savings from this feature alone more than compensates for all the
new system cost. A side benefit is the future cost visibility given to budgeting personnel.
Different i ives are being idered to use of draft PRs.
Staff_Minutes/Activity Days Activity Delay
Component Baseline |Alternative |Baseline Alternative
Department 18.07 18.07] 14.99 0.50
Division 15.31 15.31] 2.93) 0.50
Config. Mangt. 114.00 7.05 1.40 0.10
Property 14.08 0.00 1.76 0.00
[Assign Buyer 15.13 1.99 0.99 0.24
Budget Approve 15.0 0.00 1.00 0.00
Obtain Bids 336.00 252.00 0.80 0.49
Evaluate Bids 145.20 72.60 4.55 2.28
Purchase Order 18.18 9.09 0.62) 0.31
Expedite 63.50 25.91 1.00 0.00
Invoice 12.0 1.80 0.6 0.02)
Reconcile 15.00 0.00 0.50 0.00,
Audit 30.0 15.00 3.00 0.13)
[Total $11.47 418.72 34.05 4.48
Figure 2 Purchasing Activity Staff Minutes and Delay
The MITRE purchasing model
Figure 3 is an extract of some of the key sectors in the purchasing model. In the upper lef
seasonal workload is projected into the future for both present and alternative systems. Peopli
tend to spend their budget as soon as they get it or before they loose it at the end of the fisca
year. The activity times also change. For example more checking must be done at the end o
the fiscal year to see if there is sufficient budget funds available. To handle the increase:
workload overtime is authorized and temporary staff are hired. The greater staff pressure ani
new employees lead to a higher error rate that must be corrected after the peak loading
Personnel that normally ile and audit pleted purchasing paperwork are reallocated t
handle the increased workload. Thus backlog in auditing increases. This is captured in the audi
System Dynamics : Methodological and Technical Issues, page 36
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
backlog sector. Five Years of historical MITRE archive purchase requisition documents were used
to initialize the workload projection sector.
The staffing actions sector captures the decision rules used to approve overtime, hire
temporary personnel, and hire permanent _p 1. It also all these p to
departments based on the data pi d in figure 2. th ing staff lowers staff utilization and
reduces process queuing delays : at the expense of greater staff cost. This is captured in the delay
calculation sector. This sector employs analytic queuing described in another conference paper.
hasing system per is d in terms of overall purchasing delay, delay variation,
audit backlog, purchasing error rates, and operations cost. The model provides for staffing cost
versus purchase requisition delay trade-offs for both the present and alternative systems. Note
that there are separate staffing, delay calculation, audit backlog, requisition and vendor expediting
sectors for baseline and alternative processes. Only the baseline process is shown in figure 3. The
ition and vendor expediting sector effort required as a function of process delay.
‘System Dynamics : Methodological and Technical Issues, page 37
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
Workload Projection Requisition & Vendor Expediting
Growth
InitialPRperDay season
RegCalisPPR ExpReq ExpTot
SmoothDays
DelayAvg
OP prvendorExp
Delay CallsPVendor
OverhdFix ArivRpPert Daily Cash Flow © Personnel Overtime Temp
Delay Calculation Costpit CumCost
ExpTot
Cost% —CumCostt
DaysAuditBkig U cvs
OM1 Bulks: Personnell Overtime! Temp1
Staffing Actions ey .
AuditBklg TechTm Audit Backlog
Personnel TechPers
Util p f
‘Audtm ersonnel
AuditR:
Backlog
7 FinPer
Overtime ArR DaysAuditBkig
Figure 3. Key Purchasing Model Sectors
System Dynamics : Methodological and Technical Issues, page 38
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
Note that the there is iderable ii ion between purchasing activities. For example, hiring
staff reduces purchasing delay which reduces the staff required for requisition expediting.
The daily cash flow sector compares baseline and alternative costs over the system
lifetime. Discounted present value of the cash flows is done in another sector not shown in figure
3. Note the significant bulk savings obtainable with the alternative system are included in this
sector.
Sensitivity and uncertainty analysis
Figure 4 presents the DEMOS purchasing sensitivity analysis model. The baseline system is
shown on the left and the alternative system is on the right. Delay and audit performance is
presented in sub-models accessed through clicking on the lower left and lower right bold buttons.
Rounded rectangles contain data inputs or relationshi i Square les contain lists
of parameter values for sensitivity analysis. Key process parameters are monthly operations and
maintenance cost (OM), _ staff utilization (Util), PR process time, overhead proportion
(Overhead), dollars per staff month ($Staff), initial PRs per month (PR/mo) and initial monthly
PR dollar value (PR$). Other inputs include the new system investment cost (Invest$) and the
discount rate (Discount) and the system life (simulation time). Sensitivity analysis parameters
include annual growth rate in number of PRs (PR Growth), annual growth rate in PR dollar value
(PR$Growth), proportion of purchase requisitions with bulk saving potential (Pr Bulk) and the
bulk saving percent discount (Bulk Save). Figure 5 gives a sample of the sensitivity analysis
model output of discounted present value of savings when all alternative investment operations
and maintenance costs are included. It assumes 15% bulk discount on 40% of the purchase
requisitions. Even with no growth in purch isitions and a 5% reduction in the number of
PRs there are $12.53 million savings over 5 years assuming a 7% annual discount rate. At a 5%
annual growth in PR value, and a 15% increase in the number of PRs, $36.85 million savings are
achieved over 10 years.
Present Value
Delay & Audit
Performance Discount
Figure 4. DEMOS Sensitivity Analysis Model
Delay! & Audit1
Performance
System Dynamics : Methodological and Technical Issues, page 39
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
Discount Rate
System} Annual 0% | 7%
Lifetime] PR $ Annual Growth In Number Of PRs
Years |Growth| -5% | 5% | 15% | 5% | 5% | 1%
5 | 0% | 15.70] 16.59] 17.83) 12.53] 13.24) 14.21
5% | 17.95] 18.84) 20.08] 14.31) 15.02) 15.99
10 [0% | 35.16] 38.74] 45.86) 2409) 26.37] 30.71
hh | 44.91) 48.49) 55.61} 30.23) 32.50} 36.85
Figure 5. Sample Present Value Of Savings In Million Dollars
Figure 6 shows the impact of bulk savings. It is based on a 5% annual growth rate in PR dollar
value and a 15% annual increase in the number of PRs. Even at no bulk savings (0% bulk
purchases) there are half a million dollars savings over a 5 year system life and $7.33 million
savings over a 10 year system life. The chart shows that percent bulk purchases and the
proportion of PRs that have bulk purchase potential are two of the most significant model
parameters.
System} % Bulk Percent Bulk Purchases
Life | Savings 0% 30% 40% 50%
10% 0.50 8.25 10.83 13.40
5 15% 0.50 12.12 15.99 19.87
20% 0.50 15.99 21.16 26.32
10% 7.33 22.09 27.01 31.93
10 15% 7.33 29.47 36.85 44.23
20% 7.33 36.85 46.69 56.53
Figure 6. Sample Bulk Savings in Million Dollars
A similar DEMOS model to the one shown in figure 4 was developed for uncertainty analysis.
Here the input parameters are given beta uncertainty distributions based on minimum, maximum,
mean, and standard deviation estimates. The model calculates the discounted present value of
savings which is presented in figure 8. In our case the discounted (7% annual rate) present value
of savings ranged between $14.59 million and $20.22 million over a five year system life with a
mean savings of $16.85 million. This case was for a 5% annual growth in PR dollar value and a
15% annual growth in the number of PRs. It included a 15% bulk purchase discount on 40% of
the PRs. The mean discounted value of $16.85 million differs from the mean value without
uncertainty analysis of $15.99 due to the skew of the input parameter uncertainty distributions
System Dynamics : Methodological and Technical Issues, page 40
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
and nonlinearity of nodal ee relationships. The Hgure: lists all of the uncertain parameters
in the model. specified with esti mean, and standard
deviation sintistion A beta “distribution | is fit to each oe the uncertain parameters. It is also fit to
the resulting savings measure of effectiveness. The fit beta a and b parameters are given in the
figure 7. The analytic uncertainty modeling technique employed is described in another
conference paper. The beta density function is given by:
Probability density =C x@l) (-: xyo- 1) where:
X= (value - mini ) / (maxii = mini )
C = normalizing coefficient so the integral of the density
between minimum value and maximum value is one
Figure 8 plots the discounted present value of savings cumulative distribution function. In this
case it is easy to decide to go with the alternative system. In general however this might not be
the case. Alternative savings distributions can overlap, or there can be a non-zero probability of
loss.
fariable Minimum) Mean |Maximum|Std.Dev. |Variance} a b
Base Staff Days/PR] 0.4900 | 0.5490 | 0.6050 [0.022361 |5.00E-04| 2.8722 | 2.7309
JAlt.Staff Days/PR_| 0.2500 | 0.2830 | 0.3200 |0.010954|1.20E-04! 4.3254 | 4.8469
jase OM S/Mo 28,000 | 31,490 | 37,000 | 1,000 |1.00E+06] 7.0692 | 11.1610
IAILOM $/Mo 16,650 | 18,500 | 22,200 | 707 | 500000 | 4.2300 | 8.4600
PR Growth %/Yr § 15 20 | 0.7746 | 0.6000 | 3.9626 | 1.9814
PR$ Growth %/¥r | 0 5 10 | 0.4899 | 0.2400 | 3.8403 | 3.8402
Po Bulk PRs: 1 40 60 1 1 | 4.4946 | 2.3051
fo Bulk Savings 5 15 25 | 10.95 | 120.00 | 4.5000 | 4.5000
Jnvestment SM | 3.00 | 4.44 | 7.00 | 0.84 | 0.70 | 1.5359 | 2.7304
Cum. Savings SM | 18.95 | 21.15 | 24.49 | 1.18 1.39 | 1.6970 | 2.580
Present Value $M | 14.59 | 16.85 | 20.22 | 1.08 1.47 | 2.1939 | 3.2816
Figure 7. Uncertainty Analysis Inputs And Results
System Dynamics : Methodological and Technical Issues, page 41
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
se
“5
&
= L
= =
2o
sh a g
So cs q
2
a
@
I
=
&
a
—
=
oO
Figure 8. Discounted Present Value Distribution Of Savings In Million Dollars
Conclusions
We have summarized a full functional economic analysis along with sensitivity and uncertainty
analysis. Both analytic queuing and analytic uncertainty analysis techniques described in othe:
conference papers were employed. The alternative system suggested is presently being
implemented at MITRE. Staff are looking forward to speedy execution of purchase orders.
References
Richmond,B. et. Al.., 1991, i think , The visual thinking tool for the 90's: i think User's Guide,
High Performance Systems Inc., Hanover, New Hampshire.
Henrion,M.et.Al. 1993, DEMOS Professional Tutorial and Reference, Lumina Decision Systems,
Inc., Palo Alto, California.
System Dynamics : Methodological and Technical Issues, page 42
