Technology Budgets and Total Cost of Ownership

The university approved four million for the first three years of its student-records replacement. The required service covered 22,000 students, but the estimate priced licenses for only 400 administrators. It also omitted transition and retention work. At the end of the same three-year period, cash spending allocated to that replacement totaled nine million. The university is fictional; the reconciliation below holds the currency, reporting period, and intended service coverage fixed. Assume no general price inflation in this example. These are gross cash costs, before any claimed savings, and each payment is assigned to one line.

Three-year cash reconciliation	Million currency units
Originally priced subscription, implementation partner, and other delivery work; actual spending on these lines matched the estimate	4.0
Additional parallel-operation payments during eleven months of transition	1.1
License correction to the contract’s student counting unit	1.2
Retained archive access and related legacy support beyond the planned cutover	0.6
Data transfer from the old provider	0.3
Additional paid training and temporary staff cover	0.5
Replacement’s allocated share of integration-platform construction and operation	1.3
Total actual cash spending over the same three years	9.0

The five-million difference is unbudgeted spending within this stated service scope. The two legacy-operation rows cover different services and periods; they do not count the same invoice twice. Finance also recorded one million of existing staff time used for cleansing, testing, and learning. That allocated labor cost is outside the nine-million cash reconciliation: it was already in payroll and is not another million of additional cash paid. Its value in an economic comparison depends on the work displaced or otherwise required. The shared platform’s other consumers retain their own allocations. The old archive still ran, so the forecast retirement saving remained a separate, unverified benefit claim.

This case exposes a budget assembled from a partial quote. It cannot establish how often technology overruns have this cause. A useful technology cost model connects planned spending, demand forecasts, accounting treatment, shared-cost rules, and the assumptions behind benefits. It also makes uncertainty inspectable through evidence-based adjustments and sensitivity tests. Ninety-Four Projects, Every One Approved examines how those estimates inform portfolio choices; Live for a Year, Used by a Third follows the evidence after delivery.

Building a budget, revising the forecast, and assigning costs

This article separates several responsibilities within budget management. Building the budget records authorized spending and its purpose. Forecasting projects what will be spent, and when, using demand, prices, capacity, and delivery evidence. Allocation assigns costs to budget owners or explicitly retains them centrally. These views should reconcile, whether held in one system or several documents. Run, grow, and transform are one possible spending classification, used in the portfolio example, rather than mandatory categories. Give each cost one primary allocation and use separate tags for overlapping purposes. An obligation to maintain support or meet a regulation needs a funded response, but does not automatically select a particular renewal: compare permitted upgrades, replacements, continued support, or ending the affected activity.

Motion	Question	Inputs	Output	Fails as
Build	What spending is authorized, and why?	Portfolio choices; obligations and permitted responses; service commitments	Dated lines with owners, reasons, and the organization’s chosen classifications	Carrying forward last year’s costs without reviewing their purpose
Forecast	What will we actually spend, and when?	Demand and usage trends; metered consumption; capacity plans; renewal calendars; retirements	A time-phased projection with ranges, revised on a cadence	A flat twelfth of the annual figure each month; a cloud bill that surprises finance every quarter
Allocate	Whose budget bears each cost?	Written allocation rules; account and tag structure; consumption data	Every cost has a bearer, or is explicitly held centrally with a reason	Shared platforms in nobody’s budget; the university’s integration layer

Google Cloud’s cost guidance recommends monitoring current costs and developing a growth projection from business requirements, demand patterns, and sustainable capacity. Consumption can change faster than a formal budget. Distinguish usage monitoring and anomaly response from a rolling forecast and from an approved budget revision. Set review frequency by volatility, materiality, data availability, and the decisions it must support; trigger an extra forecast when a material demand or price assumption changes. Monthly review may suit one service, while another needs faster intervention or a different formal planning cycle. Give the people changing consumption access to the forecast and a route to explain departures.

Reconcile economic choices with accounting and affordability

Capitalization and operating-expense treatment affect accounts, budgets, and spending authority. A different accounting label alone does not create a benefit. An actual change from ownership to a subscription can alter prices, payment timing, commitments, tax treatment, operating work, and risk, so it can change the decision. Before approval, compare acceptable alternatives on benefits, lifecycle costs, uncertainty, and affordability, with the applicable accounting rules in view. TCO supplies the cost side; it does not determine whether the investment is worthwhile.

The UK’s Green Book distinguishes social appraisal from the financial case. Its economic comparison uses real discounted values, whereas its financial case records nominal, undiscounted amounts under public-sector accounting rules. Depreciation belongs in the accounting view rather than being charged again on top of the same resource cost in appraisal. This is a scoped public-appraisal framework, not a rule that every private financial analysis must omit discounting. Use consistent prices, timing, and discount assumptions for the chosen method, and reconcile the resulting resource, cash, and accounting views.

Unrecoverable past expenditure should not decide which remaining option is better. Continuing to use a paid-for resource can still carry an opportunity cost, and termination or replacement can create future payments. Preserve those differences. For an accounting-based unit-cost measure, the GOV.UK Service Manual includes depreciation of capitalized startup items and excludes the corresponding capitalizable startup spend from the running-cost total. It also specifies capital charges where applicable. That convention avoids counting acquisition and depreciation twice in that measure; it does not replace the cash schedule used to fund the purchase.

Total cost of ownership, and the items it leaves out

Total cost of ownership estimates acquisition, operation, and eventual replacement costs over a stated horizon and workload. Compare options on a common service scope, demand, and cost basis. GOV.UK’s purchasing guidance connects lifecycle cost with the outcomes sought, market alternatives, and the organization’s ability to deliver and support them. In the university case, the intended student population was known, but the license estimate and transition scope did not cover it. The following checklist helps investigate such omissions; it is not a claim that every project misses the same items.

Cost to investigate	Why it can be missed	How to estimate it	The university
Internal staff time	Existing payroll makes the work look free	Hours by role and an explained valuation; distinguish redeployed time, incremental hiring, and cash already included elsewhere	One million of allocated existing-staff time, reported separately from additional cash
Dual running	The estimate assumes old services stop at cutover	Separate retained services and periods; count each payment once	Eleven months of additional parallel-operation payments; archive support on a separate line
Data movement and network	Equipment purchase is confused with the full cost of connectivity	Google distinguishes upfront routers, switches, and cabling from cloud usage pricing. Include recurring on-premises circuits, support, power, and operations too; price each cloud path, processing charge, and fixed connection fee	Transfer charges omitted from the quote
Operating the environment	Product pricing omits surrounding services	Google’s TCO guidance includes maintenance and support as well as hardware and software. Identify monitoring, security, backup, and shared infrastructure costs separately	The replacement’s integration-platform allocation
License scope and growth	Priced at today’s count and the buyer’s reading of the terms	The vendor’s counting unit, verified in the contract, over the forecast population; true-up terms	Every student was a user
Retirement and retention	The old system’s end is assumed to be free	Archive, retention access, decommissioning, and the license until the last dependency moves; the system retirement checklist	The old system still running for its archive
Leaving	Exit is outside the initial quote or appraisal scope	The exit cost: moving data out, rewriting what depends on the provider, rebuilding operations, over the realistic horizon	Not estimated; the next migration’s opening line
Planning and learning work	Time and displaced work are not supplier invoices	Google includes migration planning and training. Reconcile these activities with the staff-time and paid-training lines above rather than adding a second allowance for the same hours	Existing payroll time and additional paid cover are different records

Build the estimate with the people responsible for the work: registry staff for data preparation, operations for the environment, and records owners for retention. State the horizon, workload, cost basis, and exclusions at the start. A three-year estimate is not a fixed fraction of a ten-year cost: an upfront cost of 90 plus one per year gives 93 over three years and 100 over ten, before discounting. Choose a horizon suited to the decision, compare alternatives consistently, and state how later obligations and any residual value are treated. A shorter budget window can coexist with a longer appraisal; reconcile them instead of extrapolating by years alone.

Shared costs: a rule, written down

The integration platform eventually served the replacement and four other systems. Its steady-state operating budget was 140,000 a year, held centrally; the replacement’s three-year allocation in the reconciliation included its share of construction and operation, not the other consumers’ shares. Keep the organization-wide total separate from those allocations. A transfer between internal budgets does not add an external cost.

The Service Manual gives examples of allocating shared costs by team full-time equivalents and a shared call center by time spent on the service’s calls. The useful key depends on the service and purpose of the allocation. Cost attribution connects spending to consumers through accounts, tags, and usage records. Showback reports it; chargeback formally assigns it to accounting budgets. FinOps treats neither as inherently more mature. Declare whether a split measures consumption or applies a policy proxy, report unallocated costs, and explain central funding. Where an allocation is intended to change behavior, identify which costs the recipient can influence. Review the rule when its basis changes and preserve the reason for any revision.

Estimating the effect honestly

Correct for optimism first

Optimism bias is a systematic tendency to underestimate costs or duration and overestimate benefits. The Green Book requires explicit corrections in its public-appraisal context, informed by comparable historical forecast errors. A technology organization borrowing this approach should document the estimates compared, their scope and stage, and the judgment involved in choosing a reference class. A contingency is an allowance for residual uncertainty, not evidence that the underlying estimate is accurate.

Where suitable evidence is absent, the Treasury’s supplementary guidance supplies generic starting values based on a 2002 public-project study. Its equipment/development capital-cost range is 10–200%, and its outsourcing operating-cost range is 0–41%. These are not possible minimum and maximum outcomes. The upper values refer to outline business cases and are the prescribed starting points within that method; justified reductions require evidence of mitigation. Select the appropriate project components and cost categories, and assess specific risks as well. Do not apply an old generic range automatically to every technology budget.

For the university’s next, separate finance-system proposal, suppose four earlier comparable implementation packages had matched-scope overruns of 20%, 30%, 50%, and 60%. Their unweighted mean is 40%. The student-records case’s incomplete whole-program estimate is excluded from that reference class until its scope can be reconstructed. Four observations provide limited evidence, so the finance proposal retains sensitivity tests. Its implementation calculation, in million currency units, is:

Implementation allowance for the finance proposal	Calculation / amount
Unadjusted implementation estimate	4.0
Initial optimism adjustment	4.0 × 40% = 1.6
Prevention work added to core costs	0.2
Documented reduction in the initial adjustment	0.6
Remaining optimism allowance	1.6 − 0.6 = 1.0
Separately assessed residual risk, excluded from the reference-class allowance	0.3
Core implementation budget	4.0 + 0.2 = 4.2
Reserve held by the approving authority	1.0 + 0.3 = 1.3
Implementation funding envelope	4.2 + 1.3 = 5.5

These figures are invented. Assume an independent review supports the 0.6 reduction and confirms that the separate 0.3 risk is not already in the historical uplift; otherwise the reserve calculation must change. Prevention work is paid once in core costs. In a financial plan, convert appraisal values to the appropriate nominal payment schedule. This illustration assumes zero inflation and uses undiscounted cash totals. The reserve is a funding allowance, not a forecast that every unit will be spent, and is not added again to the 5.5 implementation envelope.

Then find what would change the answer

A switching value locates a boundary for a specified comparison: for example, when a financial net benefit reaches zero or when another option becomes preferable. It does not automatically identify the only boundary or settle the whole investment decision. Define the outcome function, hold other variables fixed for an individual test, and then examine important joint changes. The Green Book uses switching values within its broader assessment of value for money.

Consider a simplified five-year cash screen for the finance proposal, against continuing the existing service. The interval starts at approval; both options must meet the same required service conditions. All amounts below are millions in one currency, with zero inflation and no discounting solely to make this cash calculation reproducible. It is not an NPSV calculation or a complete economic appraisal. Let C be implementation spending; use the full 5.5 funding envelope for a conservative screen. Other incremental operation and end-of-horizon exit costs total 1.0. The new contract charges 0.001 million per licensed finance user over this horizon, so 400 verified users cost 0.4. Assume the population is constant. These costs are separate from implementation and legacy overlap.

The maximum modeled avoided legacy payments over five years are 8.0. Let a be the fraction attainable after service and retirement constraints, before transition delay; it is not automatically the fraction of staff who adopt the system. Let d be months of delay relative to the model’s earliest feasible retirement. Each month reduces that saving by 0.1 through continued legacy payments, with no separate charge for those same payments elsewhere. Assume this linear relationship holds over 0–12 months and excludes the fixed retained-record provision already in the 1.0 line. The screen is net cash benefit = 8a − C − 1.0 − 0.1d − 0.001n, where n is licensed users. At C = 5.5, a = 1, d = 5, and n = 400, it gives 8 − 5.5 − 1 − 0.5 − 0.4 = 0.6.

Variable varied alone	Base assumption	Financial break-even value	Direction producing a negative result
Implementation C	5.5, including the reserve	8 − 1 − 0.5 − 0.4 = 6.1 million	C above 6.1
Retirement delay d	5 months	(8 − 5.5 − 1 − 0.4) / 0.1 = 11 months	d above 11, within the stated modeled range
Attainable saving fraction a	1.0	(5.5 + 1 + 0.5 + 0.4) / 8 = 0.925	a below 92.5%
Licensed finance users n	400	(8 − 5.5 − 1 − 0.5) / 0.001 = 1,000	n above 1,000

At each boundary the modeled result is zero, not already negative. These are five-year financial thresholds for the finance proposal; they are not inferred from the student-records project’s three-year spending. They also differ from a threshold that merely consumes the first two years of savings. Joint changes can matter before either individual threshold is reached: a = 0.95 and d = 8, with C and n unchanged, give 7.6 − 5.5 − 1 − 0.8 − 0.4 = −0.1. Check whether such combinations are plausible, how uncertain the inputs are, and whether another feasible option is preferable. Thresholds alone provide no probability of failure. A binding obligation or valuable noncash outcome still needs its own assessment; crossing a cash threshold is a reason to review alternatives, not automatic permission to abandon the service.

After delivery, preserve the approved baseline while updating forecasts separately. The GOV.UK Service Manual defines cost per transaction using all-channel service costs, including assisted support, divided by completed transactions. Under that guidance, the first quarterly reports use cumulative 3-, 6-, 9-, and 12-month data from public beta; later quarterly reports use the preceding 12 months. An internal service may choose another justified window, but should state it rather than calling any quarterly number the same measure.

Report total cost alongside unit cost and compare equivalent work, quality, staffing, and accounting boundaries. A cost of 100 across 100 transactions is 1 each; 110 across 220 is 0.5 each, although total expenditure rose. Neither this movement nor a before-and-after comparison alone identifies the system’s causal contribution. Released staff capacity is distinct from reduced payroll, and an avoided future payment needs a supported comparison. Verify actual changes in paid contracts, overtime, or retired resources and account for transition and continuing costs before claiming financial savings.

Judging a budget and a cost model

Review whether planned spending and forecasts reconcile to owned cost lines, whether scope and timing are comparable, and whether exclusions are explicit. Economic value, required outcomes, accounting treatment, and affordability belong in the approval decision together. Check the evidence behind reference-class adjustments, mitigation, reserves, and sensitivity assumptions, including overlap between risk allowances. A reproducible model makes the claim reviewable; it does not make it true. Keep observations, attributed effects, and realized cash changes distinct when comparing outcomes with the original case.

What a reviewer should ask

1. Why was this evaluation horizon chosen, and how are costs, benefits, obligations, and residual value beyond it treated?
2. Which of the usually omitted items are priced, which are excluded, and who priced the staff time?
3. How long will both systems run, who estimated that, and from what comparable migration?
4. What is the license’s counting unit in the contract, and over what population is it forecast?
5. What does it cost to leave, and to retire the old system with its retention obligations?
6. Which shared costs does this create or consume, and under what written key are they allocated?
7. Which historical estimates support the adjustment, what mitigation justifies reducing it, and how were overlapping risks excluded from the reserve?
8. Which decision criterion does each switching value test, under what fixed assumptions, and what joint changes could reverse the result?
9. What measurement window and comparison distinguish observed unit cost, attributed benefit, and realized cash savings?

In one continuation of the fictional finance case, implementation finishes at 5.0 million against the 4.2 core budget and 1.3 reserve: 0.8 of the reserve is used and 0.5 remains. The council reports that outcome against the implementation envelope, not against the whole five-year TCO or the student-records project’s nine-million total. The central platform has an operating owner and a documented allocation rule, and records staff test retained access before retiring an old path. At the seventh quarterly review, the preceding 12 months of transaction data meet the unit-cost target. Finance also reports total spending, workload mix, quality, and remaining uncertainty about attribution. The implementation is within its funding envelope; the five-year saving claim remains under review. Those are separate conclusions, with different evidence and owners.

References:

• HM Treasury, The Green Book (2026), chapter 6 (Open Government Licence v3.0).
• HM Treasury, Green Book supplementary guidance: optimism bias (PDF), table 1 and application steps; based on Mott MacDonald’s 2002 study (Open Government Licence v3.0).
• GOV.UK Service Manual, Measuring cost per transaction, updated February 2021 (Open Government Licence v3.0).
• GOV.UK, Define your purchasing strategy (Technology Code of Practice point 11; Open Government Licence v3.0).
• Google Cloud, Migrate to Google Cloud: Assess and discover your workloads (CC BY 4.0).
• Google Cloud, Migrate to Google Cloud: Minimize costs (CC BY 4.0).
• FinOps Foundation, Allocation (CC BY 4.0).
• FinOps Foundation, Invoicing & Chargeback (CC BY 4.0).