AI Total Cost of Ownership Framework

An AI total cost of ownership (TCO) framework should explain what the organisation will pay, which internal capacity it will commit and which costs will remain if a workflow changes or ends. For a portfolio of AI workflows, the difficult part is often shared infrastructure and responsibility. Counting the same connector or evaluation environment in every business case can overstate the portfolio; omitting it from every case understates it.

This guide is for Saudi finance leaders, technology owners and portfolio managers building a lifecycle view across several AI capabilities. It provides two reconciled views: the total cost of the portfolio and the incremental cost of a particular decision. The worked example is fictional and uses declared Saudi-riyal assumptions, not supplier quotations or market benchmarks.

Choose the cost object and the decision horizon

Name the object being costed. It might be a workflow, a shared capability or a portfolio. Then identify the decision: approve a new workflow, compare an operating arrangement, renew a platform or retire a service. The right cost view depends on the question. A fully allocated workflow cost can help explain portfolio consumption, while the incremental cost of adding that workflow answers a different question.

Set an explicit horizon and start date. Include implementation, operation, planned change and exit activities that occur within it. If an obligation continues beyond the horizon, disclose that tail rather than implying the cost stops at the spreadsheet boundary. Use a consistent treatment of timing, currency, taxes and financing across alternatives, with the applicable assumptions reviewed by the responsible finance team.

Keep previous expenditure separate from future decision costs. A past experiment may be relevant to understanding the programme's history, but it cannot be avoided by rejecting tomorrow's purchase. Record sunk expenditure where useful without letting it distort the forward comparison. Similarly, a shared contract already committed for the horizon may have different incremental consequences from a genuinely expandable service.

Build a cost dictionary before the spreadsheet grows

Give each cost a stable identifier, owner, time period, source and treatment. Identify whether it is one-time, recurring or contingent; direct or shared; cash or internal capacity; committed or avoidable. These labels describe different dimensions. A recurring cost can be fixed for the current contract, and a one-time cost can still be incremental to the decision.

Cost dictionary for an AI portfolio

Cost family	Examples to investigate	Key ownership question
Establish	Design, data preparation, workflow construction, testing and transition	Which approved release creates this commitment?
Operate	Subscriptions, metered resources, support and retained review	Who monitors the driver and authorises changes?
Maintain	Source updates, evaluation refresh and planned compatibility work	What event or calendar requirement triggers the work?
Share	Identity connection, common environments and portfolio coordination	How is the cost recorded once and allocated?
Exit	Exports, transition assistance, retirement and record handling	What remains payable or operationally necessary after use ends?

Use implementation costing to estimate establishment work, automation operating costs for completed-process demand and managed-service pricing for support boundaries. The TCO ledger combines those inputs without replacing their detailed assumptions. Preserve a reference to each underlying estimate so a changed driver can update the relevant line.

Record cash and internal capacity in separate ledgers

Cash records payments or other defined financial commitments under the chosen model. Internal capacity records work performed by existing people, such as source approval, evaluation, review and coordination. Multiplying hours by a planning rate can provide an economic comparison, but it does not automatically create an additional payroll payment or a cash saving.

Keep the hour assumptions visible even if management requests one combined resource total. A portfolio may fit the external budget while exceeding the availability of the only qualified reviewer. Conversely, a team with spare relevant capacity may absorb some work without increasing cash expenditure. The model should show both facts rather than treating time as either free or automatically incremental payroll.

Define what the planning rate includes and avoid charging the same labour twice. If a supplier fee already covers a task, do not add the supplier's estimated hours as another internal line. If an employee is funded through a separately counted project charge, reconcile that treatment with finance before adding their capacity value again.

Work through a three-workflow portfolio

Consider a fictional company deploying three assisted workflows: A checks supplier-document completeness, B prepares service-email routing suggestions, and C produces a maintenance-priority briefing. People retain the respective acceptance and operational decisions. The workflows share an identity connector and evaluation environment, while each has distinct setup and recurring costs.

The horizon is three operating years after initial setup. Figures are nominal teaching assumptions in Saudi riyals. The model excludes unlisted taxes, financing and unrelated corporate overhead. It is a planning illustration, not an accounting policy or a claim that these amounts suit any real Saudi organisation.

Illustrative external cash assumptions

Item	A	B	C	Shared
Initial setup	24,000	18,000	12,000	36,000
Recurring per year	18,000	12,000	9,000	24,000
Planned refresh in year two	0	0	0	9,000
Exit at end of year three	4,000	3,000	2,000	12,000

Initial cash is SAR 90,000. Annual recurring cash is SAR 63,000, comprising SAR 39,000 direct and SAR 24,000 shared. Three-year recurring cash is therefore SAR 189,000. Add the SAR 9,000 refresh and SAR 21,000 exit: total portfolio cash is 90,000 + 189,000 + 9,000 + 21,000 = SAR 309,000.

The timing view is SAR 90,000 at setup, SAR 63,000 in year one, SAR 72,000 in year two and SAR 84,000 in year three. These periods reconcile to the same SAR 309,000. Showing the timing prevents a neat average from hiding the initial funding requirement or the final transition work.

Add the retained capacity without calling it new cash

Assume initial internal work of forty hours for A, thirty for B, twenty for C and thirty shared: 120 hours in total. Annual retained work is twenty-four hours for A, eighteen for B, twelve for C and forty-two shared: 96 hours annually. End-of-horizon exit requires another twenty internal hours.

At an invented planning value of SAR 150 per hour, setup capacity is SAR 18,000. Three years of recurring capacity is 3 × 96 × 150 = SAR 43,200. Exit capacity is SAR 3,000. The total capacity value is SAR 64,200, producing a combined planning-resource view of SAR 373,200 when added to the SAR 309,000 cash ledger.

The company must still schedule the 428 internal hours represented by 120 + 3 × 96 + 20. The rate does not prove that those hours are available. Identify the tasks and people behind the shared forty-two annual hours, particularly when several workflows need evaluation at the same time. Use the capability and skills assessment to distinguish a missing specialist from a simple scheduling gap.

Allocate shared cost without changing the portfolio total

FinOps's cost-allocation guidance calls for an explicit strategy for identifying shared costs and documenting how they are apportioned. The practical requirement here is reconciliation: allocation changes which workflow carries a cost, not whether the portfolio incurred it. Retain the original shared ledger and the allocation rule so the view can be explained and changed.

For the example, allocate the SAR 36,000 initial shared setup equally: SAR 12,000 per workflow. Allocate annual shared operation by an assumed driver of 50% to A, 30% to B and 20% to C, giving SAR 12,000, SAR 7,200 and SAR 4,800 annually. Allocate the shared refresh and shared exit equally: SAR 3,000 and SAR 4,000 respectively per workflow.

Reconciled three-year fully allocated external cash

Component	A	B	C
Direct and allocated setup	36,000	30,000	24,000
Three years direct and allocated operation	90,000	57,600	41,400
Allocated refresh	3,000	3,000	3,000
Direct and allocated exit	8,000	7,000	6,000
Total	137,000	97,600	74,400

The totals sum to SAR 309,000. These allocation choices are illustrative, not a universal fair-sharing formula. A different driver may better represent actual consumption or responsibility. Agree the purpose of allocation: management visibility, internal charging or investment comparison. Avoid using a convenient percentage as if it proved what cost would disappear when a workflow is removed.

Use incremental cost for an incremental decision

Suppose A and B already justify the shared platform and the organisation is deciding whether to add C before committing its direct setup. Assume shared capacity is sufficient and every shared cost would remain the same without C. C's incremental external cash over the horizon is then SAR 12,000 setup + 3 × SAR 9,000 operation + SAR 2,000 direct exit = SAR 41,000.

That is different from C's SAR 74,400 fully allocated view. Neither number is inherently wrong. The first answers what extra cash the stated decision creates; the second assigns a share of the whole portfolio to C. Using the fully allocated amount as avoidable cash would overstate savings under the assumptions. Using only the incremental amount for every workflow would leave shared costs unexplained.

Test the spare-capacity assumption. If adding C instead requires a genuine SAR 3,000 annual shared-platform uplift, its incremental three-year cash rises by SAR 9,000 to SAR 50,000. Add any other causally required costs. Do not hold shared costs fixed merely to make the new workflow look attractive, and do not allocate unrelated committed spending as if C caused it.

Expose stranded costs when a workflow ends

A workflow can stop while some costs continue. A minimum subscription, retained environment, contracted support period or required transition activity may remain. Identify what can be cancelled, when it can be cancelled and what work is needed to end it. This matters particularly when a portfolio business case assumes that unsuccessful experiments can be stopped immediately at no further cost.

Separate sunk cost from continuing obligation. Past setup is already spent; a future committed payment is still cash to plan. Reallocating a shared platform cost to the remaining workflows does not create a saving for the portfolio. It may simply make their reported unit costs rise because fewer services share the same fixed commitment.

Inspect software pricing rules for commitments and unused allowances, and capability ownership for practical replacement options. A financially attractive exit assumption needs an operational route: records, responsibilities and accepted work must move or be retired deliberately. Keep required records and access decisions within the relevant organisational process rather than treating shutdown as a billing-only action.

Use sensitivity tests tied to real decisions

Change one major driver at a time before combining scenarios. Test workload, review effort, service coverage, licence commitments and a specific refresh requirement. Explain which line moves and why. A general percentage contingency can be useful for an agreed planning purpose, but it should not replace investigation of a known dependency that can be estimated directly.

For portfolio decisions, test whether shared capacity grows smoothly or in steps. The fourth workflow may require a new environment even when the first three fit the existing arrangement. Also test whether a shared change affects every workflow at once. A source interface change can create coordinated testing work that separate business cases fail to recognise.

Keep uncertainty distinct from the base case. Record the triggering event, range, owner and response for each material contingency. Do not add the most expensive outcome of every scenario and label it the expected cost. Conversely, do not omit a plausible contractual obligation because the team hopes not to use it.

Reconcile the model to contracts, work and rollout dates

Run three boundary checks before approving the total. First, trace each external line to a quotation, contract assumption or clearly labelled estimate. If a service fee includes the evaluation environment, the same environment should not reappear as a separate external charge without a reason. Preserve the bundle description and split it for analysis only when the parts still reconcile to the original total.

Second, trace internal effort to tasks. A shared review meeting should not be counted in full under every workflow and then again in portfolio coordination. Record the common event once and allocate its effort if useful. Conversely, a meeting attended by three people consumes three people's time; calling it one event does not mean counting only one participant's hours. State the measurement convention so both omissions and duplication can be found.

Third, check dates against the rollout plan. The worked example assumes all three workflows operate for the same three years. If C instead starts in year two within the unchanged portfolio horizon, it has two operating years, not three. Under otherwise unchanged assumptions, C's direct recurring cost falls by SAR 9,000 and portfolio cash becomes SAR 300,000. The initial setup merely moves in time; it does not disappear.

That shorter operating period is not automatically an efficiency gain. The business receives C's service for one fewer year, and any expected benefit must use the same period. Similarly, a delayed shared component may hold back several workflows. Record the timing consequence rather than improving the cost case by shortening its operating period while retaining three years of benefit.

Maintain a compact reconciliation record with cost identifier, commercial source, operating owner, start and end dates, included work, allocation rule and last verification. Mark unresolved estimates explicitly. If a line is removed, record whether the obligation ended, moved to another line or was discovered to be a duplicate. Deleting it without that explanation weakens the audit trail of the decision.

Ask finance and the operating owner to inspect the same record together. Finance can confirm the commitment and timing; the operating owner can confirm what must actually happen. Neither view alone catches every problem. A technically complete service can have an omitted renewal, and a financially reconciled invoice can still exclude the internal work needed to use its output.

Keep cost identifiers stable when a line is renamed or its owner changes. If one line splits into several, preserve the relationship to the original item. Explain whether a movement reflects a changed quantity, rate, scope or allocation, so a later reviewer can understand the difference without reconstructing the entire model.

Maintain the ledger through renewal and retirement

Assign a cost owner who can reconcile the financial records with the operating inventory. Review new workflows, retired services, changed pricing units, unused commitments and shared-driver assumptions. Preserve the previous version so management can see whether a forecast changed because of volume, scope, a rate, an error or a different allocation policy.

Track realised costs against their original purpose. If the planned evaluation refresh is postponed, show the remaining obligation and the operational consequence rather than presenting the underspend as efficiency automatically. If a shared component is reused, distinguish avoided new setup from a reduction in already committed cash.

A decision-ready TCO summary shows portfolio cash, internal capacity, timing, allocation rules, incremental decision cost and material remaining uncertainties. The Saudi AI authority guides connect these financial questions with adoption and architecture. Open for AI consulting can help frame the investment, while integration and delivery can scope shared technical work. Discuss your AI portfolio cost boundary with the services and commitments that must be counted once.