On this page
An AI proposal can pass its technical and risk tests and still be the wrong thing to fund. The model can work and the risk review can come back clean while the budget and people it needs would have produced more value on other work. Whether a proposal is the best available use of that money and capacity is a Strategic Portfolio Management decision, and an AI proposal can avoid it entirely if it never goes through the portfolio.
AI strategy miscalculation is now on the risk register
Gartner's third-quarter 2026 Emerging Risk Report, announced on November 2, surveyed 308 enterprise risk management leaders, auditors and senior executives. AI strategy miscalculation appeared in the report for the first time and ranked as the third most concerning emerging risk, behind AI discovery of cyber vulnerabilities and geopolitical energy supply shocks. Gartner describes it as investing in AI too aggressively or selecting unsuitable use cases, including moving ahead before the organization can manage the consequences.
The worry behind the ranking is that organizations could make major decisions based on expected benefits that fail to materialize. Gartner's advice is that organizations "should evaluate AI as a portfolio of business decisions", which for a portfolio team means putting AI proposals through the same intake for comparison and staged funding as any other investment.
AI spending can grow outside the portfolio view
AI work can add up to a large total without any single item reaching the investment committee. A pilot funded from a business unit's discretionary budget may fall below the threshold for executive review, which is a sensible guardrail for small work, as we set out in Rethinking Approvals: How Guardrails Enhance Agility. AI features added to an existing software agreement, or model usage billed to a cloud account, may never be recorded as an initiative at all. Together, these decisions can produce AI spending that has not been compared with other work.
A pilot may have a capped budget and review period, but once it reaches production the system carries running costs for compute and usage and for the people who check its output. McKinsey's November 2026 article for CFOs notes that this human oversight can be a significant ongoing expense. Those costs continue during operation and compete for the same funding as the rest of the portfolio, and our review of the first half of 2026, Enterprise Portfolio Management Trends: 2026 So Far, found AI budgets were more commonly reallocated than added in the organizations we speak with. The investment review should therefore establish what other spending would fund an AI system in production.
AI proposals belong in the same intake as other work
AI proposals should enter the intake the portfolio already uses. Each proposal should arrive with the problem it solves and the baseline it will be measured against, the business unit that owns the benefit, the implementation cost and expected running cost at full use, with a measurable outcome. A short initial review at this point lets the investment committee reject weak proposals before the team prepares a full business case, a step we describe in Why Business Cases Fail and How to Fix Them.
The benefit owner matters because the team that builds an AI tool is not always the team whose results it is meant to improve. An AI initiative, like any other, should have an accountable benefit owner in a single business unit, the one where most of the benefit lands. We explain why in Is Transformation Ever Accountable for Delivery?
The comparison should then run across the whole portfolio rather than only across AI proposals. An AI assistant for accounts payable might look strong next to other AI ideas and weaker next to finishing the supplier data cleanup it would depend on. Forrester argued in November 2026 that planning should narrow a long list of AI use cases down to a few investment portfolios, with the CEO deciding what gets deprioritized to pay for them.
An AI governance review assesses whether a system can be deployed safely and lawfully, and that review belongs in the process, but its approval does not establish whether the investment deserves the money more than the alternatives.
What an AI proposal records at intake
Example: an AI assistant that matches supplier invoices to purchase orders in accounts payable
Recorded before the initial review- Problem and baselineInvoices that need manual matching today, and how long each one takes, measured before any work starts
- Benefit ownerThe head of accounts payable, whose team's results the assistant is meant to improve
- Change costBuild or license, integration with the finance system, and training for the team
- Running cost at full useCompute and usage charges at production volume, plus the staff time spent checking the output
- Outcome measureShare of invoices matched without manual work, reported by accounts payable
- Compared againstOther work competing for the same budget and people, including the supplier data cleanup the assistant depends on
A readiness check comes before the money to scale
Readiness should be validated before scaling. In portfolio terms this is a check between the pilot and the larger funding needed to roll it out. It covers whether the data and definitions the use case depends on exist, what the system will cost to run at production volume, whether the governance review is complete with someone named to check the output, and whether the teams whose work will change have the capacity to change it.
How much data readiness to demand depends on the use case. AI that reports on portfolio health needs agreed definitions first, because a summary built on inconsistent status data reads as authoritative while hiding the inconsistency, the problem we set out in You Can't AI Your Way Out of a Data Problem. A tool that checks supplier invoices against contract terms could start with the data already available if it is sufficient to test the outcome, consistent with McKinsey's guidance for finance teams. The readiness check should ask what this particular use case needs rather than apply one data standard to every proposal.
Funding should follow evidence in stages
Approving the full rollout of an AI system at the same time as the pilot can weaken the funding checkpoint at which the evidence should be assessed. Drawdown, releasing funding in tranches tied to evidence, suits AI because the initial commitment can be limited to testing the case. The pilot tranche pays for a test against the baseline. The scale tranche is released once the pilot has demonstrated the agreed improvement and the readiness check is complete, and each later portfolio review asks whether the running cost still buys more value than the alternatives.
Each of these reviews is also where the organization can decide to stop or change course. Agreeing at approval what result would end the work, and measuring value by its effect on business decisions as well as on hours saved, gives each review a clear test to apply.
How funding for AI work follows the evidence
| Step | What is tested | Decision |
|---|---|---|
| 1Intake and initial review | Problem, baseline, benefit owner, change and running cost | Build the full case, reshape it, or reject it |
| 2Portfolio comparison | Value against other work competing for the same money and people | Fund the pilot or defer it |
| 3Pilot tranche | Improvement against the baseline by an agreed review date | Proceed to the readiness check, or stop |
| 4Readiness check | Data the use case needs, running cost at volume, governance sign-off, capacity of the teams that will change | Release the scale tranche, or close the gaps first |
| 5Scale tranche | Rollout against the outcome measure agreed at intake | Continue the rollout, adjust it, or stop |
| 6Portfolio review | Whether the running cost still buys more value than the alternatives | Keep funding, change it, or retire it |
- 1. Intake and initial reviewTested: Problem, baseline, benefit owner, change and running costDecision: Build the full case, reshape it, or reject it
- 2. Portfolio comparisonTested: Value against other work competing for the same money and peopleDecision: Fund the pilot or defer it
- 3. Pilot trancheTested: Improvement against the baseline by an agreed review dateDecision: Proceed to the readiness check, or stop
- 4. Readiness checkTested: Data the use case needs, running cost at volume, governance sign-off, capacity of the teams that will changeDecision: Release the scale tranche, or close the gaps first
- 5. Scale trancheTested: Rollout against the outcome measure agreed at intakeDecision: Continue the rollout, adjust it, or stop
- 6. Portfolio reviewTested: Whether the running cost still buys more value than the alternativesDecision: Keep funding, change it, or retire it
Shaded steps are where money is released.
Which AI work gets funded is a portfolio decision
An AI proposal deserves funding when it compares well against the other uses of the same money and capacity, and when an owner in the business unit will be measured on its result. Rollout funding should wait until the pilot has shown its improvement and the readiness check for that use case is complete.
An inventory can capture the AI work already running outside the portfolio, including pilots on discretionary budgets and usage on existing contracts, with an owner and expected running cost recorded for each commitment. Once that work is visible, the investment committee can compare the next AI proposal with existing commitments and judge whether it deserves the same money and capacity.