A 2027 AI budget needs room for the systems already operating, the work required to improve them and the evidence needed to choose new investments. Funding only the next wave of pilots leaves those obligations competing for people after the budget has been approved.
Plan the portfolio around delivery and operating capacity. Give each proposed investment a business owner, a defined next decision and an explicit claim on shared resources. Hold some capacity and money for what the organisation learns. The aim is a portfolio the team can evaluate, deliver and support throughout the year.
This framework uses evidence available through 28 September 2026. It is a planning method for 2027, not a report of Q4 outcomes or a forecast of next year's returns.
Use sector growth to set context
The commercial scale of AI is visible in supplier reporting. NVIDIA's August results release reported US$89 billion of Data Center revenue for the fiscal quarter ended 26 July 2026, up 117% year on year. Microsoft's July release, accompanying June-ended results, said Microsoft 365 Copilot exceeded 30 million paid seats.
These are meaningful supplier-demand and purchasing signals. They do not tell an executive how many additional projects their organisation should fund. Paid access, active use and correctly completed work still need to be examined inside the business.
There is also evidence of cost scrutiny alongside adoption. KPMG's September 2026 U.S. AI Pulse release reported that 74% of respondents included cost reviews in AI approval processes. The survey covered 314 U.S. executives at organisations with at least US$1 billion in annual revenue, with July–August fieldwork. It offers context from that population, not a Singapore or ASEAN benchmark.
The practical implication is to make the portfolio legible: what is being funded, what it produces, what it consumes and when the next decision is due.
Begin with existing commitments
Build an inventory of live workflows, paid tools, pilots and signed commitments. For each, record the business owner, users, outcome measure, renewal date, recurring spend and support requirement. Include products purchased through departmental budgets and capabilities bundled into broader contracts.
Then identify overlaps. Two teams may be buying similar document-processing capability; a new proposal may duplicate a feature in an existing platform. Consolidation is worth evaluating, but do not assume it is free. Migration, retraining, contract terms and differences in controls can make a nominal licence saving expensive.
Treat operational health as part of the inventory. An application with no effective owner, unreliable source data or unresolved incidents already has a claim on next year's capacity. Hiding that work behind a small licence line understates the cost of the portfolio.
Allocate money by the work it must support
Separate ongoing operation, bounded new commitments, shared capabilities and flexibility. This makes it easier to see whether the organisation is funding more promises than it can fulfil.
The following is an illustrative allocation of a S$600,000 incremental budget envelope. It is not a recommended spending ratio or a market benchmark. The organisation would still need to identify baseline staff and platform costs outside this envelope.
| Budget purpose | Illustrative allocation | Evidence needed to use it |
|---|---|---|
| Operate and improve approved workflows | S$220,000 | Usage, service performance, support plan and accountable owner |
| Evaluate and deliver new bounded commitments | S$160,000 | Individual investment memos with capped stages and acceptance evidence |
| Shared delivery capabilities | S$120,000 | Named projects that depend on the capability and a measurable benefit |
| Reserve for expansion or unexpected work | S$60,000 | A new decision showing why the priority changed |
| Migration and orderly closure | S$40,000 | Contract, record-retention and continuity obligations |
| Total | S$600,000 |
Shared work might include a reusable evaluation process, a common integration or a reliable source-data service. Require a specific demand for it. A broad internal platform programme can absorb the budget while the business workflows it was meant to support remain blocked.
The reserve should have an owner and a release process. It preserves the ability to expand a strong result or respond to a material operating issue without pretending those decisions are known at the start of the year.
Check the people constraint separately
Cash does not describe all the scarce resources. Business specialists, security reviewers, data owners and experienced engineers can become the limiting factor even when funding is available.
For example, four illustrative projects might each need 20 engineer-weeks and six reviewer-weeks. A plan with 80 engineer-weeks appears fully resourced until it is compared with only 16 reviewer-weeks. The projects require 24 reviewer-weeks. The resulting gap cannot be solved by repeating the budget approval.
If those 16 reviewer-weeks are the total available in the planning period, the portfolio owner must reduce scope, secure another eight reviewer-weeks or defer work beyond that period. Sequencing helps with overlapping peaks; it does not remove a total capacity deficit. Named people should validate estimates against their other responsibilities. Do not count the same finance expert's time in several projects as though it were independently available to each.
Country rollout adds another constraint. A regional implementation may reuse technology while still requiring local process ownership, data assessment and acceptance evidence. Schedule those decisions as work. Treating them as a final sign-off can leave an otherwise complete system waiting for months without an approved operating arrangement.
Give each investment a next decision
A portfolio review becomes useful when it can change commitments. Assign each item a stage and state the evidence needed to advance, remain at its current scale or close.
| Stage | Question for the portfolio review | Possible decision |
|---|---|---|
| Problem assessment | Is there a valuable process with accessible evidence and an owner? | Fund measurement, defer or close |
| Bounded evaluation | Does the approach meet the agreed quality and economic requirements? | Fund delivery, revise the approach or stop |
| Initial operation | Can the team sustain performance, controls and support? | Improve, hold scope or authorise expansion |
| Wider operation | Are benefits and operating costs still acceptable? | Continue, consolidate, renegotiate or retire |
A stopped project can be a good use of an evaluation budget if it resolves an expensive uncertainty early. Record what was learned and any reusable assets, but do not invent a financial return from “learning” to make every line look successful.
Review renewal dates and dependencies alongside performance. A weak product might need a managed exit; a promising project might need to wait for a shared data service. Those constraints belong in the decision, rather than in a separate technical report that the budget owner never sees.
The first 90 days
In the first month, reconcile the inventory and agree definitions for cost, completed work and benefits. In the second, challenge the largest assumptions in the leading proposals and confirm delivery capacity. In the third, approve a sequenced set of commitments, publish the review dates and retain a visible reserve.
Use the investment and portfolio records: one page per workflow for the investment decision, then a single table for money, people, dependencies and next decisions. The executive team's task is to choose the combination it can operate well. A shorter list with funded ownership and credible evidence can support a more ambitious business outcome than a budget filled with projects competing for the same few people.
Read our editorial principles for evidence, contribution and correction standards.