Plan a 2027 AI budget around a short list of use cases, each with an owner, a success measure and a date to stop or scale, then fund what those use cases depend on: clean data, integrations, access control, testing and running costs. Budget the foundations as their own line, not as leftovers from the pilot.

Picture a head of operations asked to put "AI" into next year's plan. This year the company paid for a chatbot pilot, some assistant licences and a consultant's workshop. The pilot impressed in the demo, then stalled because the answers depended on data spread across three systems and nobody owned the cleanup. The easy move is to ask for more of the same. A better one is to ask what the pilot was missing, and pay for that first.

What should a 2027 AI budget pay for?

A 2027 AI budget should pay for a few use cases that can reach production, and for the shared foundations they need to get there. Licences and model usage are the visible part. The data, integration, governance and support work underneath is usually larger and is what decides whether anything ships.

It helps to name the parts:

• Use cases are specific jobs AI will do, such as drafting invoice matches or answering order-status questions.

• Foundations are the shared things every use case relies on: data quality and definitions, integrations, access control, evaluation and monitoring.

• Running costs are what it takes to keep a use case live after launch: model usage, hosting, support and the people who review its output.

• Change costs are the training, process redesign and time teams need to work with the new system.

A budget that funds only the first item tends to produce pilots. One that funds all four produces systems people use.

What are analysts saying about AI budgets for 2027?

Analysts expect AI spending to keep rising in 2027, and they are increasingly clear that more spending alone won't produce results. The advice has moved from "start experimenting" to "fix the foundations and stop the pilots that can't scale".

Gartner's September 2026 forecast puts worldwide AI spending at $2.7 trillion in 2026, up 49.5%, with a further 36.2% rise forecast for 2027. Most of that is infrastructure and AI built into software and devices companies buy anyway, so it says little about what your own budget should be.

Forrester's 2027 Budget Planning Guides, published on 15 July 2026 and based on a survey of more than 2,600 business and technology decision-makers, are more useful for planning. Forrester recommends increasing spend on "enterprise context that agents can act on", meaning governed, machine-readable knowledge about policies, processes and systems. It recommends cutting "tech debt that slows AI and productivity" through targeted data fixes, and cutting "AI pilots that scale activity without organizational readiness", meaning initiatives without governance, clear ownership, success criteria or a path to scale. Its chief research officer, Sharyn Leaver, put it simply: "The organizations that outperform in 2027 won't be those that spend the most on AI."

Why do AI pilots stall before production?

AI pilots usually stall because the work around the model was never funded: the data isn't ready, nobody owns the result, and costs and risks only show up at scale. The model itself is rarely the blocker.

Gartner has put numbers on two of these causes. In February 2025 it predicted that through 2026, organisations will abandon 60% of AI projects unsupported by AI-ready data. In June 2025 it predicted that over 40% of agentic AI projects will be cancelled by the end of 2027, "due to escalating costs, unclear business value or inadequate risk controls."

Each of those causes maps to a budget line that was missing:

• Data not ready: no money for cleanup, definitions or a semantic layer.

• Escalating costs: model usage and review time were never estimated at production volume.

• Unclear value: no baseline was measured before the pilot started, so nobody can show it worked.

• Inadequate risk controls: access control, logging and testing were left for later.

Our post on getting BI ready for AI agents with a semantic layer covers the data side in detail.

How should you split an AI budget between foundations and use cases?

Split it by dependency: fund the foundations that your chosen use cases need in 2027, and no more. Foundations bought without a use case to serve become shelfware, and use cases funded without foundations become stalled pilots.

A practical way to lay out the budget is in four lines:

• Foundations: What it covers: data cleanup and definitions, integrations, identity and access, evaluation and monitoring. When it pays off: across every use case that shares them. Typical mistake: funding a company-wide data programme instead of the parts the first use cases need.

• Use cases: What it covers: design, build, testing and rollout of each use case. When it pays off: when the use case reaches production and is used. Typical mistake: too many at once, none of them finished.

• Running costs: What it covers: model usage, hosting, vendor licences, support and human review. When it pays off: every month the use case is live. Typical mistake: estimating from pilot volume, not production volume.

• People and change: What it covers: training, process redesign, an owner for each use case, time for teams to adapt. When it pays off: when the process changes, not just the tool. Typical mistake: treating it as optional.

The right proportions depend on how ready your data and systems already are, so be wary of anyone who quotes a fixed ratio without having looked.

How do you size an AI budget for 2027, step by step?

Size it bottom up, from a short list of use cases, rather than top down from a percentage of revenue or IT spend. Each use case gets its own estimate, and shared foundations are costed once.

1. List candidate use cases. Collect them from the teams doing the work, not only from leadership.

2. Measure the baseline. For each, record how the work is done today: time, cost, error rate, volume. Without a baseline you cannot show improvement.

3. Pick a few. Choose the ones with a clear owner, available data and low risk if the output is wrong. Start with work where a person reviews the result.

4. Map the dependencies. For each chosen use case, list the data, systems and access it needs, and what is missing.

5. Cost the foundations once. Group the shared gaps, such as one customer data cleanup used by three use cases, and estimate them together.

6. Estimate running costs at production volume. Use the vendor's current price pages and your real volumes. Add the time people will spend reviewing output.

7. Set stop-or-scale dates. Give each use case a review date and the result it must show by then.

8. Hold a reserve. Keep part of the budget unallocated for the second half of the year, when you know which use cases earned more.

If you are choosing models for these use cases, Claude, GPT or Gemini? Choosing a Model for Business Workflows explains how to compare them by task.

When should you stop funding an AI project?

Stop funding an AI project when it misses the result it agreed to by its review date and the reason is not a fixable gap. Stopping a pilot on schedule is part of good budgeting, not a failure.

Signs it is time to stop

• No one uses it without being asked to.

• The result needs so much review that the time saved is gone.

• The data it depends on cannot be fixed within the year's budget.

• Its running cost at real volume is higher than the work it replaces.

• No business owner is willing to sign off on it.

Signs it deserves more

• It met its success measure, and the owner wants it in more teams.

• Its failures trace back to one foundation that, once fixed, would also help other use cases.

How we approach it at Vizio AI

We start with the work, not the technology. Together with the teams involved, we list the jobs AI could take on, measure how they are done today, and pick the few with a clear owner, usable data and a manageable risk if the output is wrong.

For each one we map what it depends on: the data, the systems, the access rights and the review step. Shared gaps become one foundation project instead of three. We then build the first use case on those foundations with an evaluation set and logging from day one, and agree up front on the result that would justify scaling it, and the result that would mean stopping. When agents need to reach business systems, we set access control first, as described in Connecting an AI Agent to Your CRM? Secure the MCP Server First.

This is part of our DX Studio work.

This post is general guidance for planning, not financial advice.

Sources

• Forrester, Forrester's 2027 Budget Planning Guides (15 July 2026): https://www.forrester.com/press-newsroom/forresters-2027-budget-planning-guides/

• Gartner, Gartner Forecasts Worldwide AI Spending to Grow 49.5% in 2026 (September 2026): https://www.gartner.com/en/newsroom/press-releases/2026-09-16-gartner-forecasts-worldwide-ai-spending-to-grow-49-point-5-percent-in-2026

• Gartner, Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027 (25 June 2025): https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027

• Gartner, Lack of AI-Ready Data Puts AI Projects at Risk (26 February 2025): https://www.gartner.com/en/newsroom/press-releases/2025-02-26-lack-of-ai-ready-data-puts-ai-projects-at-risk

• CIO.com, AI spend will jump 49.5% in 2026, says Gartner: https://www.cio.com/article/4225785/ai-spend-will-jump-49-5-in-2026-says-gartner.html