Executive Guide · Investment

Making Better AI Investment Decisions

What leaders should challenge before committing capital, including business value, readiness, dependencies, implementation risk and measurable return.

Read the Guide

Executive Summary

AI investment cases are frequently approved with less scrutiny than a comparable capital request in almost any other category, not because leaders are careless, but because AI proposals are unusually good at sounding rigorous while omitting the questions that matter most.

A well-produced AI business case can include a technical architecture diagram, a vendor comparison, and a confident projected return, and still leave out the three things that actually determine whether the investment will succeed: whether the organisation is genuinely ready to use what it is buying, what has to be true elsewhere in the business for the projected value to materialise, and how the return will actually be measured once the system is live. Independent research on AI investment consistently finds a wide gap between reported spend and reported enterprise-level value, a gap that traces, in large part, back to business cases that were approved without these questions being properly pressed.

This guide sets out the questions a leadership team should put to any AI investment proposal before committing capital, not to slow good ideas down, but to catch the ones that look strong on paper and would not survive contact with the organisation's actual data, processes and people.

Challenge the Business Value Before the Technology

The first and most important question is rarely asked directly enough: what specific business outcome does this investment produce, and what happens if it does not?

A strong business case names the metric it intends to move, using the four-category framework set out in Where AI Can Create Real Business Value, states a target, and states the baseline the target is measured against. A weak business case describes what the technology does, in technical or feature-list terms, and leaves the leadership team to infer the business value for themselves. The second kind of proposal is far more common than it should be, largely because it is easier to write convincingly: describing a capability is straightforward, while committing to a measurable, falsifiable business outcome is a genuine act of accountability that not every sponsor is eager to make.

A useful discipline for any capital committee: decline to approve an AI investment proposal that cannot state, in one sentence, which specific metric will move, by roughly how much, and by when. If the sponsor cannot answer that question with any precision, the investment is not yet ready for capital; it is ready for a smaller-scale pilot designed specifically to answer it.

Test Readiness Before You Test the Business Case

A technically excellent proposal built for an organisation that is not ready to use it is not a good investment; it is a good idea arriving too early.

Readiness has three components worth asking about explicitly and separately, echoing the capability discussion in Building an AI Strategy That Delivers Value: is the underlying data available, accessible and of sufficient quality; is the process this will touch well enough understood and consistent enough for the system to learn a stable pattern from it; and are the people who will actually use the output, not just the sponsor pitching the investment, bought in, trained, and organisationally positioned to change how they work as a result. A proposal that has not honestly assessed all three is asking for capital before it has established that the capital can actually be put to work.

This is where investment committees can add real value beyond simple financial scrutiny: pressing specifically on the data, process and people dimensions of readiness, rather than accepting a general assurance that "the team is ready." Vague reassurance on this point is one of the more reliable predictors of an investment that underperforms its projected return, not because the technology failed, but because the organisation was not actually positioned to use it.

Five Questions Before You Commit Capital

01 Value

Which metric moves, by how much, and by when?

02 Readiness

Is our data, process and people position honestly assessed?

03 Dependencies

What else has to be true elsewhere for this to work?

04 Implementation Risk

What is the realistic cost and timeline, including change management?

05 Measurement

How, specifically, will we know if this worked?

The Hidden Costs That Rarely Make It Into the First Business Case

The visible cost of an AI investment, licensing, infrastructure, initial build, is usually only part of the real cost. The rest is easy to underestimate and expensive to discover late.

Change management is the most commonly underestimated line item: the training, workflow redesign and internal communication required to get a system genuinely adopted, rather than technically deployed and quietly ignored, is a well-documented and substantial share of total implementation effort in most serious technology change programmes, and AI initiatives are no exception. Ongoing monitoring and governance is the second: a system that requires no attention after go-live does not exist for any consequential use case, and the resourcing for oversight, discussed fully in What Leaders Need to Know About AI Governance, should be part of the original business case rather than a surprise budget request eighteen months later.

The third hidden cost is dependency risk: an AI initiative that assumes a separate data migration project, a different system integration, or another team's roadmap will land on schedule is exposed to that dependency's own risk of delay, and a business case that does not surface these dependencies explicitly is quietly betting the investment's timeline on someone else's project plan.

Where AI Investment Decisions Most Often Go Wrong

Technical detail substituting for business value. An architecture diagram and a vendor comparison can make a proposal feel rigorous while quietly avoiding the question of what business outcome actually changes.

Readiness assumed rather than assessed. A general assurance that "the team is ready" is not the same as an honest, specific evaluation of data, process and people readiness.

Change management and governance left out of the budget. These costs do not disappear by being omitted from the original business case, they simply arrive later, as an unplanned request.

No defined measurement plan before launch. Without a metric and a baseline set in advance, "success" becomes whatever can be argued for after the fact.

Key Takeaways

Press for a named metric, an honest readiness assessment, and a measurement plan, before you press for a return projection.

Better AI investment decisions come from asking the questions technical proposals are often designed to avoid: what business outcome actually changes, whether the organisation is genuinely ready to use what it is buying, what hidden costs and dependencies sit outside the visible budget line, and how success will be measured once the system is live. A proposal that cannot answer these clearly is not yet ready for capital.

Continue this series with Scaling AI Across the Organisation, or explore how INTAI helps investment committees evaluate AI business cases directly through our Executive AI Briefings and Board AI Briefings.

Talk To Our Leadership Advisory Team

Would your AI business cases survive scrutiny?

We pressure-test AI business cases, separating what's ready from what isn't.

Book a Consultation

Scroll