Many AI investments that disappoint do so for reasons that have little to do with the technology. They go wrong much earlier.
AI was aimed at the wrong work. The expected value was never properly tested. Something the organisation already owned could have done the job. Or nobody decided what success would actually look like before the investment was approved.
By the time those problems become visible, the contract has been signed, the implementation has started and people are being asked to adopt the tool. At that point, the question has changed from should we do this? to how do we make this work?
That is an expensive change of question.
AI has made buying easier than deciding
There is no shortage of things organisations can buy. AI is being added to Microsoft 365, ServiceNow, Salesforce, SAP and almost every other major enterprise platform, and specialist vendors are appearing for individual functions, workflows and industries. Organisations can also build things themselves.
The result is not a shortage of technology but an abundance of options, and abundance creates a different problem. For almost any workflow, there may now be four reasonable choices:
- Do nothing.
- Use AI already included in something you own.
- Build or configure something yourself.
- Buy another product.
Those choices have very different costs, risks and operating consequences. Yet organisations often start with the product rather than the workflow. Someone sees a demonstration, a team wants a licence, a vendor presents a business case or a competitor announces an AI initiative, and the conversation begins with should we buy this?
It should begin somewhere else.
Start with the work
In an earlier article, I used Ronald Coase’s work on transaction costs to argue that AI investment struggles to deliver when ownership is unclear and the information behind decisions is scattered. Clear ownership and visible reasoning make better decisions possible. They don’t tell you which work to change.
That starts with the work itself.
Before evaluating an AI product, describe the work it is supposed to improve, not the technology or the vendor.
Take something ordinary: processing incoming customer requests.
Before discussing AI, you should know roughly:
- How many requests arrive?
- How much employee time does the process consume?
- Where are the delays?
- How much rework occurs?
- What systems and data does the process depend on?
- Who owns the outcome today?
Now there is something against which an investment can be tested.
Suppose the process consumes 6,000 employee hours a year.
An AI proposal claims it can reduce the effort by 40%.
That sounds attractive.
But 40% of what?
Perhaps only half of the process can safely be supported by AI. Perhaps employees still need to review every output. Perhaps integration work removes part of the saving. Perhaps the organisation already pays for a platform capable of doing most of it.
The useful question is not:
What can this AI do?
It is:
What changes in this workflow if we introduce it?
That is a much harder question for a vendor to answer, and it is the one leadership needs.
The business case should survive without the vendor
There is something slightly strange about the way technology investments are often assessed.
The company selling the product is also asked to explain the value of buying it.
That does not make the vendor dishonest. Vendors know their products better than anyone else.
The vendor’s job is to show what is possible.
Your job is to decide what is probable in your organisation.
Those are different things.
If a business case depends entirely on vendor assumptions about productivity, adoption or savings, it is not really your business case yet.
The organisation should be able to state independently:
- This is what the workflow costs us today.
- This is the part we believe AI could change.
- These are the assumptions behind that belief.
- These are the alternatives we considered.
- This is how we will know whether we were right.
Only then does the vendor proposal become evidence rather than the starting point.
The cheapest AI may be the one you already own
There is another reason to start with the workflow.
It changes the buying decision.
Imagine a team identifies a good opportunity for AI. The next question should not automatically be which vendor do we buy from? It should be:
What is the most sensible route to this outcome?
The capability may already exist in a platform you pay for. A process change may remove most of the problem without AI. A small internal solution may be enough. Or a specialist product may genuinely add enough value to justify another contract, integration, security review and vendor relationship.
Buying something new should be one possible answer, not the default.
This matters even more as AI capabilities become embedded across the existing enterprise stack. Organisations risk accumulating overlapping AI products while using only a fraction of the capability they already license.
The cost is not just licence spend.
Every additional tool creates another thing to integrate, secure, govern, support, train people on and eventually replace.
Sometimes the answer should be no
This may be the most overlooked part of AI governance.
A good AI decision process must be able to produce a no, not because AI is too risky, but because the investment is not good enough.
A workflow may be too small to justify the effort, the underlying process may need fixing first, the expected benefit may depend on unrealistic adoption, or the capability may already exist somewhere else. The organisation may also simply have better AI opportunities competing for the same money and attention.
AI investments should therefore be compared, not only assessed one by one.
If ten opportunities exist and the organisation has enough capacity to implement three properly, the important question is not whether all ten have positive business cases.
It is:
Which three should go first?
That is an investment allocation problem.
Governance should begin before approval
Most governance becomes visible after an organisation decides to proceed.
Security reviews the tool. Legal reviews the terms. Privacy assesses the data. IT determines how it will connect to the existing environment.
All of that matters.
But there is another governance question that comes first:
Why are we making this investment at all?
There is a governance reason for keeping that record too. Existing privacy, consumer protection and sector rules are increasingly being applied to AI-enabled decisions, making clear accountability and documented evidence more important. But even without the regulatory pressure, the management case is straightforward: an organisation should be able to explain why an investment was made, what it expected to happen and who owned the decision.
Source: Gartner, AI Liability: How CIOs Can Reduce Legal Risk.
That decision should have an owner.
And the organisation should keep a record of:
- the workflow being changed,
- its current baseline,
- the expected value,
- the assumptions behind that value,
- the alternatives considered,
- the chosen route,
- the person accountable for the decision,
- and when the decision will be reviewed.
This does not require a 40-page business case.
It requires enough evidence that, six or twelve months later, somebody can compare what actually happened with what the organisation thought would happen.
That is how AI investment decisions get better.
Not by making every prediction correct, but by making the assumptions visible enough to learn from them.
Where the decision fits
AI gateways and platform controls can govern which models people use and what data they can reach. They cannot decide what AI should be used for. That decision comes first, in a sequence that is relatively simple:
Baseline → Decide → Implement → Govern → Measure → Reassess
And there is an important possibility inside that sequence:
Decide not to invest.
That is not a failed AI initiative.
AI can remove friction from work. But removing friction is only valuable when you have chosen the right work to change.
Sometimes the highest-return AI decision you can make is not to invest at all.
Operient helps leadership teams decide which AI investments to make, in what order and on what evidence.