Most enterprises do not have an AI idea problem.
They have an AI investment problem.
Customer service copilots. Intelligent search. Forecasting. Document automation. Predictive maintenance. Decision support. Agents. Every business function can now produce a credible list of opportunities.
Many of them will sound valuable.
But there is never unlimited budget, delivery capacity or executive attention.
That changes the question.
It is no longer simply:
What can we do with AI?
It is:
Which initiatives deserve investment now — and what evidence makes that decision defensible?
That is a much harder question.
And increasingly, it is the question that separates AI activity from AI value.
More AI activity does not automatically create more AI value
The first phase of enterprise AI was largely about discovery.
Find use cases. Run pilots. Give teams access to new technology. Build experience.
That made sense.
But the measures that indicate momentum in an experimentation phase — number of pilots, tools deployed, users activated, ideas collected — become less useful as AI moves into enterprise investment.
A portfolio with 40 initiatives is not necessarily stronger than one with 15.
Ten successful pilots do not tell you whether those pilots deserve to scale.
And high adoption of an AI tool does not, by itself, tell a CFO whether capital should continue flowing towards it.
Recent enterprise research reflects this tension. Organisations continue increasing AI investment while leadership remains under pressure to demonstrate tangible financial and organisational outcomes. Gartner also finds that activity-based metrics such as projects, pilots, deployed tools and usage are increasingly disconnected from what CEOs and boards ultimately expect from AI.
The management challenge is therefore changing.
AI leaders need to move from stimulating activity to steering investment.
A roadmap is not an investment portfolio
Most organisations already have some form of AI roadmap.
That is useful.
But a roadmap typically tells you what the organisation intends to explore, build or deploy.
An investment portfolio must answer harder questions:
Why this initiative?
Why now?
What is it worth?
What needs to be true for that value to materialise?
Can the organisation actually execute it?
What are we not funding because we chose this?
This distinction matters because prioritisation without trade-offs is not really prioritisation.
If every initiative is described as strategic, innovative or high priority, leadership still has no mechanism for allocating scarce resources.
A mature portfolio therefore needs decisions, not just rankings.
Fund. Validate. Prepare. Defer. Stop.
Those decisions force a much more useful conversation than moving an initiative from a score of 74 to 81.
A business case is only as strong as the assumptions behind it
Financial estimates can create a false sense of precision.
An initiative may show €2 million in projected benefit. Another may show €1.4 million.
The temptation is obvious: fund the larger number.
But those figures are meaningless unless leadership can understand what sits underneath them.
What baseline was used?
How much process volume is actually affected?
What level of adoption is assumed?
Does the benefit depend on headcount reduction, productivity release, revenue growth or risk avoidance?
Are those benefits cashable?
What implementation, integration, model, data and change costs have been included?
What evidence comes from the organisation itself, and what comes from an external benchmark?
Two initiatives with two ROI percentages are not comparable if the logic underneath those percentages is fundamentally different.
A credible AI investment process therefore makes assumptions visible before it makes numbers impressive.
This is also where finance needs to be involved earlier. Gartner’s work on AI funding argues for common investment standards across AI spend and for business cases to explicitly validate their cost components, expected value and benefit-realisation criteria.
The objective is not to eliminate uncertainty.
AI investments will contain uncertainty.
The objective is to make that uncertainty visible enough to manage.
High potential is not the same as readiness
There is another reason the biggest business case should not automatically win.
An initiative can have significant economic potential and still be a poor investment today.
Perhaps the data foundation is insufficient.
Perhaps the required integrations do not exist.
Perhaps critical skills are missing.
Perhaps operating processes are not ready.
Perhaps ownership is unclear.
Perhaps governance requirements would delay deployment materially.
This is why value and readiness should not be evaluated separately.
But there is an important distinction.
The useful question is not:
“How mature are we at AI?”
It is:
“How ready are we for the AI portfolio we are actually trying to execute?”
Those are very different questions.
An organisation does not need to become world-class across every AI capability before it can create value.
It needs sufficient readiness in the capabilities that its chosen portfolio requires.
That makes readiness portfolio-specific.
A customer-service portfolio may place heavy requirements on knowledge management, integration, data quality and operational adoption.
A predictive-maintenance portfolio may require a completely different combination of industrial data, engineering capabilities and workflow integration.
A high-level maturity score hides those differences.
What leaders need to see is:
Current readiness → Required readiness → Gap
And then ask whether that gap should change the investment decision.
Every AI portfolio is also a capability and workforce portfolio
AI investments do not stop at technology.
They change work.
That is one reason productivity estimates are often more complicated than they first appear.
Saving thirty minutes of a task does not automatically create thirty minutes of economic value.
What happens to the released capacity?
Does the role change?
Does the work move to another function?
Does remain necessary?
Which skills become less important, and which become more important?
Which organisational capabilities need to strengthen before the AI solution can scale?
Research on enterprise AI value increasingly connects value realisation with changes in workforce productivity, roles, responsibilities and organisational structure.
This is why capability and workforce impact should be considered before scaling decisions, not after deployment.
If five initiatives all depend on the same scarce capability, that is a portfolio issue.
If ten initiatives collectively remove significant effort from one function, that is a workforce issue.
If several initiatives require the same data, platform or governance capability, that may be a reason to fund the shared capability first.
Viewed this way, the portfolio becomes more than a list of applications.
It becomes a picture of where the organisation itself needs to change to realise the expected value.
Governance should focus attention, not slow everything down
Governance is often treated as something that happens after an initiative has already been selected.
That is too late.
Risk, compliance, privacy, security, human oversight and accountability can materially change both feasibility and economics.
But the answer is not to subject every experiment to the maximum possible governance process.
Good AI governance should help the organisation distinguish between levels of criticality and apply the appropriate oversight.
A low-risk internal productivity experiment should not necessarily move through the same process as an AI system influencing employment, credit, safety or other consequential decisions.
The principle is simple:
govern what matters most, at the level it requires.
Governance then becomes part of investment discipline rather than a separate compliance exercise.
For each initiative, leadership can ask:
- What could go wrong?
- What controls are required?
- Who is accountable?
- What evidence is required before the initiative moves to the next stage?
- Does the governance burden materially change the business case?
This also makes governance an enabler of scale.
It is much easier to accelerate an initiative when everyone already knows the conditions under which it is allowed to progress.
Funding should follow evidence
Perhaps the biggest mistake in portfolio management is treating the original approval as the end of the investment decision.
It should be the beginning.
AI initiatives change as they move from exploration to pilot, deployment and scale.
Evidence improves.
Costs become clearer.
Adoption becomes observable.
Risks emerge.
Assumptions prove right or wrong.
Funding should change with that evidence.
Gartner’s research on AI funding makes a similar case for explicit decision gates and criteria to scale, modify or terminate initiatives rather than letting funding continue by default.
That leads to a more useful operating rhythm:
Fund when the value case is credible and the organisation is ready enough to execute.
Validate when the upside is attractive but critical assumptions still need evidence.
Prepare when value is clear but important must first be addressed.
Defer when the initiative may make sense later, but timing or dependencies work against it today.
Stop when new evidence no longer supports continued investment.
Stopping is not failure.
Continuing to fund an initiative after the evidence has changed is.
The goal is not a perfect score. It is a better decision.
Enterprise AI will remain uncertain.
No scoring model will perfectly predict which initiative will generate the greatest return.
Nor should it try to.
The purpose of portfolio discipline is not to replace executive judgement with an algorithm.
It is to give that judgement a better foundation.
A strong AI investment process makes the important things visible:
Value.
Feasibility.
Readiness.
Assumptions.
Capability and workforce impact.
Governance and risk.
Evidence over time.
Once those dimensions can be compared consistently, leaders can have a much better conversation about where budget and capacity should go.
That is the shift enterprises now need to make.
From:
“How many AI initiatives do we have?”
to:
“Which AI investments can we defend — and what should we stop funding to make room for them?”
A more disciplined approach to Enterprise AI Value Realisation
At MetroMinds, this is how we think about Enterprise AI Value Realisation.
AI initiatives should not live as disconnected experiments, business cases or risk assessments. They need to be evaluated as part of one portfolio, where expected value can be compared with feasibility, organisational readiness, capability impact and governance requirements.
Because the hardest AI decision is rarely finding one more initiative to start.
It is deciding, with evidence, which ones deserve to continue.
How ready is your organisation to turn AI ambition into measurable value?
Assess your AI value readiness → www.metrominds.io/valuerealisation
