What should an AI strategy consulting engagement actually produce?

Most organisations are not short of AI ideas. Ask each function for use cases and a substantial backlog will appear quickly: copilots, agents, forecasting, service automation, content generation, decision support and dozens of local productivity opportunities.

That activity can be useful. It is not yet a strategy.

A strategy must make choices under uncertainty. It should connect the organisation's priorities to a small number of opportunities, explain why those opportunities matter more than the alternatives and define how the decisive assumptions will be tested.

The quality of an AI strategy consulting engagement can therefore be judged by what leadership is able to decide at the end.

1. A material outcome and economic baseline

The work should begin with an outcome the business already cares about: growth, margin, working capital, capacity, service, quality or risk. “Adopt AI” is not an outcome.

The current position also needs to be visible. How is the outcome produced today? Where is value created, delayed or lost? Which costs are fixed, variable or avoidable? Which service or quality measures matter? Without a baseline, almost any improvement can be described as value and almost none can be verified.

A strategy does not need perfect data before it starts. It does need a clear account of what is known, what is estimated and what must be measured next.

2. A view of the work that determines the outcome

Organisation charts and system maps are useful, but they rarely show how value-producing work actually happens.

A good engagement looks at the decisions, handoffs, exceptions, queues, reviews and information flows that shape the outcome. It identifies where human judgement is valuable, where it is compensating for a broken process and where new capability could change the economics of the work.

This prevents the strategy from becoming a catalogue of tools applied to existing activity. It also reveals opportunities that would be missed if the scope were defined by current departmental boundaries.

3. A focused portfolio of value hypotheses

Each priority opportunity should have an explicit value hypothesis:

If we change this capability or workflow in this way, we expect this operational measure to move, creating this financial or strategic consequence.

That statement forces important distinctions. Time saved is not necessarily cost removed. Faster output is not necessarily more revenue. Higher model accuracy is not necessarily better customer or business performance.

The hypothesis should identify the assumptions carrying the case. Those assumptions become the focus of the next test.

4. A defensible order of priority

A useful prioritisation model goes beyond impact and feasibility. Become tests opportunities against five connected questions:

• Strategic fit: Does this move an outcome that matters?
• Value potential: Is there a credible and measurable path to value?
• Time to value: Can the decisive uncertainty be resolved soon enough to inform action?
• Workflow leverage: Does this change consequential work rather than add another tool?
• Delivery feasibility: Can it be tested safely with the available people, data and operating conditions?

The output should be a short portfolio with a clear reason for the order—not a two-by-two chart in which every idea remains “high potential”.

5. A path from strategy to evidence

The engagement should define what happens next for each priority opportunity. That means more than a delivery roadmap.

For the leading opportunity, leadership should know:

• what must be true for the value case to hold;
• what Minimum Viable Capability could test it;
• which users and real work need to be involved;
• what baseline and measures will be used;
• what control and risk conditions apply;
• what result would justify scaling, revising or stopping.

This is how strategy becomes an investment discipline rather than a statement of intent.

6. Ownership, funding and decision rights

AI initiatives often sit between the business, technology, data, risk and transformation teams. Ambiguous ownership is therefore a design problem, not an inconvenience to resolve later.

The strategy should identify a business owner for the outcome, the roles required to build and operate the capability, how decisions will be made and how funding will move as evidence changes.

A steering committee is not a substitute for accountability. Someone must own whether the value appears.

7. A clear list of what not to do

The most useful output may be the activity the business chooses not to fund.

A strategy should identify opportunities that are attractive but immaterial, premature, duplicative or too weakly connected to a business outcome. It should also make visible the platform, data or capability work that is genuinely shared—and distinguish it from infrastructure being built in anticipation of use cases that may never earn investment.

Red flags in an AI strategy engagement

Be cautious when the process:

• begins with a preferred technology rather than a business outcome;
• celebrates the number of use cases generated;
• ranks ideas without an economic baseline;
• treats readiness as a generic maturity score;
• ends with a roadmap but no Build and Prove design;
• assumes every pilot should progress;
• separates strategy from the people who understand the real work;
• leaves ownership of value unclear.

The standard to use

At the end of the engagement, leadership should be able to say:

These are the few AI opportunities that matter. This is the outcome and value logic behind each one. This is what we will prove first, how we will judge it and what we are choosing not to fund.

That is what AI strategy consulting should produce.