The category has a credibility problem, and it is worth naming honestly before anything else: a great deal of what is sold as AI strategy is a slide deck describing what AI is, a list of use cases anyone could have written, and a roadmap with no owner. It costs a lot and changes nothing.
That is not an argument against the work. It is an argument for knowing what the work should produce.
What a strategy engagement should actually deliver
Not a technology assessment. Not a survey of what competitors are doing. Those are inputs, and they are the cheap part.
The output that matters is a short list of specific decisions, each with a named owner, a cost, and a way to tell whether it worked. Something closer to:
"Automate tier-one support triage. Owner: Head of Support. Build cost roughly ÂŁX, run cost ÂŁY monthly. Success is deflecting 30% of tickets without CSAT dropping. Reviewed in 90 days. If it fails, we stop."
If what you receive cannot be converted into sentences of that shape, you did not buy a strategy. You bought a summary of a field you could have read about.
The questions that actually matter
Most engagements should spend their time on four questions, and most spend it on the first alone.
Where would this create value? The easy one. Every business has candidate processes, and identifying them rarely needs outside help.
Which of those can we actually staff and run? Much harder and much more often skipped. A use case nobody owns after the consultants leave is not a use case. Most AI initiatives fail here rather than technically.
What does the data situation permit? This is where roadmaps quietly die. If the data is scattered, undocumented, or legally constrained, half the plan is fiction. An honest engagement finds this in the first fortnight and rewrites the plan around it rather than presenting the plan anyway.
What are we not doing? A strategy that endorses eleven initiatives has made no choices. The value of the exercise is largely in what it rules out, and a deliverable with no rejected options should be treated with suspicion.
Where consultants genuinely help
Seeing failure modes early. Someone who has watched a dozen of these knows which pilots stall, which vendor claims do not survive contact with production, and which "quick wins" have long tails. That pattern recognition is real and hard to acquire from inside one company.
Saying the uncomfortable thing. Sometimes the honest finding is that the data is not ready, or the process should be fixed before it is automated, or a proposed project exists because an executive wants it rather than because it pays. An external voice can say that survivably.
Concentrating a decision. Organisations circle these questions for a year. A structured six-week engagement can force resolution, and the forcing is worth something independent of the content.
Where they do not
Building it. Strategy and delivery are different skills. Firms that sell you a strategy recommending a large implementation they happen to sell have an obvious conflict, and it is reasonable to ask about it directly.
Knowing your business. They will know AI. You know your operations, your customers and your constraints. If the engagement does not extract that from your people, the output will be generic â and generic is exactly the failure mode the category is known for.
Making it stick. Adoption is an internal problem. No deliverable substitutes for someone senior owning the change after the invoice is paid.
Practical guidance on buying it
Scope tightly. Prefer six focused weeks over six exploratory months. Long engagements drift toward describing the landscape.
Ask for a redacted deliverable up front. Not a case study â the actual artefact from prior work. If it is a general overview of AI with a use-case matrix, you now know what you would receive.
Insist your people are in the room. If the work happens through interviews and a final presentation, the knowledge leaves with the consultants.
Agree what "no" looks like. An engagement that cannot conclude "do not do this yet" is not evaluating anything.
Pay for judgement, not headcount. The value is in a few experienced people making calls, not a large team producing volume.
When you should not buy it
If you have not tried anything yet, a strategy engagement is premature. One small project delivered end to end will teach you more about your real constraints â data, staffing, procurement, adoption â than any assessment. Strategy is most valuable when you have enough experience to have specific questions.
If you already know what to build and are looking for validation, that is a more expensive way to get permission than simply deciding.
If nobody senior will own the outcome, do not start. The most common failure is not a bad strategy. It is a good one that nobody was accountable for.
The test
A useful engagement ends with your team able to explain, without the deck, what you are doing, what you decided against, who owns each piece, and how you will know in ninety days whether it worked.
If that is not the state you are in when it finishes, the work did not land â regardless of how good the deck was.
Originally published at https://www.misar.blog/@misar/articles/ai-strategy-consulting-what-it-should-deliver-and-when-to-skip-it
