AI Strategy: decide where AI belongs in the business
A strategy is the set of choices leadership will fund, sequence and refuse, with an owner and a reason beside each. We help you reach those choices before the money is committed.
In brief
An AI strategy is a set of decisions: where AI changes the business, what to fund first, what to leave alone and who answers for each choice. We reach them with the executives who own the affected work, test what each choice needs from data and systems, and record the reasoning for the board.
On this page
What does a strategy have to decide that a list of use cases does not?
A list says what AI could do for the business. A strategy commits money, and that forces decisions a list lets everyone avoid: where AI will change how the business earns or spends, the order in which money moves, the good ideas that will wait, and the person who answers for each choice.
The refusals carry more weight than the endorsements. The strategy earns its keep in the meeting where a proposal that looks ready to fund is told to wait because it needs a data source nobody owns, and the sponsor accepts the reason. We write those reasons down so they survive the meeting.
A strategy holds only if the CEO or chief AI officer, the CIO who inherits the systems, the finance partner and the head of each affected function can each repeat it in a sentence. If one cannot, it has not been decided yet.
A strategy that cannot tell a funded idea to wait is a wish list with a budget line.
From a long list to a funded sequence, in the order the work happens
- 01
Interview the people who do the work before the people who sponsor the ideas
We start with the heads of the functions where the work happens, and with finance and risk, before any candidate is pitched. We bring away a map of where decisions are slow, information is keyed in twice and a mistake is expensive. Each idea is then judged against that map instead of its own slides.
- 02
Restate every candidate as a decision, not a project
A candidate is rewritten in one form: the task or decision it changes, who performs it today, what better looks like, and what a wrong answer costs the business. One that cannot be written this way goes back to its sponsor.
- 03
Test what each candidate needs before it can start
The data lead, the security lead and the owner of the target process each say what exists today: whether the data can be reached, whether the systems accept the change, whether anyone can approve use. A candidate with a missing prerequisite is placed behind the work that supplies it, and that work gets an owner.
- 04
Agree the ranking rule first, then rank in one room
Leadership settles the criteria before it sees any scores, so the order cannot be argued backwards from a favorite. We run the session and the executives decide. What leaves the room is a funded sequence, with a stage gate and a stop condition on every commitment.
- 05
Write the not-now list and the rule for looking again
Deferred and refused ideas are recorded with the reason and the condition that would bring each back. The sponsor of a deferred idea reads that entry before the board does.
How does each candidate appear on the board paper?
Every candidate leaves the work with a verdict, and the paper shows the verdict and its conditions instead of the pitch.
| Verdict | What has to be true | What the paper must state |
|---|---|---|
| Fund now | The data is reachable, an owner is named and the process owner agrees to change how the work is done | The stage gates, the stop condition and the person with authority to stop it |
| Fund after a prerequisite | The data, access or approval it needs is being supplied by a piece of work that is itself funded | The dependency, who owns it and when leadership will check it |
| Test in a bounded experiment | The question is cheap to answer and the answer would change the ranking | What the experiment must show, and who decides afterward |
| Wait | The value is real, but another candidate needs the same scarce people or data first | The reason, and the condition that brings it back |
| Do not do | An error would fall on customers or patients in a way no control can bound | The reasoning, kept so the idea is not proposed again unchanged |
A pattern we are called into
A retailer with an assistant, a forecast and a content tool, all approved separately
A multichannel retailer has an assistant live in customer service, a forecasting trial in merchandising and a content generator in the e-commerce team. Each was approved on its own merits, in a different quarter.
The budget cycle is close, the board wants a position, and the chief AI officer has been told every one of them should scale.
- Map the work. We sit with the heads of service, merchandising, e-commerce and finance to see where time and error accumulate.
- Test the prerequisites. The forecast relies on sales history that stores and the web site keep differently, and nobody owns the reconciliation. The assistant answers from a policy library that legal has never reviewed for this use.
- Rank against rules agreed first. Leadership sets the criteria: what an error costs a customer, how ready the data is, how much the work must change. The assistant goes ahead once its library has an owner. The forecast waits behind the reconciliation, funded first. Content generation continues as a bounded experiment with a human editor.
- Record it once. The not-now list, the owners and the review dates go to the board in a single paper.
Where it endsFunding follows a sequence instead of separate requests. The reconciliation, the least glamorous piece, has an owner and a budget line for the first time.
Is a strategy engagement what you need now?
Strategy work is sometimes the wrong first step.
It fits when
- More than one function is asking for AI funding and no rule decides between them
- The board has asked for a position and management has a list
- Pilots are competing for the same data, people or reviewers
- A change of leadership or an acquisition has reset what the business will fund
Look elsewhere when
- One use case is already chosen and the need is to build it, which is delivery work such as AI Consulting
- No executive can release money, so a ranking would have nobody to act on it
- The question is which vendor to buy from, which is an evaluation with its own written scope
- The aim is a document for investors that will not be allowed to change a decision
What you give up is speed on the first commitment. Nothing is funded until leadership has agreed how it will choose. For a business already running isolated experiments, that pause is cheap, because it comes before money goes on prerequisites nobody planned.
What does the board hold at the end of the work?
- A decision memo: each choice, the reasoning, and the alternatives that were considered and lost
- The ranked portfolio: each candidate as a decision, with prerequisites, owner and stage gates
- A readiness finding on data, systems, skills and approvals, written for the people who must fix it
- The ranking criteria and how they were applied, so a later candidate can be scored without us
- A not-now list with a reason and a return condition against every item
- A short board paper, and a calendar of review points each with the evidence it needs
What do we need from you, and where does our part stop?
We need time from the executive who owns the decision, the heads of the functions affected, finance and whoever holds the data. Working material matters more than presentations: process documents, system inventories, recent incident reports and existing AI vendor contracts, all handled under the NDA that covers every engagement.
Decisions stay with your leaders. We advise, run the sessions in which choices are made, and write the record. We do not run delivery, and we recommend vendors only where an evaluation is in the written scope. We do not promise financial returns from any AI investment. We state the conditions under which each choice would pay off, so the first review can test them.
Where can you test your own list before we start?
The prioritization matrix and the investment decision brief are the working tools behind the ranking and the board paper, and both can be completed on your own list first. The research on stalled pilots describes the prerequisites that go unplanned.
Questions leaders ask before commissioning a strategy
How is an AI strategy different from a list of use cases?
A list records what could be done. A strategy commits: it says what is funded first, what waits, what is refused, who answers for each choice and which prerequisites come first. If striking out any one idea would change nothing about what anyone does on Monday, the document is still a list.
Do we need a strategy before running any pilots?
Not for a bounded experiment, which can teach you something cheaply. You need the strategy once pilots begin to compete for the same data, people and reviewers, because that is when sequence and ownership start to matter.
Who has to be in the room from our side?
The executive who owns the decision, the heads of the functions where work would change, a finance partner and whoever holds the data. Small enough to decide, wide enough that no owner meets a choice for the first time after it is made.
What happens after the strategy is delivered?
Your leaders own it, and the review points keep it current. If you want help building what the strategy chose, that is a separate written scope, for example AI Consulting, so the advice and the work stay distinct and the strategy can still be questioned.
Related
Does your AI strategy make choices?
Describe what you are weighing in a few lines. A senior practitioner replies with a suggested first conversation.