You walk out of the meeting holding a vague brief and a budget. The board has decided the business needs to “get AI in”, and they have handed it to you, because you are the person who makes things happen, not because you know anything about AI. There is already an update in the diary. The instinct, sitting there with the clock running, is to buy something quickly so there is progress to point at.
That instinct is the trap. The operators who come out of this well do almost the opposite of what the pressure tells them to do. This post is the first of six on the AI mandate, and it is about the one decision that shapes everything after it. Do you start with a tool, or do you start with the problem.
What is the choice you are actually facing?
The real choice is not which tool to buy but whether to lead with the tool or with the problem, and the two paths diverge in week one. Tool-first means picking a product, deploying it, and showing the board movement. Problem-first means reading the real state of the business and the problems worth solving before any purchase. The pressure pushes you towards the first. The evidence sits with the second.
The difficulty is that you have been set up in an awkward position. You are now accountable for AI outcomes, but you have not been handed the governance, the budget control, or the decision rights that would let you guarantee them. Korn Ferry calls this the AI readiness paradox, organisations giving AI leadership to strong operators who lack the AI-specific competencies the task needs. High expectation, low preparation. That gap is not your fault, and naming it to yourself early matters, because it changes how you weigh the next move.
Founders often delegate AI precisely because it feels technical, and because keeping a strategic distance from a technology that affects the firm’s valuation suits them. Spencer Stuart describes the pattern. The delegate absorbs the board pressure while the founder keeps the resources and the real decisions. You cannot fix that on day one. You can refuse to make it worse by spending the budget before you understand what you are spending it on.
When is starting with the tool the right move?
Almost never at this stage, though there is a narrow case where it works. A tool-first move is defensible when the problem is already crystal clear, the workflow is already understood, and the only open question is which product solves a known job. If your business has a documented bottleneck that a specific AI tool plainly addresses, and you can name the outcome before you buy, then moving fast is fine.
That is rarely the situation a freshly handed delegate is in. What looks like a clear problem in the boardroom is usually a slogan, “get AI in”, not a defined job to be done. Buying against a slogan produces what Addepar’s team call AI theatre, flashy demos with no measurable outcome. The tool gets installed, a few people try it, and ninety days later there is nothing in the numbers to show for it.
MIT’s research found that roughly ninety-five per cent of generative AI pilots fail to show a measurable P&L impact. Read honestly, that figure says less about whether AI works and more about how pilots get launched. Many are funded without a real problem behind them, and the person who launched them carries the blame when they stall. Speed feels safe in week one. It stops feeling safe at the board update where you have movement to report and no result.
When is starting with the problem the right move?
In almost every case where you have been handed a broad mandate rather than a defined job. Starting with the problem means resisting deployment and asking what the business is actually trying to solve, then reading the current state before you commit a penny. It looks slower and produces something the board can trust far faster. It is the move that separates the initiatives that survive from the ones that disappear.
The test Addepar’s team offer is the one to carry into every conversation. Would this initiative still matter to the business if it did not involve AI? If yes, you have found a real problem and AI might be the right way to solve it. If no, you are about to fund a demo. That question filters the noise out of an AI mandate before any money moves.
There is a second reason problem-first wins that the boardroom never mentions. Your real current state is far wider than the sanctioned projects. Over ninety per cent of employees already use personal AI tools for work, so the actual AI footprint inside your business is mostly invisible until you go looking. Any strategy written as if AI use starts the day you arrive is built on a false baseline. Surfacing that shadow activity first tells you what is already working, where the data risk sits, and where adoption will be easy because people are already halfway there.
What does it cost to get this wrong?
The cost lands on you personally, which the board rarely says out loud. When a tool-first pilot stalls, the delegate becomes the natural scapegoat. Employee distrust hardens into leadership scepticism, the founder re-engages with unrealistic demands, and your credibility takes the hit for a structure you did not design. A meaningful share of executives already report fearing AI-related job loss within a year, and that fear pushes good operators into fast, hollow movement.
The internal damage is just as real. Staff fears about AI, job loss, inaccuracy, opacity, data insecurity, do not announce themselves. They turn into passive resistance, workarounds, and people feeding bad inputs to prove the tool is worse than they are. HR Executive’s research is blunt about this. Roll a tool out without addressing the fear and you get adoption on paper and refusal in practice. A tool-first launch skips exactly the trust-building that adoption depends on.
There is also the wasted-funding cost. MIT’s data shows the heavily funded sales and marketing pilots tend to return the least, while less glamorous back-office automation returns the most. Spend your first budget on the flashy front-office demo because it is visible, and you may have backed the lowest-return option in the business while burning the goodwill you will need later. Getting the first move wrong is not a neutral delay. It spends money, trust, and your own standing at the same time.
What should you ask before you decide anything?
Five questions, and none of them is “which tool”. What problem is the business actually trying to solve, with a number attached if one exists. Would that problem still matter without AI. What AI are your people already using off the books. Is your data in a state any tool could work with. And what will the board count as success in twelve months, because the answer is rarely “we bought something quickly”.
Those questions feed one deliverable, an honest single-page read of the current state across five dimensions, the shadow AI already in use, data readiness, technology infrastructure, organisational capability, and risk exposure. A shortlist of tools is not the job here. One page, because the board will read one page and ignore a deck. Honest, because a flattering current-state read sets you up to be caught out later. This is the document that earns you the room to run the next ninety days properly.
The reframe that helps here is in the language you use with yourself. In week one you are “building AI readiness” rather than “implementing AI”, and “identifying opportunities” rather than “deploying solutions”. Read that as an accurate description of the only work that makes the later work succeed, and it takes the pressure off the false deadline in your head. The next post in this series turns that one-page read into a structured first thirty days. The work starts with the honest page, not the purchase order.
If you are the operator holding that brief and would rather not work out the first ninety days alone, that is the kind of problem worth a conversation. Book a conversation and we will look at where your business actually stands before anyone talks about tools.
