Strategy, not pilots.
The cheapest way to look like you have an AI strategy is to fund a portfolio of pilots and rebrand it. The signal that you actually have one is a written no list - the use cases, tools, and teams you have decided not to bet on this year. McKinsey's annual State of AI survey finds that the small group of companies capturing meaningful EBIT impact from AI almost all share a defined portfolio with explicit no decisions, not just an open list of pilots.
Without that, every team runs every experiment, no investment compounds, and the third quarterly review feels exactly like the first.
The eight steps.
Step 01
Name the business outcomes.
Three to five outcomes AI is meant to move in the next 12 months. Stated in business language: revenue, retention, cost, time-to-market. If you cannot say it without the word 'AI', it is not an outcome yet.
Step 02
Pick the bets.
Decide which use cases, teams, and tools you are betting on - and which you are explicitly not. Write the no list. Without it, every team runs every pilot.
Step 03
Connect to customer value.
At least one bet must show up for the customer. Internal-only AI strategies underperform because no one outside the team feels the change.
Step 04
Map the people work.
Per role: what changes, what they need to learn, how their work is measured differently, and who their model user is. Generic AI training is the cheapest part; this is the expensive part.
Step 05
Redesign the workflow.
Pick the workflows touched by each bet and rebuild them with AI in the loop. Layering AI on an unchanged workflow returns time-saved that disappears into more of the same output.
Step 06
Choose the platform.
Now - not first - decide the tool stack. Build vs buy, sanctioned vs tolerated, who owns each. Most strategies that lead with this step never reach the workflow step.
Step 07
Set the measurement.
One adoption metric, one capability metric, one outcome metric per bet. Quarterly. Public to the team.
Step 08
Schedule the re-read.
AI moves quarterly. A strategy with no re-read date will be a year out of date in nine months.
Common failure modes.
- - Tool-first. Strategy starts with the platform choice. People and process get a paragraph at the end.
- - All-internal. Every metric is efficiency. Customers feel no difference. Board interest fades by Q3.
- - Leadership exempted. The CEO and CMO talk about AI without using it. Adoption stalls at the manager layer. Harvard Business Review's work on managing generative AI repeatedly returns to leader use as the single best predictor of whether direct reports take the tools seriously.
- - No no list. Every team funded. Nothing depth-cut. Nothing compounds.
- - One-and-done. Strategy shipped, never re-read. Six months later the model landscape has moved and the plan has not.
Where to start.
Before you write the strategy, get a read on where you are. The AI readiness assessment guide covers the six dimensions to score, and the free AI Alignment Snapshot is the self-serve version of that read. The longer argument behind it sits in our book, The Elephant in the Algorithm.
Questions people ask.
What is an AI adoption strategy?
A written plan for how an organization will get value from AI through the people who use it. It covers which AI bets the organization is making, which it is not, how those bets connect to business strategy and customer value, and how the workforce will be brought along. It is not a list of pilots.
Is AI adoption the same as AI implementation?
No. Implementation is getting the tool stood up. Adoption is people using it well enough, often enough, on the right work, to move an outcome. Most failed AI programs implemented cleanly and then stalled at adoption.
How long should an AI adoption strategy take to build?
Four to eight weeks for a mid-market organization, depending on how much groundwork already exists. Faster than that and you skipped the listening. Slower and the strategy is out of date before it ships.
What is the most common mistake?
Treating it as a technology strategy. The hard parts of AI adoption are about how leaders model use, how workflows change, and how value is measured - not which model to license. A good strategy spends most of its pages on people and process.