AI is already in most workplaces. Very few have decided what it's for. An AI strategy is six decisions, made in order, that settle it.
Key takeaways
- AI is everywhere, plans are rare. Most companies use AI somewhere, but formal adoption with a real plan is still unusual.
- Strategy is six decisions. The number to move, where AI goes first, your data, your rules, who owns it and how you measure it.
- Measure value, not activity. Research from MIT Sloan shows leaders often mistake quick productivity gains for lasting value.
McKinsey's 2025 State of AI survey found that 88% of organizations now use AI in at least one business function. On the surface, that sounds like the race is nearly over.
The Federal Reserve tells a different story. Its April 2026 research found that only about 18% of U.S. firms had formally adopted AI by the end of 2025. Yet roughly 78% of the U.S. workforce already works at companies using AI in some form.
So most AI at work is a chatbot one person tried, or a note-taking app someone switched on, with no plan behind it. That gap is good news for you. A short written plan puts you ahead of most of the market.
An AI strategy comes down to six decisions, and the first one carries the rest. Pick one business result you want to change this year, and write it as a number. I call this the one-number rule: every later decision has to point back to it, or it waits.
How to build an AI strategy in six decisions
Decision 1: Pick the number you want to move
Start with one specific, measurable problem. Maybe proposals take too long, support requests swamp your staff, or monthly reports eat two days of spreadsheet work.
Picture how it might go at a marketing agency. Proposals go out about a week after the discovery call. The owner writes down one goal: every proposal goes out within 48 hours. When someone pitches a new AI idea, the owner asks whether it moves the 48 hours. Most debates end there.
Why does a number work better than a goal like "faster proposals"? Because you can do the sum. Say the agency sends eight proposals a month, and each one takes a week. That's 56 days of prospects (potential clients) waiting every month. At 48 hours each, it's 16. Nobody argues with 40 fewer days of waiting. The same sum shows what a fix is worth paying for, and the guide to what AI implementation costs gives market ranges to set against it.
Sticking with it can be harder than picking it. Netflix is a useful reminder. In 2011 it split its DVD and streaming plans into separate subscriptions. It lost about 800,000 subscribers in one quarter, and its stock fell roughly 77%. It stuck with streaming anyway, and that choice built the company it became. Your stakes are smaller, but the pull to change direction every quarter feels the same.
So once you have your target, where should AI go first?
Decision 2: Start with the team closest to the number
Start with the team that owns that result. For the agency, that's whoever writes the proposals. Operations might start with data entry and status reports, and the guide to AI for operations and process improvement shows how to pick the first target.
Sales and marketing might start with follow-up emails, or with being found in AI search. McKinsey estimates that $750 billion in U.S. revenue will flow through AI search by 2028, so that money moves toward the businesses that show up in the answers. Customer service might start with faster first replies, which help you keep customers longer.
An approach I wouldn't take is starting in every team at once. Pick one, and let the others wait a quarter. One small win that everyone can see makes the next department far easier to bring along.
Decision 3: Check your data honestly
Find where the information behind your target lives. Then ask three questions. Is it complete? Is it consistent? Can other software reach it?
At the agency, past proposals might sit across three shared drives, each with its own naming habits. Staff find what they need from memory. AI can't.
So before anything drafts from those files, someone spends a week building one tidy folder of the best examples. That week can feel slow while you're in it. A week of tidying up front can save a month of poor results later. If your files already sit in one tidy place, I'd consider moving straight on to the rules.
Decision 4: Write your rules before you scale
Decide two things in writing: what gets a person's review before it reaches a client, and what data AI may use. For example, the agency's written rule might read: "Every AI-drafted proposal gets a senior team member's read before it's sent. Client budgets and contracts stay out of public chatbots." Two sentences is enough to start.
It's tempting to write these rules after something goes wrong. That's the expensive order. The guide to using AI at work without risking data leaks has a starting policy you can adapt.
Careful industries show why this decision comes early. Data from the Office of the National Coordinator for Health IT shows that 71% of U.S. hospitals used predictive AI tied into their core systems by 2024, up from 66% the year before. Hospitals aren't known for moving fast. If your clients have strict rules, that's a reason to write yours first. It's a poor reason to wait.
Decision 5: Keep the big decisions with the owner
BCG's 2026 research found that nearly three-quarters of CEOs are now their company's main decision-maker on AI strategy. That's double the share from a year before. In a small business, the decision-maker is you.
It makes sense. AI changes how work gets done, so it belongs with the person who decides how work gets done. Hand it to IT as a side project and it tends to stay a side project.
Then name one person who runs each AI-assisted task day to day. If it were my business, I might pick whoever already writes the most proposals, even if they aren't the most senior person. At the agency, the owner sets the 48-hour goal and signs off the rules. The operations lead runs proposal drafting and reports the result each month.
Which leaves the last question: how do you know any of it is working?
Decision 6: Measure the number, not the logins
Research from MIT Sloan's Center for Information Systems Research points to a common trap. Leaders often mistake productivity gains for real business value. A pilot looks promising, everyone declares success, and the work that drives margin and growth stays the same.
Logins tell you people opened the app. The number tells you whether anything changed. For the agency, that's days from discovery call to proposal, plus the win rate on those proposals.
Check it monthly for the first quarter. If it moves, add the next team. If it stalls, look at the work and the data before you blame the software. A stalled result is information, and it often points back to an earlier decision.
Your AI strategy checklist
- One business outcome, written as a number to move
- One team chosen to start, closest to that target
- The data for that outcome found, checked and tidied
- Written rules for human review and client data
- The owner as decision-maker, plus one named person per AI-assisted task
- A monthly check on that target, kept separate from usage
Think back to the Federal Reserve numbers: most workplaces use AI, and fewer than one in five firms has formally adopted it. With every line on this list filled in, you're in the smaller group. If a line is still blank, that's your next task. Once it's complete, turn it into a 90-day plan with the AI roadmap template for small businesses. Not sure your business is ready to start? The guide to AI readiness assessments explains what to check first.
Ready to build your AI strategy on a real plan?
If one number already came to mind while you read this, start there. The next step is putting a plan and a target behind it.
Our complimentary AI Readiness Assessment shows where your business stands today. It includes an estimate of the productivity value sitting in your current operations, so you know where to start.