A week-by-week AI implementation plan for a 5 to 50 person business. Thirteen weeks, one job per week, and a running workflow at the end instead of a roadmap deck.
Part of our guide to AI enablement for small business.
Every AI plan you can find online has the same shape. Eight steps. Five phases. Define your goals, assess your data, choose your technology, build a cross-functional team. IBM and Microsoft both rank on page one for exactly that, and neither of them is wrong.
They are also useless to you on a Monday morning.
None of those frameworks tell you what to do in week three. They do not say who owns the thing, what artifact each week produces, or how you know by Friday whether you are ahead or behind. Business owners do not stall out because they lack a five-phase diagram. They stall out because week one arrives, nobody has an hour blocked, and the plan quietly becomes a document.
So this is the calendar version. Thirteen weeks, one job per week, with the owner, the artifact, and the pass condition named for each one. We run this on ourselves. More than 40 agents work inside WE•DO's own operation right now across GA4, Search Console, Google Ads, Shopify, WordPress, and ClickUp, and every one of them arrived through this sequence. The costs and returns from that fleet are published in what running 40+ agents actually costs and saves.
Read it once end to end, then work it a week at a time.
What the 90 days actually produce
Judge the plan on what you keep, not on what you learn. By day 90 you should be holding five things:
- A scored shortlist of every recurring job in the business, ranked by hours times frequency, with the ones AI should not touch marked and explained.
- One live workflow running inside the tools you already pay for, with output landing in your system of record instead of a chat window.
- A one-page SOP for that workflow: trigger, inputs, output, owner, review rule, and what to do when it fails.
- A measured before and after on one number you chose in week two, taken the same way both times.
- A named owner who is not you, plus the next two jobs already scored and queued.
Notice what is not on that list. No AI strategy deck. No tool evaluation matrix. No maturity score. Those are artifacts of consulting engagements, and they are why enterprise AI programs run 12 to 18 months and six figures. You are not running that program. You are handing off one job and proving it stuck.
Before day one: three things that have to be true
Check these before you start, because each one kills the plan quietly if it is missing.
- Someone owns it by name. Not a committee, not "the team." One person with 3 to 4 hours a week on their calendar for 13 weeks. If you cannot free that up, you do not have an AI problem, you have a capacity problem, and this plan will not fix it.
- The work already lives in shared systems. If your records are on paper or scattered across personal drives, spend two to four weeks organizing first. That is not a delay, it is the highest return work available to you.
- You will measure one number. Pick it in week two, take the baseline before you build anything, and take it again the same way in week 10. Without that, you get opinions about whether it worked.
If the plan does not survive one person taking a vacation, it is not a system yet. It is a hobby.
Days 1 to 14: Pick the lane
Two weeks on choosing sounds slow. It is the highest leverage time in the whole plan. Almost every failed rollout we have looked at failed here, on scope, and then spent the next 11 weeks paying for it. The full set of failure patterns is in why most AI agent pilots fail.
Week 1: log where the hours actually go
Do not brainstorm the list. Log it. For five working days, everyone on the team writes down the recurring jobs they touch, how long each one took, and what triggered it. A shared sheet with four columns is enough: job, trigger, minutes, output.
The list you get back will not match the list you would have guessed. It never does. The jobs that hurt are rarely the glamorous ones, they are the assembly work: pulling numbers into a report, retyping a call into tasks, chasing the same three approvals, formatting the same deliverable. Our guide to what to automate first covers the categories where those hours usually hide.
Week 2: score the candidates and commit to one
Rank by hours times frequency, then apply one filter that matters more than the math: does the middle of the job follow rules? Judgment at the edges is fine, that is what the human keeps. Judgment all the way through means leave it alone.
Good candidates share three traits. They fire on a predictable trigger, the middle is mechanical, and the output has an obvious home. Then commit to exactly one and write the number you will move next to it.
Days 15 to 30: Write the contract
Weeks three and four produce paper, not software. That feels like stalling. It is the difference between a workflow your team trusts and one they quietly stop using in month two.
Week 3: map the trigger, the inputs, and the landing spot
Write down what starts the work, what the work needs before it can run, and where the finished output has to land. Be specific enough that a new hire could follow it. "Pull the analytics" is not an input. "Sessions, conversions, and top landing pages from GA4 for the trailing 30 days, compared to the prior 30" is an input.
Then write down what the workflow should ignore. Most performance problems are context problems, and the fastest way to make output inconsistent is to hand it every document you own.
Week 4: write the SOP and the one-page gate
The SOP is one page: purpose, owner, trigger, input checklist, output example, review rule, and the escalation path when it goes wrong. If you cannot fit it on a page, the scope is still too big.
The gate is the other half, and it is the part teams skip. It answers three questions and nothing else: what can run without review, what needs a person to sign off before it leaves the building, and what AI never touches. Write it before the build. Writing it after the first bad output reaches a client is a much more expensive way to learn the same lesson.
Days 31 to 60: Build it and run five real cases
Thirty days on one workflow is generous on purpose. You are not building until it runs. You are building until it runs reliably, which is a different finish line.
Weeks 5 and 6: connect the sources, ship version one
Build inside the systems the work already lives in. If a person still has to move the output by hand, you automated the easy half and kept the annoying half. Then run case one against real work, with the owner watching. This is also the week the platform question gets answered, and our tool guide by category is built to be read at exactly this point and not before.
Weeks 7 through 9: five real cases, including the ugly ones
Idealized test cases teach you nothing. Run five that look like your actual week:
- One clean case, where everything is where it should be.
- One messy case, with a bad transcript or an incomplete brief.
- One case with a source missing entirely.
- One case where the right answer is to stop and escalate.
- One case at real volume, on the day the work normally lands.
Log every failure in one place with the input that caused it. Then fix the instructions and the source connections before you touch the model. In our experience the model is almost never the problem. Week 8 is for deliberately breaking it, and that week is the reason the workflow survives month four.
Days 61 to 90: Measure, train, and hand it over
The last month is about making the workflow independent of the person who built it. Skip this and you have not removed a bottleneck, you have moved it.
Week 10: measure against the old way
Take the same number you baselined in week two, the same way. Hours returned per week, missed steps eliminated, time from trigger to finished output. Volume of AI output is not a result, so do not report it.
Weeks 11 and 12: train on the workflow, then transfer ownership
A tool demo does not create adoption. Budget 2 to 4 hours per person up front, then 30 minutes a week of guided practice for the first month. Your team needs to know when to reach for it, what to feed it, how to spot a confident wrong answer, and where to report a failure.
Then hand it over in writing. Name the owner, set the review cadence, and put the SOP where the work happens instead of in someone's inbox.
Week 13: score job two
Go back to the week one log, re-score it with everything you now know, and pick the next job. This is where compounding starts. Job two takes a fraction of the time because the hard parts, the gate, the source access, the review habit, are already built.
What the 90 days cost
Enterprise AI enablement runs $200,000 or more across 12 to 18 months. A 20 person company does not need that, and the real budget looks nothing like it.
Add it up and a self-run 90 days lands somewhere between $200 and $1,200 in software plus roughly 30 to 50 hours of internal time. The hours are the expensive half, and they are the half most plans never book.
If you would rather not spend those hours, this same sequence is what we install for clients as the 90-Day AI Operations Buildout, starting at $5,000 a month for a three month engagement, with a $997 workflow audit if you want the scored shortlist and nothing else first.
Five ways the 90 days go wrong
- Starting with a tool. The tool is a setting, not a strategy. Start with the job that eats your week and pick the tool in week five.
- Scoping a department instead of a job. "Run our marketing" is a wish. "Turn every client call into assigned tasks within an hour" is a workflow.
- Skipping the gate because you are small. You are not too small for one page of rules. You are exactly the size where one bad client-facing output costs real money.
- Testing on clean inputs. A workflow that only works when the transcript is perfect does not work.
- Never transferring ownership. If the builder is still the only operator on day 91, you built a dependency.
FAQ
What is a 90-day AI implementation plan?
It is a dated schedule for handing one recurring job to AI: two weeks to pick and score the job, two weeks to write the SOP and the review rules, four weeks to build it and run it against real cases, and four weeks to measure, train the team, and transfer ownership. It differs from an AI strategy in that every week produces an artifact you keep, and the plan ends with a workflow running rather than a document.
Is 90 days realistic for a small business?
For one workflow, yes, and it is usually generous. Most of the calendar is deliberately spent on choosing the job and testing the messy cases, because that is where rollouts fail. A second workflow typically takes four to six weeks because the access, the gate, and the review habit already exist.
What should the first AI workflow be?
One with a predictable trigger, a mechanical middle, and an obvious home for the output. Meeting notes into assigned tasks, client status reporting, inbound request routing, and content QA against an approved brief all qualify. Anything where judgment runs through the whole job, like pricing a custom project, does not.
Do we need clean data before we start?
You need organized data, not perfect data. If you can export your customer list as a usable file and your team works in shared systems rather than personal drives, you are ready. If the records live on paper or in untracked spreadsheets, spend two to four weeks organizing first.
How do we measure whether the plan worked?
Pick one number in week two and baseline it before you build. Hours returned per week, missed steps eliminated, or time from trigger to finished output all work. Take the measurement again in week 10 the same way. Do not report volume of AI output, because more output is not the same as more capacity.
Do we need a developer?
Usually not for the first workflow. You need a developer when the job crosses several systems that have no existing connection, requires a private API, or needs controls your current tools cannot enforce. Decide that in week five, not week one.
Find your week one job in 30 minutes
Bring the log from week one and we will score it with you on a fit call. If you would rather have the whole sequence installed, that is the 90-Day AI Operations Buildout.
Book a 30-minute fit call or see how we build AI into an operation.
