How to Implement AI in a Small Business: A 90-Day Plan That Does Not Start With a Strategy Deck
You have been told to do something with AI. Here is the 90-day version: a week of looking properly, three weeks of building one thing, and two months of measuring it. What it costs, who needs to be involved, and what to ignore.
- The 90-day shape: week 1 find the workflow, weeks 2–4 build it live to 10% of traffic, days 30–90 measure, widen, start the second.
- Cost of the first 90 days at Wope: $3,600–$6,300 on Starter or Core, plus roughly $100–$500 a month in model costs. Typical agency equivalent: $15,000–$40,000.
- You need one business owner who does the work today, not a steering committee.
- The things to ignore: maturity models, "AI transformation" programmes, and any tool bought before the workflow is chosen.
- Success is a number: hours a month before, hours a month after, written down on day one.
To implement AI in a small business, spend one week finding the single workflow that loses the most hours and is cheapest to automate, three weeks building it live to a slice of real traffic with a test set and a human handoff, and the next two months measuring the saving before starting the second one. The whole first cycle costs a few thousand dollars with a fractional engineer and needs one person on your side who knows how the work gets done.
That is the entire plan. The rest of this article is the detail: what happens in each phase, what it costs, who needs to be involved, and the three things that derail it most often. It is written for owners and operators of 10- to 300-person businesses who have been told to "do something with AI" and would like that something to be real.
Before day one: two decisions
Decide who owns it. One person who does the affected work today and will be its first user. Not a sponsor, not IT, not a committee. They will spend two to three hours a week on this for the first month and they need the authority to say "switch it on".
Decide how you will get it built. The options are in-house, fractional or agency; the comparison is here. For a first system, a fractional engineer working in your own repositories is almost always the fastest and cheapest route, and the knowledge stays with you.
Days 1–7: find the workflow
Do not buy a tool yet. Do not write a strategy. Run the opportunity map: list every repetitive workflow, ask the people who do them how many hours they lose a month, have an engineer estimate weeks to build, and plot the two. The top-right corner, most hours and fewest weeks, is your first project. The bottom-left is a list you write down and do not build.
For most small businesses the winner is one of: repeat support questions, sales-call notes into the CRM, quote and proposal drafts, invoice chasing, or inbox triage. The seven workflows worth automating first covers each.
By Friday you should have one sentence: "We lose 62 hours a month on repeat tickets; we expect this to take 40; three weeks to build." If you cannot write that sentence, you are not ready to build, and finding that out in week one is a good outcome.
Days 8–30: build one thing, live
- Week 2. Real data ingested, deduplicated and inspected. A test set of 100–200 real cases with answers the business owner approves. First commit in your repository.
- Week 3. The system built against the test set. Accuracy scoring wired into the pipeline so any change that makes it worse fails the build. A confidence threshold with human handoff below it. A feature flag.
- Week 4. Live to 10% of real traffic. The owner reviews every handoff. First numbers.
Something real, in front of your team, inside a month. If a builder's plan does not have a live date before day 30, the scope is too wide.
Want the 90 days run for you?
A senior AI engineer, a fixed share of their week, inside your systems. Week one is the map; weeks two to four are the build; every month after is measure and next. Plans from $1,200 a month, cancel any time.
See how Wope works →Days 30–90: measure, widen, repeat
Day 30: compare hours a month now against the sentence from week one. If it is on track, widen from 10% to 50% of traffic. If it is not, the engineer should tell you why and what changes; sometimes the answer is to stop, and that is fine at this cost.
Day 45: widen to most of the traffic. Train the team on what it does, what it does not, and how to read the handoffs. Running costs are now visible; at small-business volume they are usually $100–$500 a month in model and hosting fees.
Day 60: the second item from the map starts. The first continues to be measured; the map is rescored with what you learned.
Day 90: two systems live, one number per system, and a decision about what comes next: stay fractional, hire, or stop here because the obvious wins are taken.
What the first 90 days cost
| Route | Build cost, 90 days | Running cost | Notes |
|---|---|---|---|
| Wope Starter (40 hrs/month) | $3,600 | $100–$500 / month | Map included; one system live, second started by day 90 |
| Wope Core (80 hrs/month) | $6,300 | $100–$500 / month | Map included; two systems live by day 90 |
| Typical fractional retainer | $12,000–$54,000 | $100–$500 / month | At $4k–$18k a month |
| Agency fixed project | $15,000–$40,000 | $100–$500 / month | Usually one system; scoping phase first |
| Consultant assessment, then separate build | $1,000–$5,000 + $5,000–$25,000 | $100–$500 / month | Handoff between advisor and builder |
| Off-the-shelf tools only | $0 | $20–$500 / seat / month | Works for standard tasks; not for your documents or systems |
Sources for the market ranges are in our fractional AI engineer cost guide and AI consultant cost guide.
Who needs to be involved
- The business owner of the workflow (two to three hours a week in month one, less after). Approves the test set, reviews handoffs, says "switch it on".
- The engineer (a fixed share of their week on a Wope plan). Scopes, builds, ships, measures.
- Someone with admin access to the systems involved, for an hour in week one.
- The decision maker, for a 45-minute call in week two and a 30-minute review at day 30 and day 90.
That is the whole team. No steering committee, no transformation office.
The three things that derail it
- Starting wide. An "AI programme" with six workstreams and a Gantt chart. This is how 95% of pilots fail. One workflow, one month, one number.
- Buying the tool first. A subscription to an "AI platform" before anyone has chosen the workflow. The tool then goes looking for a problem, and finds one that was not worth solving.
- Skipping the test set. Without it, nobody can say whether the system got better or worse, so nobody touches it, and it quietly decays. The test set is a week of work and it is the difference between a demo and a system.
The one-page checklist
- One owner named, who does the work today.
- Workflows scored on hours lost and weeks to build; top-right picked; bottom-left written down.
- One sentence: hours now, hours expected, weeks to build.
- Real data ingested and inspected in week two.
- Test set approved by the owner.
- Live to 10% by day 30, behind a flag, with human handoff.
- Measured at day 30; widened or stopped.
- Second item started by day 60.
- Decision at day 90: fractional, hire, or done.
Frequently asked questions
How do I start implementing AI in my small business?
Pick one repetitive workflow that loses the most hours and is cheapest to automate, build it in three to four weeks with a test set and a human handoff, measure the saving at day 30, then decide on the next one. Do not start with a strategy document.
How much does it cost to implement AI in a small business?
A first system built by a senior engineer costs from $1,200 a month at Wope, or $5,000–$25,000 as a fixed project at most firms, plus $100–$500 a month in model and hosting costs. Off-the-shelf tools run $20–$500 a month per seat.
Do I need technical staff to use AI in my business?
Not to start. You need one person who knows how the work gets done and can answer questions a few times a week. The engineering can be a fractional engineer; the knowledge of your business cannot be outsourced.
What is the most common mistake small businesses make with AI?
Starting too wide: an 'AI transformation' with six workstreams and no test set. The businesses that succeed pick one workflow, ship it in a month, and measure it.
How long does AI implementation take for a small business?
The first useful system should be live to a slice of real traffic within four to six weeks. The 90-day plan here covers finding it, building it and proving it paid.