AI automation for small business: a realistic 90-day roadmap
What to do in the first 30 days, the next 30, and the final 30 — with the traps every SMB falls into and the numbers you should actually measure.
Ninety days is roughly the window in which a small or mid-sized business can go from "we've been meaning to look at AI" to "AI is materially running three or four things every day." Longer than that and momentum dies; shorter and you are shipping toys, not systems. What follows is the plan we run, more or less unchanged, inside client businesses between 20 and 300 people.
Days 1–30 — See clearly, then ship one thing
Week one: honest inventory. Every workflow in the business that touches a customer, a dollar, or a decision. Not the org chart. The actual moves your operators make in a day. Two-thirds of the win is knowing what you're looking at.
Week two: pick one workflow. One. Ideally something that (a) runs daily, (b) is measurable, and (c) costs a human operator more than four hours a week. The best first candidates are almost always at the top of a funnel — speed-to-lead , intake, first-touch — or in document handling (parsing invoices, statements, applications).
Weeks three and four: ship it end-to-end. Not a pilot. A production system your team uses on Monday. This is the single highest-signal thing you can do in the first month. The organisation learns more from one shipped system than from ten decks.
Ninety days is the window between "we've been meaning to look at AI" and "AI is running three or four things every day." Anything longer, momentum dies.
Days 31–60 — Extend and connect
With one working system in production, three things become possible that were not possible before.
Extend the win. If you automated intake, extend into qualification and routing. If you automated document parsing, extend into approvals. Compounding matters more than breadth here.
Connect the systems you already own. Most SMBs run a hundred pieces of software . AI adds real leverage only when it can read across them. Days 31–60 is where you build the plumbing.
Start the second workflow. Now pick a workflow that touches a different part of the business — usually finance or operations. Two shipped systems in two departments is what makes AI feel real to a company, not one shipped system in one department.
Days 61–90 — Institutionalise
By day 60 you have systems. By day 90 you need them to be durable — meaning someone owns them, someone monitors them, and they survive a change of personnel.
Owners named. Every automated workflow has a human owner. Not a committee. One person whose job includes checking that it still works and knowing what to do when it doesn't.
Metrics wired in. Time saved. Errors caught. Response time. Revenue captured. These are the numbers that keep the CFO patient and the CEO honest.
The next quarter's backlog. A written list, prioritised, of the next five workflows — because momentum is the only thing that matters now.
What most SMBs get wrong
The most common failure mode is starting with a large, ambitious project — usually a "customer-facing AI" — before shipping any small, internal one. The customer-facing project takes six months, produces nothing, and burns the appetite the organisation had for the whole topic. Internal, boring, measurable systems build the muscle and the trust required to earn ambition later.
The second most common failure is treating AI as a technology purchase rather than an operational change. Buying a subscription does nothing. The work is in redesigning the processes so the software has something useful to do.
What ninety days should look like
Two to four workflows automated end-to-end. Named owners. Measured impact. A written backlog for the next quarter. An organisation that has stopped asking whether AI works for it and started asking what to ship next. That's the finish line — and it is entirely reachable inside a quarter for any business that decides to.
If any of this maps to a decision you're weighing, a call is the fastest way to know whether we're the right people to build it. If we're not the fit, we'll say so.