AI workflow automation for small businesses
Most small businesses do not need an AI strategy. They need three or four repeating jobs taken off somebody's desk, done the same way every time, with a person still reading the result before it goes anywhere.
What it is
Workflow automation means picking one job that repeats on a schedule and building software that does the mechanical part of it. The Monday report someone assembles by hand. The quote that needs a follow-up on day three and again on day ten. The intake email that gets retyped into three different systems. The photos and notes from a job that never make it into a record anyone can search.
The AI's role here is narrower than the marketing suggests. It is good at drafting, extracting, summarizing, and sorting. It is bad at deciding, and it is confidently bad, which is worse. So I build the deciding part in plain rules you can read, use the AI for the language work, and keep a person on the last look before anything reaches a customer.
If a job only happens twice a year, I will tell you not to automate it. Anything you own, you maintain.
What I've actually done
I run my own business this way. Scouts run every week without me: they read search-console data, rank the content gaps, and leave research briefs on my desk on Monday morning. None of them send anything. Every one stops at the point where judgment starts.
Tiller, the lawn app I built and ship, is the same shape at product scale. An agronomy rules engine decides what is true; the AI writes the coaching voice on top. If a combination is invalid, the engine refuses before the AI ever speaks. Its 125 published guide pages were AI-generated and every one passed a verification gate before going live. Generation is cheap. Verification is the product.
On a client's online store this year, the automated part was the diagnosis. I pulled the search-console coverage export, cross-checked it against the live site, and turned 1,356 known URLs into a ranked list of what actually needed attention. The alarming headline was roughly a thousand pages Google would not index. The real number was about 86 genuinely broken pages, plus one theme setting that had been inviting Google to crawl an unbounded set of search and filter URLs. That finding cancelled a copywriting project nobody needed to run.
How an engagement runs
- A plain-English conversation. Where do the hours actually go? No deck, no jargon. We find the one workflow worth fixing first.
- A small first build, fixed scope. You see it working on your real data before you commit to anything bigger. If it does not earn its keep, we stop there.
- Handoff and enablement. Your team runs it. We grow it only if the first piece proved itself.
Who it's for
Owner-operated businesses, roughly two to fifty people, where the bottleneck is one person's evenings rather than headcount. Trades and home services, professional practices, small retail and e-commerce, anyone whose weekly admin has grown into a shape nobody designed. If you already have an internal engineering team, you probably do not need me.
Questions I get asked
How much does it cost?
The first build is fixed-scope and quoted before it starts, so you know the number in advance. I would rather scope it small and be asked to do more.
What happens when the AI gets something wrong?
That is the design problem, not an afterthought. Rules decide, the AI drafts, and a person approves anything customer-facing. Where I cannot build a check, I do not automate the step.
Do I have to move my data somewhere new?
No. Most of what I build reads from the tools you already pay for and writes back into them.
What happens when the engagement ends?
You keep the software and the documentation, and your team has been trained to run it. No retainer is required and there is no lock-in.