Custom AI tools for small businesses
Off-the-shelf software makes you work its way. A small tool built around how your business already runs is often a couple of weeks of work rather than a platform migration, and you own it at the end.
What it is
A custom AI tool is a small piece of software that does one thing your business needs and nothing else. A quoting tool that knows your catalog and your margins. An intake form that reads a customer's messy email and fills in the record. A search box over ten years of job notes that actually finds the answer. A checklist that writes itself from the photos a crew took on site.
The reason to build rather than buy is not that software has become cheap. It is that the fifth subscription you bend your process around costs more, in hours and in frustration, than one tool fitted to the process. The reason not to build is that anything you own, you maintain. When buying is the right call I will say so, and I have no affiliate deals that would make me say otherwise.
Small and sharp beats broad and half-used. One tool that gets opened every day is worth more than a platform nobody logs into.
What I've actually done
Tiller is the long answer. It is a live AI lawn coach with an agronomy rules engine that decides what is true, an AI layer that only writes the coaching voice, a Stripe billing rail running in production, and 125 verified guide pages. Product strategy, design system, front end, engine, billing, and content pipeline: one person. That is the depth behind a two-week client build.
Inside this studio, an AI-visibility scoreboard asks AI assistants the questions customers actually ask and scores whether a business gets named in the answer, with the real citations attached. I built it for my own product first. It is now the basis of a client service, which is my rule for tooling: run it on my own business before selling it to yours.
For a client's online store, the catalog had 100 products with no search title and 33 with no description. The build was not a rewrite of the store. It was a script that drafts the missing fields from the real product data and hands them back for row-by-row approval by the owner, because the copy is his voice and his call.
How an engagement runs
- A plain-English conversation. I want to see the actual workflow, including the spreadsheet with the tabs nobody outside the business understands.
- A small first build, fixed scope. A working tool on your real data, quoted before it starts. If it does not earn its keep, we stop there.
- Handoff and enablement. Your team runs it, the documentation is written for them, and the accounts are in your name from day one.
Who it's for
Businesses of roughly two to fifty people with a process that is genuinely their own: a quoting method, a routing rule, an inspection sheet, a catalog with quirks. Those are the places where off-the-shelf software fits worst and a small custom tool pays for itself fastest. If your process is standard, buy the standard product.
Questions I get asked
How long does a first tool take?
Most first builds are one to three weeks of work. The scope is set before it starts, so the timeline is a commitment rather than an estimate that drifts.
Who owns the code and the accounts?
You do. The code, the documentation, and any API accounts are yours and in your name from day one.
What if I would rather buy something off the shelf?
Then buy it. I will tell you when an existing product covers the need, because a tool you do not have to maintain is cheaper than one you do.
Does it have to use AI?
No. Plenty of useful tools need no AI at all, and some of the best results come from ordinary rules with an AI layer only where language is involved.