AI aimed at the job that eats your week
For businesses where somebody spends hours every week doing something a computer should be doing, and nobody has had time to fix it.
What usually goes wrong
Most AI projects in small businesses start from the wrong end. Someone reads that AI will transform the industry, buys a tool, and before long it is a browser tab nobody opens. The technology was never the problem, so the technology was never going to be the answer.
Meanwhile the actual cost is somewhere much more boring. Quotes typed out by hand every evening. The same figures re-keyed from one system into another. A question customers ask forty times a week that somebody answers forty times a week.
How I do it
The work starts from the task, not the technology. What does someone here repeat, how often, and what does it cost when it slips. Only then does it matter what the fix is built from.
A lot of the time the honest answer is that it does not need AI at all. A form, a template and a reminder will solve more small business problems than a model will, and they are cheaper to run and far easier to trust. When I tell you that, it is not me talking myself out of work, it is the difference between something you still use and something you abandoned.
Where a model genuinely earns its place, reading messy documents, sorting unstructured text, drafting from your own material, it gets scoped narrowly and given a way to be checked. Anything that can be wrong in a way that costs you money gets a human in the loop by design.
- Estimating and quoting tools that use your own pricing and history
- Back-office automation across the systems you already pay for
- Document and data extraction, replacing manual re-typing
- Internal assistants grounded in your own content, not the open internet
Evidence for this specifically
Toggled iQ
Commercial building experience, pointed at your home or business.
Before this practice I ran product for a smart building platform used across commercial portfolios in more than thirty states. The problems are the same as a home, only less forgiving: devices from vendors that disagree, networks that were never planned, and automations somebody has to trust at two in the morning.
This site
You are already looking at a sample of the work.
Every page here is real HTML rendered ahead of time, so it loads fast and reads correctly to search engines and AI assistants. There is no cookie banner because nothing is tracking you that would need one. The assistant in the corner answers questions about my work from the site's own content.
Questions people ask about this
- Do I actually need AI?
- Often no, and I will tell you when the answer is no. Plenty of what gets sold as AI automation is a form, a scheduled job and a notification. If that is what your problem needs, that is what I will build, and it will be more reliable and cheaper to run than the version with a model in it.
- Is my business data safe, and does it train someone's model?
- Business data goes to a commercial API under terms that exclude training on it, and I set that up explicitly rather than assuming it. Nothing gets pasted into a consumer chatbot. Which systems hold what, and who can see it, is written down as part of the work rather than left implicit.
- What happens when it gets something wrong?
- It will, sometimes, and anything built as if it will not is badly built. The design question is what a mistake costs. Low-cost mistakes can run unattended; anything touching money, a customer commitment or a legal document gets reviewed by a person before it goes out. You should be able to see what it did and why.
- How small can we start?
- One workflow. Pick the thing that annoys you most, get that working, and judge the rest on whether it actually helped. Starting with a whole strategy is how these projects end up as a slide deck rather than something running on a Tuesday.
Tell me what you are trying to build
A few sentences is enough. If this is not the right fit I will say so and point you somewhere better.