Practical AI for small businesses.
Most small businesses have one or two jobs that eat hours every week and don't need a person doing them. I find those, build something that handles them, and stay on until your team is running it without me. In practice it looks less like "AI" and more like a Monday email that says: here are the six invoices worth a second look.
20 minutes. No deck. If there's nothing here worth building, I'll say so. Not ready to talk? The audit gives you the same read on your week, without me in the room.
Three things, done in order.
Find the workflow
Which one or two jobs are worth the effort, and just as often, which ones aren't.
Build it
AI doing the heavy lifting, plus enough custom code to fit the way you already work.
Stay through adoption
The build isn't the hard part. Getting it used is. I stay until it's running without me.
It doesn't look like "AI." It looks like this.
Catching invoice errors and price creep
A restaurant takes in about forty emailed invoices a week. Every line gets checked against what they paid last time. Monday morning the owner opens one email: six items moved this week, three of them nobody agreed to. Four minutes to read, and the credit request is already drafted.
The mail that needs you
Every inquiry gets read and sorted before it reaches you. Twice a day you get a short brief: what needs a decision, what's just noise, and the routine replies already drafted and waiting on a yes. Your first response goes out in hours instead of Thursday.
Drafting and comparing documents
A twelve-page vendor contract lands. You get back the four clauses that differ from the last one you signed, in plain English, with the changes marked. You still make the call, you just make it in ten minutes instead of over a weekend.
When two systems disagree
Your POS says one thing, your books say another, and someone loses every month-end finding out why. That comparison runs overnight instead, and the mismatches arrive as a short list with the likely cause attached.
The retyping
A new client fills out an intake form. Someone retypes the same dozen fields into the CRM, then the scheduler, then the invoicing system, and gets one wrong often enough that nobody's surprised anymore. Now it's typed once, lands in all three, and anything that doesn't line up gets flagged instead of guessed at.
Any one of these is small. Together they're your Sunday night. If something here rings a bell, there's probably a project worth talking through.
You can point straight at the problem, even if you can't yet describe what to do about it.
That's exactly the right starting point. Owner-operated businesses, real pain, no IT department. If you've been told you "should be doing something with AI" and don't know where to start, that's the gap I close.
Paperwork that piles up
Contracts, invoices, quotes, proposals, BEOs (banquet event orders), intake forms. Operations where the paper arrives faster than anyone can get to it.
Where small leaks matter
Restaurants, agencies, professional services, contractors. A few dollars a week on the wrong line adds up to real money by December.
Close enough to the work
You know where the time goes because it's your time going. Nobody has to interpret the problem for you.
A small, honest process with defined edges.
A short conversation Free · 20-30 min
We talk about your operation, where the time and the money seem to go, and what you've already tried. I'm honest about whether there's a worthwhile project. Sometimes there isn't, and saying so is more useful than pitching one.
A scoped engagement You see the number before you commit
If there's a fit, I write up a small, specific project: what gets built, what success looks like, and what it costs. One price, in writing, no multi-month "discovery phase" in front of it.
The build Usually a few weeks
You see progress weekly, not at the end. I build it alongside the people on your team who'll be using it, so it fits how they work before it ever goes live.
Adoption and handoff Through the first stretch of real use
I stay on while it beds in, tune what isn't working, document what needs documenting, and step back when the workflow is yours.
Minerva Logic is run by Craig Finch. Thirty years of building software, most recently on where these AI tools hold up under real work and where they fall over. Based in the New York metro area, working with clients there and remotely.
The practice is deliberately small. I take on the projects I can do well, and I'd rather refer you to someone else than overcommit. If we work together, you're working with me - not a team of contractors you've never met.
What's the first thing you'd hand off?
If you were referred by someone, mention who. It helps me figure out where to start. If you found me cold, tell me a little about your business and where the time goes. I read everything and usually reply the same day.