What a small business AI consultant should actually do
Most small business AI consulting is a workshop and a slide deck. You leave with a roadmap and a list of tools, and nothing runs. We start from the other end: we sit with the person who does the work, watch a week of it, and write down where the hours go.
Then we rank those jobs by how much time they eat and how safely an agent can take them over. You get a short list with real numbers, not a vision statement.
- A written map of the manual work in your business, by person and by hour
- The two or three jobs worth automating first, and why
- A fixed-price proposal to build and run the first agent
- A clear line on what stays with a human
We build on the systems you already run
Our consultants are the people who write the agents. When we recommend something, it's because we've built it and run it in our own company, not because a vendor pays a referral.
Salesforce, QuickBooks, BILL, Gmail and Slack are where every one of our agents ran first, on real work, before we offered it to anyone.
- Salesforce and QuickBooks Online for quotes, orders, invoices and accounting
- BILL and Stripe for getting paid
- Google Workspace, Gmail and Slack for the conversations around the work
- Airtable and spreadsheets for everything that never got a proper system
AI solutions for small business that keep a human in charge
An agent that sends the wrong invoice or books the wrong appointment costs more than the hours it saved. So every agent we build has an approval step by default: it drafts, a person clicks approve, and only then does anything leave the building.
As you trust it, you loosen the leash. Nothing changes about how your team works until you decide it should.
What the first thirty days look like
The engagement is deliberately short before anything is built, because the expensive mistake in this work is committing to a scope nobody has watched happen yet.
Weeks one and two are observation. We sit with whoever actually does the job, at the time they do it, and write down where the hours go. Not a workshop, and not a survey — the work itself, including the exceptions that never make it into a process document.
Weeks three and four are the decision. You get the ranked list, the recommendation, and a fixed-price proposal for the first agent. If the honest answer is that nothing here is worth automating yet, that is what the document says, and the engagement ends there.
- Week 1–2: watch the work, map the systems, count the hours and the exceptions
- Week 3: rank the candidates by hours saved against how safely an agent can take them
- Week 4: a written recommendation and a fixed-price build proposal for job number one
- No slide deck, no maturity model, no eighteen-month roadmap
What we will tell you not to automate
A consultant who finds an agent-shaped answer to every question is selling, not advising. Some of the most useful output of a discovery is the list of things to leave alone.
Work that happens twice a month is rarely worth the build and the maintenance. Work where the rules live entirely in one person's head has to be written down before anything can be automated, and that writing-down is the real project. Anything where a wrong answer costs more than the hours saved keeps a person in the loop permanently, not temporarily.
- Genuinely low-volume work, where the build costs more than the hours it saves
- Judgment nobody has written down yet — fix that first, then revisit
- Decisions where being wrong is expensive enough that a human check never comes off
- Anything your existing software already does, if somebody would just switch it on
Who this is wrong for
We are a poor fit for a few situations, and it is cheaper for everyone if that surfaces on the first call rather than in month three.
- Companies wanting an AI strategy document to show a board, with no intention of shipping
- Anyone who needs a model trained from scratch on proprietary data — that is a different discipline
- Businesses under about five people, where the manual work usually is not enough to pay for an agent
- Teams that want the code handed over on day one with nobody running it afterwards
What it costs
Discovery is a fixed fee, scoped on the first call. Building an agent is a fixed project price. Running it is a monthly fee that covers hosting, monitoring, and a person on our side who reads the logs.
We'll publish price ranges on this site. Until then, ask, and you'll get a straight number in the first conversation.