What an AI roadmap consultant should build for a small business

A good AI roadmap for a small business is short. It's a ranked list of two to four jobs to hand to AI over the next 6 to 12 months, and for each one: the software it touches, who approves its work, what stays with a person, and how you'll know it's working. If someone hands you a 40-slide deck with a "transformation journey" on it, you've paid for a presentation, not a roadmap.

This guide covers how to sequence the work, what to do in each stage, and what an AI roadmap consultant should leave you with.

Why a small business needs the roadmap more than the tools

Most small businesses have tried AI. Few have put it to work inside the business. The Federal Reserve's survey of small employer firms found that of those using AI, about half were experimenting, 44% had partially integrated it and just 7% had fully integrated it. For firms that planned to adopt it, the top challenges were finding tools that meet their business needs (54%) and the time to implement and train people (37%).

Size makes it harder. The Census Bureau's business survey put overall AI use between 17% and 20% of firms, with fewer than 20% of firms with four or fewer employees using it, against 37% of firms with 250 or more. A shop with six people doesn't have an IT department to sort through vendors. A roadmap does that sorting once, so the next purchase isn't a guess.

What belongs on the roadmap

Pick jobs, not tools. In a service business the candidates are nearly always the same:

• Calls and requests. Missed calls, after-hours requests, web forms nobody answered.
• Estimates. Quotes sent and never followed up, declined work nobody revisited.
• Jobs. Work that stalls between steps, or finishes with no invoice.
• Money. Invoices sent but unpaid, payments not matched in the books.
• Owner reporting. One morning message instead of five logins.

For each candidate, write down four things: what happens today, what it costs you when it slips (in your numbers, not a benchmark), which software holds the data, and who would approve the AI's work.

How to sequence it

This is our opinion, from watching what goes wrong, not a formula.

Months 0 to 2: one job, finished. Pick the job where a miss is most visible and the data is already in your software. For most service companies that's quiet estimates or missed calls. Run it with a person approving every draft. The goal isn't savings. It's learning what the agent gets right and wrong in your business.

Months 2 to 5: loosen, then add a neighbor. Once you've read enough drafts to trust the first job, let it do the low-risk parts without a click, and add a second job that uses the same connections (for example, estimates first, then declined work, since both read the same records).

Months 5 to 9: reporting and money. Add the owner's morning report and invoice or collections follow-up. These read across systems, so they depend on the connections you built earlier.

Months 9 to 12: review and decide. Look at what ran, what you turned off and why. Then decide the next two jobs. Don't plan past that. Tools and models change too fast.

Two rules hold across all of it: no more than one new job in flight at a time, and every job has a person whose name is on its approvals.

What an AI roadmap consultant should deliver

Ask for these in writing before you sign anything:

1. A ranked list of jobs with the reason for the order. Not "Phase 1: Foundation."
2. The systems each job touches and how they'd connect, including whether the software's API even allows it.
3. The approval rule for each job. What the AI drafts, who approves, what never goes out without a person.
4. What it needs from you: logins, a person's time each week, decisions.
5. How you'll measure it, in your own numbers, with a baseline taken before the work starts.
6. What it will not do. A plan with no exclusions hasn't been thought through.

Be wary of a promised return before anyone has seen your numbers. Compliance frameworks like the NIST AI Risk Management Framework (voluntary, organized around Govern, Map, Measure and Manage) matter for companies that build AI. A ten-truck service business needs something more practical: a plan it can run.

Your own numbers are the ones that count. If your dispatcher or office coordinator spends hours a week chasing quotes and balances, put a price on it with your real payroll. For scale, the Bureau of Labor Statistics lists the average dispatcher at HVAC, plumbing and electrical contractors at $50,090 a year. The question for each job on the roadmap is how much of that time it gives back for work that needs a person.

What we do and what we don't

At Cortex the roadmap is the output of the free AI Operations Audit: a 30-minute walk through your workflows and software, and a ranked opportunity map you keep even if you never hire us. It doesn't come with a dollar return estimate, because we won't put one on a business we haven't seen the numbers for. How the audit works is in how an AI operations audit works.

If the map shows work worth building, the path is design, build, deploy and manage, inside the software you already run (see how we work and AI implementation). After launch, managed AI operations covers the upkeep. For the thinking behind agents that read across your systems, see the AI operations layer, and for who should do the work, hiring an AI consultant.

Start with the audit. It's the cheapest way to find out whether you need a roadmap at all.