What AI Operations Actually Looks Like for a 20-Person Service Business

Most of what's written about AI operations is written for companies with 500+ employees, a CTO, and a budget for a six-month transformation. That's not who I work with.

I work with service businesses — 10 to 100 people, an owner or GM who runs everything, and a team that spends half its week on admin work that has nothing to do with why they were hired. Restoration companies. Property managers. Contractors. Consultants. The kind of businesses where the person making the AI decision is also the person signing the checks, managing the crew, and sometimes still doing the work themselves.

AI operations for these businesses doesn't look anything like what McKinsey writes about. Here's what it actually looks like.

Week 1: Find the bleed

Every engagement starts the same way. I sit down with the owner or GM and ask one question: what does your team's week actually look like?

Not the org chart. Not the process documentation (there usually isn't any). The actual week. Monday through Friday, where does the time go?

The answer is always some version of: "We spend about half our time on stuff that isn't the actual work." Dispatching. Data entry. Chasing people for updates. Rebuilding the same report. Copying information from one system to another. Following up on things that fell through the cracks.

I map that. Every hour, every task, every handoff. By the end of the week, we've identified the three workflows that are bleeding the most time or losing the most revenue. Then we pick one — usually the one that's most painful or most expensive — and that's what we build first.

Weeks 2-4: Build the first system

This is the part that surprises people. We're not building a platform. We're not migrating systems. We're not running a pilot with a committee reviewing results in 90 days.

We're building one working AI system that handles one specific workflow. It might be an SMS assistant that logs client details into the CRM automatically. It might be an automated report generator that replaces the Thursday-afternoon spreadsheet scramble. It might be a proposal tool that drafts responses in minutes instead of hours.

Whatever it is, it goes into production. Not a demo. Not a proof of concept. A system your team uses on Monday morning.

The timeline depends on complexity, but most first systems ship in two to six weeks. The constraint is usually integration — connecting to whatever CRM, scheduling tool, or billing system the business already runs on. The AI part is fast. The plumbing takes longer.

Month 2+: Find the next one

Once the first system is running, something interesting happens. The team starts noticing other things. "If we could do that for dispatching, could we do it for invoicing?" "What about the weekly report?" "What about customer follow-ups?"

That's when the engagement shifts from project to ongoing. I embed with the leadership team — typically a few hours a week — and we keep shipping systems. One at a time. Each one compounds on the last because the infrastructure is already in place.

This is what "Fractional AI Operations" actually means. It's not a subscription to a platform. It's not a consulting retainer where someone sends you a deck every month. It's a part-time operator who shows up, builds the next system, makes sure it's working, and moves on to the next one.

What it doesn't look like

It doesn't look like a six-month discovery phase. It doesn't look like a 40-slide roadmap. It doesn't look like an enterprise sales cycle with procurement, legal review, and a pilot committee.

It looks like a conversation, a build, and a deploy. Repeat.

The economics

For a 20-person service business, the math usually works out like this: the first system saves 10-20 hours per week across the team. At a blended cost of $40-60/hour (including the opportunity cost of what those people could be doing instead), that's $2,000-$5,000 per month in recovered capacity. The engagement costs less than that.

By the third or fourth system, the cumulative savings are usually 3-5x the cost of the engagement. That's when owners stop thinking of it as an expense and start thinking of it as infrastructure.

Is your business ready?

Honestly, not every business is. If you have fewer than 5 people, you probably don't have enough repeatable process to justify building AI systems around. If you're still figuring out your basic workflows, AI will automate the chaos rather than fix it.

But if you have 10+ people, established workflows that eat time every week, and a CRM or scheduling system that's half-empty because nobody has time to use it properly — you're ready. The admin bleed is real, and it's compounding every month you don't address it.

The first step is a conversation. Tell me what your team's week looks like. I'll tell you what I'd build first.