The HVAC owner is at the kitchen table with last month's P&L. Five thousand dollars is a truck payment. It's a coordinator. It's the difference between hiring this year and waiting. Dispatch has a $20 ChatGPT login. Missed-call voicemails still dump into one inbox. Estimate follow-up still depends on which salesperson remembered. He's staring at a $4,000+ line item, and he should be. If I can't tell him what that money buys on a Tuesday in his shop, he shouldn't pay it.
I am not selling a lab. I am not selling a binder of "AI SOPs" nobody will open. I am not selling a chatbot with your logo on it. I sell ownership. One workflow a month. Visible. Done. Then the next one.
That's the job of a fractional chief AI officer for a small business. Not a title on a slide. A person who picks the bottleneck, sits with the person who lives in it, and does not leave until Tuesday actually looks different. A useful distinction from Alex Lieberman is this: buying seats is procurement; a working loop is the job.
You already bought the seats. Nobody owns the work.
Look at your office. Dental: the front desk has ChatGPT open. They used it twice to rewrite a recall text. Hygiene still calls the same list by hand. Insurance follow-up still sits in a sticky-note pile. HVAC: the dispatcher has a subscription. After-hours still dies in voicemail. Law: a paralegal drafts with a model. Intake still lands in a shared mailbox that the managing partner checks at 9:40 p.m. Nobody tagged the matter. Nobody started the clock.
Agencies: the team has licenses. The CRM still has 400 "leads" with no next action. Multi-location retail: each store manager asked ChatGPT for a Google reply once. The three-star review from last Thursday is still unanswered. That's licenses. That's not a system. A retainer exists because no one on payroll is paid to finish the loop. Your office manager is running the desk. Your lead tech is on a truck. You are still the help desk after 6.
So the work stays half-done. Tools accumulate. Nothing ships. The monthly fee is not "AI strategy hours." It is someone whose job is to make one loop real, write it down so the shop can run it without them, and come back next month for the next one.
What you actually buy in a month
The month is the unit. Not a 40-item backlog. Not a "transformation." One workflow. We pick it from a ranked list you already paid to see. That's the assessment. The retainer is me executing what that map already named.
Week 1: we draw the real job, not the one in the employee handbook. Who touches it. Where it dies. What "done" means in dollars, hours, or a customer who doesn't have to call twice. We cut the dumb steps before we add a tool. Faster mess is still mess. Week 2: we build a helper for one person. Ugly is fine. The office manager, the dispatcher, the intake coordinator — they are the first user. We feed it last week's real examples, not a demo script. We write the house rules: what we never promise, what we never send, when a human has to look.
Week 3: they use it on live work with a checkpoint. Customer-facing messages still go out from a human. No auto-send on anything that sounds like legal advice, medical advice, or a promise about a job. If it breaks, we fix the step, not the slogan. Week 4: we measure the one number we named. We write what we built into a shared hub you can open without me. Recording, three takeaways, the artifact, who owns it now, what we parked. Then we pick next month's workflow from the same list.
If that sounds slow, good. Shops don't fail because they didn't buy enough tools. They fail because they tried to rebuild intake, reviews, scheduling, and the CRM in one sprint and went back to the old way by Friday.
Quote the month. Not the pile.
You will hear other people quote a year of fees against a fantasy savings number. I won't do that to you as a sticker. You hear $4,000+. That's the number. Compare it to one leak on your own books, using your numbers, labeled as a model — not a promise.
Hypothetical (not a client, not a promise): HVAC, two locations. Four estimate follow-ups a week that die in the salesperson's head. Average ticket $4,200. Close rate on a live follow-up, call it 25%. If you recovered even two of those a month, that's north of the retainer before you count the dispatcher's hours. If you recovered zero, we killed the workflow and picked a different one. That's the test. Effectiveness first. I will not invent your close rate. I will not put a made-up "annual transformation" on a slide and divide by twelve after you already feel committed.
If the math only works when we stack twelve unbuilt workflows into one scary total, the offer is theater. We quote the month. We ship the month. The assessment puts the leaks on paper with assumptions written down. Then you decide if a month of ownership is rational.
What a month looks like in HVAC or an agency
Say you run HVAC. Multi-location or heading there. Month one is not "an AI receptionist that also does billing." Month one is missed-call capture: voicemail and web form into a tagged queue, a first-draft text the dispatcher edits, a rule for after-hours vs. emergency vs. maintenance. Human sends. You measure: how many leads sat more than 30 minutes, how many estimates got a same-day follow-up.
Month two might be the estimate itself. Call notes in, first-draft scope out, in your price book language, with the "we don't lowball to win" rule in the file. Estimator still prices. The helper does the blank page. Month three might be review replies and the ask — after the job is actually done, not while the customer is staring at a hole in the wall. Each month, one loop. You can walk into the shop and see it. If I disappeared in month four, you would still have months one through three.
That's Efficiency: the dispatcher stops living in voicemail. That's Effectiveness: more estimates leave the building. That's Quality: the customer isn't calling twice to ask if you got the message. If a tool doesn't hit one of those three, we don't install it. I don't care how good the demo was.
Agencies already understand a retainer. You know what "hours with no artifact" feels like from the other side. Don't buy that from me. For an agency, month one is usually the graveyard: CRM statuses that mean nothing, the follow-up that only happens when a salesperson feels guilty. We define next action. We draft the sequence. A human still hits send on anything that could be a promise. Month two is often proposal first drafts from the call transcript. Month three might be the weekly client update that currently takes an afternoon of archaeology. If it produces every month, you price every month. That's how you already bill your own clients.
What I will not do on your dime
I will not spend month one writing a 40-page strategy you could have gotten from a blog. I will not roll out a tool to the whole office before one person trusts it. I will not auto-send patient explanations, legal conclusions, or "we can definitely get that permit." Human still sends. If that slows us down, we stay slow. Quality is a door. We don't skip it to look modern.
I will not keep a secret stash of prompts in my account. The hub is yours. The context files are yours. The workflow is written so the office manager can run it when I'm not on the call. I will not pretend a custom app is the answer because it bills more. Most shops need a cleaned-up process plus a helper sitting on top of the software they already pay for. I will not call this a lab. It's an operator seat. I show up. We pick. We ship.
Renewal is not a vibe. Open the hub. You should see: the workflow we named, the before (how it actually ran), the after (who does what on Tuesday), the artifact, the number we watched, and the next item already queued. If the hub is empty and I'm "available for questions," you're buying office hours. Don't.
Use this with a spouse or partner before you even talk retainer:
Retainer smell test (fill in, then decide):
Monthly fee: $4,000+
First workflow, in one sentence:
Person who will run it on Tuesday:
Number we will watch in four weeks:
Human checkpoint (who sends):
What we will not automate:
If this number doesn't move, we will: kill / change the loop — not invent a second science project.
The assessment map we already paid for lives at: [link or folder]
If you cannot fill this, you are not buying a retainer. You are buying a vibe.
We do not start with $4,000+. You should not either. Use the free 15-minute Mini Margin Check if you need a first look. Pay $1,500 for a Margin Scan when you want a 45–60 minute owner interview, ranked opportunities, and a seven-business-day decision. The embedded relationship comes only after that map exists.
FAQ
What does a fractional chief AI officer actually do in a small business? Owns the next workflow every month. Maps it, sits with the person who lives in it, installs a helper with a human checkpoint, writes it into a hub you can open without them, then picks the next loop from a ranked list.
Why $4,000+ a month instead of a project fee? Because a working loop produces every month, and someone has to keep it from rotting. We quote the month, not a year of unbuilt workflows stacked into a fantasy savings number.
Do I start with the retainer? No. Use the Mini Margin Check if you need a first look. Use the Margin Scan if you want the map. Continue monthly only after the map exists and you want someone to own the next loop. We will not skip the line.
If you want the ranked list and a clear "start here," that's the assessment: $1,500 for a 45–60 minute owner interview and seven-business-day analysis of your shop. Bring last week's real mess. Missed calls. The inbox. The estimate that sat. I don't need an org chart. I need the work.