The owner of a three-truck HVAC shop forwarded the Team invite from the cab on a Thursday. Dispatch got a login. Two techs asked for one because $20 is cheaper than an argument. The office manager used it twice to rewrite a recall-style confirmation text, then went back to the whiteboard. At Monday's huddle he said, "We have AI now." The after-hours voicemail box was still a black hole. The estimates still left the building on day three.
I've watched the same scene in a dental office after a lunch-and-learn, in a law firm where the managing partner thought "the associates can use it for research" was a plan, in an agency that bought seats so nobody would feel left out, and in a multi-location retailer who added ChatGPT to the "approved tools" PDF. Same mistake in five costumes.
You don't have a ChatGPT for small business strategy. You have calculators. A $20 seat is a general-purpose box. It will draft an email if someone sits down and asks. It will not fix recall, dispatch, intake, proposals, or the Saturday overflow at store two. Those are workflows. A workflow has a trigger, a person, a definition of done, and a way to see if money, time, or the customer moved. A seat has a password.
Procurement is not a plan
Buying seats is a purchasing decision. Rebuilding how a job moves through the shop is an operating decision. Those are different jobs. Owners mix them up because the invoice for seats is easy and the operating work is not.
Easy looks like this: pick a plan, enter a credit card, forward the invite, mention it at the Monday huddle. Not easy looks like this: write down how a new-patient call becomes a kept appointment. Feed the helper last week's real no-shows and the words you actually use. Sit with the coordinator until Tuesday's list is different. Measure filled chairs, not logins.
If your no-show rate is the same, your unclosed HVAC quotes still sit, your consult follow-ups still wait three days, and your store managers still copy-paste weekly notes, the seats did not fail. You asked them to be a system. They are not a system. A useful distinction from Alex Lieberman is this: buying licenses is procurement; a working loop with context is the actual work. Sequence, not seat count.
Hypothetical (not a client, not a promise): eight people on Plus is roughly $160 a month. Over a year that's about $1,900 before tax. Not a crisis. Also not a strategy. The hidden cost is the six months you spent feeling like you'd "done AI" while the same leaks ran. That's the expensive part. Run the number on your seats and your leaks. I will not invent a shop that "got ROI from Team."
What seats actually produce
In every shop I've watched from the outside — and in every honest conversation with owners — you get a split. A couple of people become dangerous. The paralegal who lives in it. The office manager who drafts recall texts on Sunday night. The account manager who outlines a deck from a call. They look like proof the rollout worked.
Everyone else logs in twice, gets generic junk, and goes back to the old way. They will nod in the meeting. They will not change Monday. That split is worse than doing nothing. The power users pull away. The rest of the office feels behind. You, the owner, see the two stars and assume the tool "took." Then you buy more seats, or a fancier seat, and the silent majority stays silent.
Depth beats headcount. One person, the good model, your rules, last week's real work, a week to rebuild one loop — that beats twelve unused logins and a prompting lunch. If you already bought the seats, the next dollar is not another seat. It is a named job attached to one person who already uses the thing without being asked.
What a strategy looks like in a shop
A strategy names the job, the person, the outcome, and the checkpoint. That sentence is the whole difference between a login and a plan.
Dental: the job is recall-and-confirm for hygiene. The person is the coordinator, not "the front desk" as a blob. The outcome is filled openings next week, not "we used AI." The checkpoint is a human sending the text. The helper drafts from the schedule, your script, and three examples of messages you were proud of. If it sounds like a spa in another zip code, you correct it like a new hire, and you save the correction.
HVAC: the job is missed-call to booked diagnostic, or first-pass estimate from the tech's notes. The person is dispatch or the estimator. The outcome is quotes out while the homeowner still cares, or after-hours calls that don't die in voicemail. The checkpoint is a human confirming the window and the price band before anything customer-facing goes out.
Law: the job is intake to first follow-up. The person is the intake coordinator or the paralegal who already owns that queue. The outcome is complete facts and a same-day note to the prospect, drafted by the helper, sent by a human. Research-summarizer seats for attorneys who don't have a research bottleneck are a vanity purchase.
Agencies: the job is call recording to first-draft proposal. One account lead. Real call, real offer sheet, real "we never claim X" list. Human edits. Human sends. Multi-location retail: the job is complaint routing or shift notes that the next manager can actually use. One store lead first. Not all locations on day one. If it works in one store, then you copy the loop, not the login count.
Notice the pattern. One job. One person. Real examples. House rules. Human send. A number you can see in two weeks. That is a strategy a $20 seat can serve. The seat is not the strategy.
The model is not the first step
Owners start with "which model?" because vendors start there. Wrong order. First: the pain that shows up every day. Empty hygiene. Voicemail. Stale consults. Proposals that take a week. Store notes that don't travel. Second: how the work actually happens, on a legal pad, with the person who does it. Not the SOP in the binder. The real path, including the text to the owner's cell.
Third: last week's real examples. Ugly ones. The voicemail that went nowhere. The estimate that needed three rewrites. Then a prototype for that one person. Then you beat it up on edge cases. Then you wire it to live data instead of a pasted screenshot. Then you train the next person. Then it becomes how the shop works, not a side quest.
Most of the work is operations. The model is a multiplier on a loop you already understand. A messy loop plus a model is a faster mess. You already knew that. You just hoped the subscription would spare you the whiteboard.
Use this to turn "we have ChatGPT" into one working loop. Paste it with last week's real examples, not a demo script:
You are a first-draft helper for one loop in my shop.
Shop: [dental / HVAC / law / agency / retail]
The job: [one sentence]
The person who will use you: [name and role]
House rules: [what we sell, who we serve, what we never promise]
Three examples of work we were proud of:
[paste]
One example we regretted:
[paste]
Today's live work:
[paste the list, notes, or transcript]
Write a first draft only. Flag anything you're guessing. Do not send. Do not invent prices, medical claims, legal conclusions, or reviews.
Wishlist hygiene, and what to do with seats you already bought
Pull the list of "AI ideas" on your phone. Be honest. Better intake form. Reminder text. Way to find last year's treatment notes. Status update to the homeowner. First draft of a proposal. Review reply (human still sends; you never fabricate reviews). Weekly numbers from the stores in one page.
A lot of that is ordinary software, a checklist, or a cleaner process. Naming that saves you from a science project. Some of it is a one-person helper with context. Almost none of it is "custom AI" or a new product you should build. Seats don't sort the list. A person with a filter does: Effectiveness (money), Efficiency (time), Quality (customer experience). If a seat isn't attached to one of those on a named job, it's garnish.
Don't panic-cancel tomorrow. Don't add more. Pick one person who already uses it without being asked. Give them the good model if you've been cheap-seating everyone on a weak one. Sit with them and pick one loop. Write the shop rules. Run that loop for two weeks. Measure the thing the loop is for — filled chairs, quotes out, follow-ups sent, complaints closed — not messages generated.
If it works, that's your strategy starting. If it doesn't, you learned something a dashboard login count would have hidden. Unused seats: turn them off. A silent majority of logins is not "optional capacity." It is clutter and a story you tell yourself.
I will not hop on a free call to "get you set up on ChatGPT." That's a YouTube video. I will not sell you more seats. I don't make money when OpenAI does.
FAQ
Is buying ChatGPT for the team a small business strategy? No. A seat is procurement. A strategy names one job, one person, one outcome, and a human checkpoint. Until those four exist, you have calculators.
Should I cancel the Team plan we already bought? Not tomorrow. Attach the seats you keep to one loop and one person. Turn off the unused logins. Adding more seats before a loop exists is how you buy another six months of feeling finished.
What should I measure instead of logins? The number the loop is for. Filled hygiene openings. Quotes out the same day. Consult follow-ups sent. Complaints closed. Messages generated is a vanity count. It will lie to you.
If you want someone to look at your shop — dental, HVAC, law, agency, or retail locations — and say which loop is first, which seats to keep, and what you're skipping, that's the paid assessment: $1,500 for a 45–60 minute owner interview and seven-business-day analysis. You get a map of opportunities scored on money, time, and customer experience. You can implement it yourself or hire anyone, including me later. The map is the product. You already spent money on seats. Spend the next dollars on a plan for the shop, not another calculator.