Twelve of Your Twenty-Two Employees Are Seasonal Taproom Staff. Per-Seat AI Pricing Was Never Built for That

How many of the people on your payroll would you actually buy a software seat for?

Count honestly. A twenty-two-person brewery with a taproom: an owner, a head brewer, two cellar staff, a packaging lead, a part-time lab tech, an office manager who also does compliance, a sales lead, a delivery driver, a taproom manager, and twelve hourly bartenders and servers, nine of whom you hired in April and will lose in October. Of those twenty-two, maybe seven have a desk. Three have a company email address they check.

Per-seat software pricing was designed for a company where everyone sits at a screen. It was not designed for an industry where two-thirds of the staff are on their feet, on a forklift, or behind a bar, and where your headcount swells fifty percent between May and September and shrinks back after the last patio weekend.

Why does per-seat pricing feel wrong in a taproom?

Because it charges you for presence and you need it for bursts. Your taproom manager would use an assistant hard for four days before a can release and then barely touch it for three weeks. Your sales lead wants it the week before a chain reset. Your office manager wants it in the first two weeks of the month when the reports are due and the excise return is looming. Nobody uses it evenly, which is exactly what a monthly per-person price assumes.

Then there is the seasonal problem. Suppose business plans run around $30 per person per month — use your own vendor's real number, this is an illustration. Adding your nine summer hires for six months is $1,620 you spend on people who will each log in twice. Most owners respond by not giving them seats at all, which means the assistant never reaches the part of the business where the customer questions actually land.

What is actually on offer now that was not a year ago

The open side of this got serious. Moonshot AI released Kimi K3, a 2.8-trillion-connection model built so that only the relevant slice of it does the work on any given question — the largest openly published model in the world, and free to download and run on your own hardware. More broadly, the open tier closed much of the gap with the paid frontier models over the past year. Two years ago "run your own" meant accepting something noticeably dumber. In July 2026 it does not, for the kinds of jobs a brewery actually has.

The choice in front of an owner is no longer whether the free option is good enough; it is whether you would rather rent by the head or own a machine and pay somebody to look after it.

That second clause is where most trade-press coverage stops and where the real cost lives. A model you host does not have a support line. It has whoever in your building is willing to own it.

flowchart TD
  A["Monday: 340 can and menu descriptions needed for the fall lineup"] --> B["Job left running overnight on the brewery's own box"]
  B --> C["Draft taproom menu lines and tap list copy"]
  B --> D["Draft distributor sell sheets by market"]
  B --> E["Draft shelf-talker and case-card copy"]
  C --> F["Taproom manager edits over coffee"]
  D --> G["Sales lead edits before the distributor meeting"]
  E --> H["Owner signs off before it goes to print"]

Where does the volume actually justify a machine?

Seats are the wrong way to think about it. Volume is the right way. There are jobs in a brewery that are enormous in count and low in stakes, and those are the ones that make a machine of your own pay:

• Product copy at scale. If you release forty beers a year across taproom, distribution and a web store, every one needs tap-list copy, a menu line, an Untappd description, a sell sheet, a shelf-talker and a social post. That is hundreds of short pieces of writing a year, all of which need a human edit and none of which need a human first draft.
• Reading depletion data every night. Your distributors send depletions through their reporting systems; most breweries look at that monthly, when a placement that stopped selling in week two has already been dead for six weeks. Reading all of it every night costs nothing once the machine is bought.
• Review and mention triage. Untappd, Google, the taproom's reservation notes. Not to reply automatically — to tell your taproom manager on Monday that four people mentioned the patio heaters and two mentioned a specific bartender.
• Recipe and batch history questions. "When did we last brew this with the old yeast strain and what did it finish at?" That is a question your head brewer asks maybe once a week and currently answers by scrolling.

The arithmetic, and it does not go the way vendors imply

Two years at a twenty-two-person brewery. Assume seven real desk users year-round, plus you would like the taproom's twelve to have access. Assume a per-person plan at $30/month. Assume a self-hosted setup on one server with a couple of capable graphics cards, bought outright.

Renting wins, and it is not close. That is the honest answer for a twenty-two-person brewery, and anyone telling you otherwise is selling servers. The picture only flips when the volume work above is large enough that per-use charges on a rented service climb past a few hundred dollars a month, or when you have someone on staff who genuinely wants to own the machine — and "wants to" matters more than "can," because the five hours a month is real and it lands on somebody who already has a job.

The scale where owning starts to make sense in this trade is usually a group operation: several brands under one roof, a contract-brewing side, or a distillery with a large barrel program and a real direct-to-consumer web store. If that is you, the open models existing at this quality is the news. If you are one taproom and 3,000 barrels, the news is simply that your negotiating position got better, because the free alternative is now credible enough to mention out loud when your vendor raises prices.

What you should not put on your own machine

Anything with a deadline and a signature. Your monthly TTB reports, your excise return, your label submissions. Not because a self-hosted model cannot help draft them, but because when the machine is down on the fourteenth you do not want the report to be down with it. Keep the compliance work on something with a support number.

Anything customer-facing in real time, at least at first. The taproom phone and the web chat need to answer reliably at nine on a Friday night, and reliability on your own hardware is a thing you earn over months, not a thing you get on day one.

And be honest about the person. If the answer to "who owns this box" is "the packaging lead, probably, he's good with computers," you have just given a critical dependency to someone who did not ask for it and who is on the line four days a week. That works until he takes a job at the brewery across town.

What to do this month

Do not buy hardware. Do two things instead. First, measure: for one month, log every job you would hand an assistant and roughly how big it was — number of descriptions, number of files read, number of questions asked. Most owners are surprised in both directions.

Second, get access into the hands of the people who are not at desks. A shared account on the taproom tablet and one on the packaging office computer will teach you more about where the value is than any amount of planning. Then, when your renewal comes up, you will know exactly how many seats you need and exactly how much volume you are doing — which is the only position from which the buy-versus-run question can actually be answered.

Frequently asked questions

Is a free model going to be noticeably worse for writing our beer descriptions?

For that job, no. Short marketing copy from a grain bill and a set of tasting notes is well within what open models handle now, and every line gets a human edit anyway. Where you may still notice a gap is long, careful reasoning over messy documents — reconciling a barrel program, or working through a multi-state shipping question. Keep the hard, rare jobs on the paid service and put the high-count, low-stakes jobs wherever is cheapest.

We only have one IT person and he is really our packaging lead. Can we do this?

You can, and plenty of breweries this size have. The question is whether you should. Five hours a month is the steady state; the first two months are more like fifteen. If that time comes out of packaging line hours during your Oktoberfest run-up, you have paid for the machine twice and not noticed.

Does running our own mean our recipes stay private?

It means they stay in your building, which matters if you contract-brew for other labels or have confidentiality language in a co-packing agreement. But read the terms of the paid services too — the business tiers generally commit to not training on your material. Privacy is a real reason to run your own; it is a contractual reason more often than a technical one.

What happens to our seats when we cut back to winter staffing?

Ask your vendor about monthly seat adjustment before you sign an annual agreement. Breweries, distilleries, and anyone with a patio have a headcount curve that annual per-seat contracts handle badly, and it is a negotiable point. Going into that conversation knowing a credible free alternative exists is worth more than any discount code.

One place seats never made sense is the phone. Your taproom line rings hardest at the exact hours every person you employ is behind the bar or on the floor — Friday evenings, the Saturday of a release, the week you announce a beer dinner. CallSphere builds voice and chat agents that answer those calls and web chats around the clock, book tours and private events, and hand real leads to your taproom manager. It is priced against the calls you are missing, not against a headcount that doubles every May.