“It’s fine when you’ve got 10 people, but not when you’ve got 200 people. And when you’ve got 200 people and you look at the cost, it just blows it out of the water.”
That was an executive at a 3,500-person care provider last week, explaining why she had ruled out a clinical AI tool her staff actually liked. Not risk. Not accuracy. Not change resistance. Arithmetic.
A few minutes later she did the sum the AI industry never does out loud: “it’s only a dollar. Well, okay, it’s a dollar, but we produce 1,000 of those, then it’s 1,000 bucks, and then it’s 12,000 bucks.” Her organisation runs aged care, disability services, retirement living and at-home care. The majority of her workforce are cooks, cleaners, gardeners and home care workers. Eighty of them hold a paid Copilot licence. Everybody else is on the free tier.
Why this matters now
The market has settled on an explanation for the frontline AI gap, and it is the comfortable one. BCG reports that the 42% of frontline workers who are regular AI users save around a workday a week, while adoption overall lags badly; Forbes ran a piece in August under the headline that AI continues to bypass half of the workforce. The prescribed fix is almost always enablement — better training, better change management, better champions, better perception.
That is not what I am hearing in the rooms. The frontline is not behind because it is resistant. It is behind because per-seat AI pricing was built around a desk worker producing desk-worker margin, and it does not survive contact with three and a half thousand people on shift. A tool that pencils beautifully for ten users is unaffordable at two hundred, and structurally impossible at three thousand. Nobody in that organisation ran a change management programme badly. They ran a spreadsheet correctly.
What organisations in this segment are getting wrong
They read their own low frontline adoption as weak demand, and go hunting for enthusiasm. Meanwhile the biggest recoverable prize sits precisely where the licence maths fails.
In that same conversation: the residential admissions team spends two hours writing up notes for every three hours of meetings. Two hours in three. Nobody had raised it as a problem, and this is the part worth sitting with — in her words, staff “don’t see a problem. They can’t look at things and go, I’m doing this manually, let’s look for a technical solution. They just keep doing it.” She has to go to them and offer to build it. Then they say yes.
Put those two facts side by side. The people with the largest measurable block of recoverable time are the people your licensing model cannot reach, and they are also the people least likely to ask. Waiting for demand from that workforce is a strategy that guarantees you never find the value. In the same week, a different client told me his adoption philosophy was that “the idea must come from the ground up.” In a knowledge-worker business that is sound. In a frontline business it is a decision to find nothing.
Three questions before you buy another seat
What proportion of your workforce is structurally excluded from your current AI licensing model — not unwilling, excluded? Count the people who could never be economically licensed at today’s per-seat price. If that number is over half your headcount, your AI strategy currently applies to a minority of your business.
Where is your largest block of unreported manual effort, and how would you know? Do not survey for it — nobody who is coping will report it. Go and time one process end to end.
Have you ever priced the alternative to a seat? Task-based, consumption-based or shared-workflow models put AI against a process rather than a person. That is a different commercial conversation, and it is the one your provider should be broking on your behalf rather than passing through a per-user price list.
And if you are a provider, as I am
This is the uncomfortable one, because there is no licence to resell at the end of it. The per-seat model is the easiest thing in our industry to sell and the easiest to renew, and for the frontline-heavy half of the mid-market it does not work. A vendor cannot say that; their pricing is the product. A reseller whose margin is a percentage of seats will not say it either.
Which leaves it to whoever is willing to be paid for the outcome rather than the headcount. If your proposal to a 3,000-person care provider, logistics operator or clinic network opens with a per-user price, you have not made them an offer. You have told them, politely, that your product is for the eighty people in head office.
The frontline was never the laggard. It was never invited.