Adoption Is the Wrong North Star

The invoice that renews anyway

Somewhere this month a fifty-person company is renewing its AI subscriptions. Nobody will question the invoice, because everybody agrees AI matters. Luke Pierce described what the usage data usually shows, writing on 24 August 2026: four people who live in the tools, ten who paste things in occasionally, and thirty-six seats untouched since the week they were bought. Nothing about how that company operates has changed since they added an AI subscription. The invoice renews anyway ^S1.

The thirty-six are not failing. Their job was never "use AI." Their job is getting projects out the door, keeping clients happy, closing the books. A workshop on prompting asks them to pick up a second profession on top of the first, then sends them back to a pile that did not shrink. The chat window becomes a tab they do not open, and that is a sensible use of a working week.

The four who live in it are being sensible too. They were already curious, they got faster at the work in front of them. Good for them. In the future, cracks will likely emerge, but that's for another day Single-player AI and the multiplayer business: the tools get used, the business does not change ^S2.

The surveys make the scene a pattern, not a story about one firm. McKinsey's Global Survey on AI, published 12 March 2025 from fieldwork in July 2024, found 78% of organisations use AI in at least one business function. More than 80% of respondents still reported no tangible impact on enterprise-level EBIT from generative AI ^S3. PwC's 29th Global CEO Survey, of 4,454 chief executives across 95 countries, found 56% had seen no significant financial benefit from AI to date. Only 12% reported both higher revenue and lower costs ^S4.

Pierce's question follows. The old one was how to get the team to adopt AI. That question treats unused seats as a motivation problem. They are a goal problem.

Named momentum, not more logins

Adoption is a spend line. It counts logins, licences, and whether someone opened the tab. It does not count whether the business moved or value was generated.

What the business needs is named momentum: a goal for the company that people can say out loud, a way to see whether this week advanced it, and the next piece of work actually done. We call that goal a North Star - the overall mission for the business, then for each part of it, kept live rather than filed after an entertaining but short-lived away-day ^S2. One sits at the top. Each department has one of its own, and it has to serve the company's, not compete with it. AI is an instrument on some of the work beneath it. AI use is not the goal.

McKinsey tested 25 organisational attributes against reported EBIT impact from generative AI. Workflow redesign had the biggest effect. Only 21% of respondents whose organisations use generative AI said they had fundamentally redesigned at least some workflows ^S3. Among twelve adoption and scaling practices, the one with the most bottom-line impact was tracking well-defined measures for the solutions themselves, and fewer than one in five organisations were doing it ^S3.

Those are not findings about models. They are findings about aim. Gartner, predicting on 25 June 2025 that over 40% of agentic AI projects will be cancelled by the end of 2027, named unclear business value as one of the three causes ^S5. A seat that nobody can attach to a named goal is that cause, bought in advance.

The spine under the principle is very simple. Start with a North Star and drill down to objective to milestone to task. Every piece of work, human or machine, traces up. If it cannot, it waits. That is goal alignment as an architectural property, not a quarterly reminder ^S2.

Choosing the North Star is the hard part, and it is not a naming exercise. Most businesses discover during the choosing that two directors have been optimising for different things for years, quite reasonably, because nobody ever wrote down which one won. A department can also name a clean North Star that does not serve the company's. That is two businesses sharing a payroll. That argument is the value. At Red First we uncover it during the Red Brief. Seats in an AI subscription were never going to settle it.

The Friday test

The Friday test is unglamorous and boring. It is this.

Before the next renewal, or the next workshop, run three questions on the work the seats were bought to touch.

If every AI seat vanished on Friday, which jobs would still finish on Monday? Those jobs were never adopted. They were already the operation.

Of the jobs that would stall, which of those move a North Star - the company's, or the department's that serves it? A faster weekly report that nobody uses to decide anything is activity. Cash collected on the terms the client agreed, or quotes that convert without eroding margin, is movement.

Of the jobs that move the goal, which actually need a model, and which need a written procedure, a deterministic step, or a person? Anthropic's advice from December 2024 still holds: start with the simplest system that works, which often means no agent at all ^S6. Extraction from an invoice, a draft update pulled from project data, a queue already sorted before anyone looks - those are the uses that earn their place, because the team meets an outcome, not a prompt box ^S1. A chat window handed to the warehouse coordinator is another tool. The Friday test is how you tell the two apart.

Held properly, the check runs in both directions. An agent that speeds up quoting but erodes margin per delivery fails, however clever it is, if the North Star is profitable growth. An agent that improves a dashboard by moving work somewhere it stops being counted fails too. The model can be right and the aim still wrong.

The operational method for choosing the North Star, ranking the work, and resolving the argument between directors stays inside certification. The principle is public. If a proposed AI use cannot name the task and the milestone it serves, it waits, whether the seat is already paid for or not.

Cancel the seats. Does the goal notice?

Time. The saving that matters is not ten minutes on a task the four power users already enjoy. It is the work that stops, because it never served the goal, and the workshops that stop, because adoption was never the job.

Control. More of it, not less. Unused seats hide in an invoice everyone is proud of. Named momentum shows up as a milestone that moved, or one that did not. You can see the week.

Risk. Aligned systems fail visibly, because a failure shows up as a goal that did not advance. Unaligned systems fail quietly for a year and then surface as an incident, or as a renewal nobody can defend except by saying AI matters.

Value. A business that can state its North Star, show the trend, and explain what moved it is a more valuable business than one that can only show a seat count. That is true independently of the number itself, and independently of whether a model sat on the path.

For a genuine solo operator this is the wrong frame. Single-player AI is the right tool for a single player, and not to say so would simply be dishonest ^S2. Their adoption is the operation. A 10-to-250-person company is a multiplayer game, and seat counts are how it pretends otherwise.

RedOS is built so every piece of work is checked against the relevant North Star - the company's first, then the department's - and a Certified Consultant is the person in the room who will make the owner say which goal wins at the top. It is heavier than signing into a chat window, and that is the trade-off. The company does not need the team to adopt AI. It needs tracked, executed movement toward a named goal, and AI only where that movement requires it.

If you cancelled the seats on Friday and the goal would not notice, the seats were never the strategy.