The research, quoted properly

73.5% of knowledge workers use AI often, but only for one-off tasks. Just 5.5% have it embedded in repeatable workflows.

Section AI Proficiency Report, 2025. 5,026 US knowledge workers. Section sells AI training, so this is a vendor survey with a real sample.

Most people need to cross the bridge from dabbling to habit, not another tool or course.

Workers whose managers require AI use score 1.5 times higher on proficiency. Role-specific training lifts scores 1.8 times.

Same survey, same caveat.

An installed routine moves people. Optional exploration does not. Our 90-day method exists to be that routine for firms with no one to enforce one.

Developers using AI on their own real projects believed they were 20% faster. Measured, they were 19% slower on that work.

METR, 2025. Sixteen experienced developers, one small study, narrow scope.

Feelings are not measurement. Every rebuild starts with a baseline, because without a before, every after is a story.

Executives estimated 4% of their staff use AI for a third of daily work. Employees themselves reported 13%.

McKinsey survey of C-suite executives and employees. Exact report title and sample size not yet confirmed on our side.

Leaders are roughly three times wrong about their own teams. Seeing your team clearly comes before any plan.

41% of workers received AI-generated work in the past month that looked finished and was not. Fixing it took nearly two hours each time, about $186 per worker per month.

BetterUp Labs and Stanford Social Media Lab, September 2025. 1,150 US full-time workers.

Machine output without a named human owner is where the junk gets through. Every judgment needs an owner.

40% of workers fear losing their job to AI, up from 28% two years ago. The fear is higher at firms furthest along (46%) than at firms barely starting (34%).

Mercer Global Talent Trends 2026. 12,000 workers and leaders.

A rollout without explanation reads as threat. Name what people gain in the same breath as what the machine takes.

42% of enterprises abandoned most of their AI initiatives in 2025, up from 17% the year before. The average firm scrapped almost half its pilots.

S&P Global Market Intelligence, 2025. Over 1,000 enterprises, North America and Europe.

Our read on that gap: no baseline, no redesigned workflow, no feedback loop, no named owner. All four are fixable by an owner who moves in order.

95% of purpose-built enterprise AI pilots never reach a measured, sustained payoff, even though general AI tools like ChatGPT succeed for most people using them personally.

MIT NANDA (Project NANDA, MIT Media Lab), The GenAI Divide: State of AI in Business 2025, July 2025. Preliminary findings, not peer-reviewed: 52 structured interviews, 153 survey leaders, review of 300+ public AI initiatives, over a 6-month window.

A pilot built without a clear bottleneck to fix is the kind that shows up in that 95%. Naming the bottleneck first is where our method starts.

Enterprises have put $30 to 40 billion into generative AI, and most of it has not shown a measured return yet.

MIT NANDA, The GenAI Divide: State of AI in Business 2025, July 2025. Same sample as above. The report calls this investment with no return so far, not waste: some of the spending may still pay off after its 6-month study window.

The size of the number matters less than what it tells a firm to check: whether its own AI spending is showing a measured return yet, not just activity.

In a McKinsey survey of almost 2,000 leaders, only about 6% say AI is moving their profit by 5% or more and calling it a real win.

McKinsey & Company (QuantumBlack), The State of AI in 2025, November 2025. 1,993 respondents, 105 countries, fielded June to July 2025. Self-reported profit impact, not audited.

The 6% who clear this bar show a measured profit gain, not just heavier AI use. That is the bar worth aiming at.

Companies that buy or partner for AI tools succeed about twice as often as companies that try to build the same tools in-house.

MIT NANDA, The GenAI Divide: State of AI in Business 2025, July 2025. Based on the report's interview sample of 52 organisations, self-reported and correlational: the report itself says this does not prove buying causes the higher success rate.

This is why we point most firms toward an existing tool for their first move, not a custom build from scratch.

Structured outside help is priced for big companies: fractional AI officers run $2,000 to $30,000 a month.

Range we have seen quoted in fractional AI officer job postings, 2025-2026. Pricing data, not results data.

A 30-person firm is priced out of the market's answer. That gap is who we build for.

Self-paced online courses complete at roughly 5 to 15%. Cohorts with dates and peers complete at several times that.

Platform-reported figures across tens of thousands of courses; exact numbers vary by platform.

Content does not change behaviour. Cadence and company do.

Sustained workplace coaching shows moderate to large measured effects on performance and self-direction.

Two independent peer-reviewed meta-analyses: Theeboom and colleagues 2014, Jones and colleagues 2016. Not AI-specific.

A recurring, personal, goal-tied relationship is the strongest-evidenced format there is. It is the shape our coach is built on.

In the studies above, none tracked AI behaviour change past a few weeks. We have not run a full literature search broad enough to say no such evidence exists anywhere, only that nothing we found shows it holding without ongoing structure.

The one randomised trial found measured four weeks. Course statistics cover single cohorts.

So we do not claim lasting change. We build the 90-day loop to be measured, and we publish what we find.