Most AI transformation ends in a deck

There's a version of AI transformation that almost every organization has now been sold. It runs about twelve weeks. There are interviews, a maturity assessment, a workshop or two, and at the end there is a document. The document is usually good. It identifies real opportunities, ranks them by effort and impact, and recommends a roadmap.

Then it goes in a drive folder, and nothing happens.

This isn't a knock on consultants. The document is genuinely the thing they were hired to produce, and producing it well is hard. The problem is structural: a recommendation is only worth what the organization can execute, and the organizations most eager to buy AI transformation are precisely the ones with no capacity to execute one.

The numbers describe a stall, not a failure

The industry has adopted AI almost universally and converted it into results almost nowhere.

McKinsey's November 2025 State of AI — 1,993 respondents across 105 nations — found that 88% of organizations now use AI in at least one business function, up from 78% a year earlier.¹ But nearly two-thirds have not yet begun scaling it across the enterprise at all.² On money, the gap is starker: just 39% report any EBIT impact at the enterprise level, and most of those put it under 5%.³

Deloitte's 2026 survey of 3,235 leaders across 24 countries found the same shape from a different angle. Only 25% of organizations had moved 40% or more of their AI pilots into production.⁴

And ISACA's 2026 poll of more than 3,400 professionals found that only 22% say AI's return on investment has met or exceeded expectations.⁵

You will sometimes see this summarized as "70% of digital transformations fail." That statistic does not hold up. The earliest version of it was labelled an unscientific estimate by its own authors in 1993, and one of them publicly retracted it two years later, writing that it had been widely misrepresented. Every later restatement traces back to that same unsourced claim rather than to a study.

The real failure mode is less dramatic and more useful to understand. Transformation programs don't usually explode. They stall — somewhere between the recommendation and the running software.

Why the stall happens where it does

The gap between a good recommendation and a working system is engineering. Not strategy, not change management — engineering. Someone has to build the thing, connect it to the systems you already run, secure it, test it, deploy it, and keep it alive at 2am.

Large enterprises absorb that with internal platform teams. Most organizations have no such option, and the adoption data shows exactly where the line falls. The Federal Reserve reports that about 18% of US firms had adopted AI as of year-end 2025.⁶ Yet 78% of the labor force works at a firm that has.⁷ Both numbers are correct. AI adoption looks near-universal only when you count employees; counted by company, most of the economy hasn't started.

So the advisory model quietly assumes a capability that most of its buyers don't have. You get told what to build by people who won't build it, and then you are expected to go find someone who will.

What we do instead

Every phase of our AI Transformation program ships working software that you own outright. Source code in your repository, running on your infrastructure, under your accounts. Not a pilot in someone else's sandbox. Not a license.

It opens with a fixed-fee assessment, and that assessment behaves differently from the usual one in two specific ways. First, it is yours to keep — the inventory, the workflow map, the costed plan. If you take it to another firm or build it yourself, that works, and it is a legitimate outcome. Second, it is credited against the build if you move forward, so you are not paying twice to start.

We map how your operation actually runs — the spreadsheets, the handoffs, the approval that lives in one person's inbox — and identify which workflows are worth automating and which honestly aren't. Then we build them, one at a time, the same way we would build any production application.

Governance is in the engagement, not a separate invoice

Here is the part most people underestimate.

ISACA found that only 38% of organizations have a formal, comprehensive AI policy, up from 28% a year earlier.⁸ A quarter have no active policy at all.⁹ That reads like a policy problem. It isn't.

The more revealing number is what happens where policies already exist. Two-thirds of office professionals report having used AI at work even though they believed doing so was not permitted under company policy.¹⁰ The rule was there. It did not bind.

A policy nobody follows isn't governance. It's a document — which is where we came in.

Real governance is built into the system: what the software is permitted to do, which data it can reach, what gets logged, who reviews what, and what happens when it is wrong. That is an engineering property, not a paragraph. It is also why agent governance lags so badly — only 21% of companies planning to deploy agentic AI report having a mature governance model for it.¹¹ You cannot write enforceable rules for a system you are not building.

That is why we don't sell it separately from the build.

Who this is for

Organizations under real pressure to adopt AI that aren't staffed to deliver it safely — small and mid-sized businesses, and the agencies and nonprofits that serve the public. If you have a technology department that could execute a roadmap on its own, you probably don't need us for this.

If you don't, the first step is small and concrete: a paid assessment that tells you what's worth automating, what isn't, and what it costs. You keep it either way.

---

Every figure above traces to a primary source, linked below and re-verified in CI. Three statistics from an earlier draft were cut because the pages they were attributed to did not contain them — including a widely repeated "70% of digital transformations fail," which no study supports.

Footnotes

1. McKinsey & Company, The state of AI in 2025: Agents, innovation, and transformation — mckinsey.com ↩

2. McKinsey & Company, The state of AI in 2025 — mckinsey.com ↩

3. McKinsey & Company, The state of AI in 2025 — mckinsey.com ↩

4. Deloitte, State of AI in the Enterprise 2026: The Untapped Edge — deloitte.com ↩

5. ISACA, 2026 AI Pulse Poll — isaca.org ↩

6. Jeffrey S. Allen, Monitoring AI Adoption in the US Economy, FEDS Notes, Board of Governors of the Federal Reserve System — federalreserve.gov ↩

7. Allen, Monitoring AI Adoption in the US Economy, FEDS Notes — federalreserve.gov ↩

8. ISACA, 2026 AI Pulse Poll — isaca.org ↩

9. ISACA, 2026 AI Pulse Poll — isaca.org ↩

10. PagerDuty, Shadow AI in the Workplace survey — pagerduty.com ↩

11. Deloitte, State of AI in the Enterprise 2026 — deloitte.com ↩