The transformation deck on your desk is a rerun — here's the 5-question test that catches it before you fund it

Somewhere in the second slide of the AI transformation deck, you felt it.

You have approved this deck before. It said Agile then, or digital, or cloud, or — if you have been at this long enough — ERP. The fonts are better now. The demo is genuinely astonishing. But the shape of the plan underneath is one you have funded, watched drift, and quietly stopped asking about.

You remember what the last one cost. You do not remember a closing report, because there wasn't one. Transformations do not end in this industry; they fade. The steering committee meets monthly, then quarterly, then the invites stop, and eighteen months later the programme name only comes up as a joke with a wince in it.

You have not said any of this in the meeting. The person presenting is good. The technology is real. And you do not want to be the executive who blocked the future twice.

This piece is for the thing you are not saying.

The unprecedented technology, and the extremely precedented plan

Everyone frames AI transformation as unprecedented, and on the technology they are right. The models are new. The economics are new. The capability curve is genuinely unlike anything you rolled out before.

The technology being unprecedented is not the interesting part.

The rollout plan is the interesting part, because the rollout plan is not new at all. Look at the beats: buy the tools first, run pilots staffed by volunteers, publish an adoption dashboard, stand up a champions network, book the benefits into next year's plan. That is not an AI playbook. That is the transformation playbook — the same five beats your Agile rollout ran, and your digital programme before it. The nouns changed. The plan didn't.

And unlike the technology, the plan has a track record. It has been measured.

The number audit

Here is where I am supposed to hit you with the statistic, so let's do it properly — which means auditing it in the same breath.

BCG's 2020 research found 70% of digital transformations fall short of their objectives¹. Broken down: 30% fully succeeded, 44% created some value but missed their targets, 26% created little or nothing. McKinsey's 2018 global survey set a stricter bar — transformations that improved performance and sustained the improvement — and found only 16% cleared it².

Now the audit. Nobody has measured "the 70%" cleanly. The BCG band includes partial wins. The McKinsey bar is strict enough to fail programmes most boards would call adequate. The two firms measured different things, and both sell transformation services — the people quoting the disease also invoice for the cure. If you take one thing from this section, take the reflex, not the number: the last transformation was sold to you on numbers with exactly this shape, and nobody went back to check them against what actually happened at your company.

That going-back is the closing report. It never got written. Which means the most expensive research your organisation ever ran — a multi-year, eight-figure experiment on how change actually behaves in your specific culture — produced findings that are currently stored nowhere except in the scar tissue of the people who lived it.

Meanwhile, the rerun is already filming. S&P Global surveyed over 1,000 enterprises and found 42% abandoned most of their AI initiatives in 2025 — up from 17% a year earlier³. On average, 46% of projects died between proof of concept and adoption⁴. The failure rate didn't creep. It more than doubled in a year, while the decks got more confident.

What the rerun always skips

Every rollout runs three stories at once.

The first story is procurement. It is measurable, it finishes on time, and it is genuinely completed — the licences exist, the security review passed, the vendor is live. The second story is usage. Seats active, prompts per employee, adoption curves. It is the story dashboards are built from, and it is a systematically misleading proxy: a thousand chatbot-polished emails read as adoption; the one team that rebuilt its claims process end-to-end reads as three users.

The third story is the redesign of actual work — which workflow, exactly, now runs differently, owned by whom, with what cycle time. It is the only story that pays for the programme. It is also the only one of the three that never gets a slide, because it is slow, unglamorous, and impossible to buy.

The transformation playbook — Agile then, AI now — is a machine for finishing stories one and two while story three quietly starves. That is not a cynical read. It is just what the beats optimise for: tools land (story one), volunteers demo (story two), and the redesign of work is left to emerge on its own. It doesn't. It didn't last time either.

The defences, named and shot

You will hear three defences of the deck on your desk. You may be rehearsing one yourself.

"This time the technology is real." It was real last time. Agile was not a fake idea; iterative delivery genuinely beats waterfall for most software. Digital was not a fake idea. The technology being real was never the failure mode. The failure mode was a plan that deployed tools into an undesigned gap.

"We'll go slower and learn from last time." Slower is not different. A rerun at half speed is still the same film.

"We hired a Head of AI." The last programme had a transformation office. A named owner of the programme is not a named owner of a workflow — and story three is made entirely of workflows.

None of these defences names the thing that actually failed. So none of them prevents it failing again.

Your scar tissue is the asset

Here is the reframe that makes the burned executive the most valuable person in the room, instead of the blocker.

You already paid for the answer key. Your organisation's last transformation established — at enormous cost — exactly three things: which parts of your culture absorb change and which parts perform it; what happens to freed-up time when nobody redirects it (it dissolves — invisibly, without a trace in any metric); and which layer of management actually converts intent into changed work (the middle one, which is precisely the layer the rollout bypassed).

The rerun asks you to ignore all three findings and run the experiment again. The alternative is cheap: write the closing report now, late, in the form of a test — and run every new deck through it.

Take the deck on your desk. Any deck. The one with the astonishing demo. This takes one sitting.

The Rerun Test

Five questions. Each one checks a beat where the last transformation actually broke — not where the post-mortem slides said it broke. Score each 0 (no), 1 (partially), or 2 (yes, with a name attached).

1 · The Closing Report. Can anyone in the room state, in numbers, what the last transformation was promised and what it delivered? Not impressions — the original deck's targets against what happened. If the answer is no, you are about to run an experiment you already ran, minus its findings.

2 · The Named Workflow. Does the plan name a specific end-to-end workflow that will run differently by a specific date — with the person who owns its cycle time? "We will empower teams to reimagine work" is not a name. "Claims triage, ninety minutes to nine, owned by Priya, by March" is a name. This is the single strongest predictor in the test. A deck that cannot name one workflow has planned stories one and two only.

3 · The Enthusiast Filter. Do the pilot results come from volunteers? Pilots staffed by enthusiasts measure enthusiasm. The last rollout's pilots glowed for the same reason and for the same reason it didn't scale: tool access was never the constraint — the sceptical middle of the organisation was. A pass is a pilot that includes the team that didn't want it.

4 · The Hour Trail. When the tools free up time — and they will — where does the plan say those hours go? Name the mechanism, not the aspiration. Freed time with no destination becomes invisible slack, and it becomes it fast enough that the productivity case silently dies while the dashboard stays green. A pass names who redirects the hours and into what.

5 · The Kill Criteria. Does the plan state, in advance, what result would cause the programme to stop, shrink, or change course — and when that check happens? The last transformation had no kill criteria, which is why it faded instead of ending. A programme that cannot fail by its own definition cannot succeed by it either. A pass is a date and a number.

Scoring. 8–10: new picture — fund it; this deck learned from the last one. 4–7: remake — the material is real but the plan needs its missing names before money moves; send it back for the specific zeros. 0–3: rerun — you have seen this film, and you know the ending; blocking it is not blocking the future, it is declining to pay twice for the same lesson.

One boundary, honestly: the test cannot tell you whether AI will transform your industry. It can only tell you whether this deck would transform your company — which is a different question, and the only one you are actually being asked to fund.

The closing report, at last

Classify the deck before you fund the deck. The investment motion follows the classification, not the other way around.

By the end of one sitting you will have a score, a word — rerun, remake, or new picture — and, for the first time since the last programme faded, a written record that your organisation's most expensive lessons were finally allowed to grade something.

Start with question two. If the deck cannot name one workflow, you do not need the other four questions this week.

References

Footnotes

1. Boston Consulting Group. (2020). Flipping the odds of digital transformation success. https://www.bcg.com/publications/2020/increasing-odds-of-success-in-digital-transformation ↩

2. McKinsey & Company. (2018). Unlocking success in digital transformations. https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/unlocking-success-in-digital-transformations ↩

3. CIO Dive. (2025, March 14). AI project failure rates are on the rise: Report. https://www.ciodive.com/news/AI-project-fail-data-SPGlobal/742590/ ↩

4. CIO Dive. (2025, March 14). AI project failure rates are on the rise: Report. https://www.ciodive.com/news/AI-project-fail-data-SPGlobal/742590/ ↩