The vendor-percentage slide audited with a three-question test — then the real barriers to adoption and Deloitte’s five steps, with data (not the model) as the true blocker.
August 2026 · Case 4 of 6
As you read — hold this question
A slide of enormous percentages. Who measured them, on how many cases, compared to what — and if AI is this good, why hasn’t it taken over?
1blog
Six jaw-dropping vendor stats — up to $1.3 trillion added to the economy — all trace back to a single 2024 vendor blog.
Every deck has the same slide: a wall of confident percentages. Put it under a brighter light and a habit worth keeping for life falls out — three questions for any AI statistic. Then the honest reason all this isn’t everywhere yet: a series of walls, and a consultant’s map through them.
The vendor slide, audited
The rest of the storyQuiz: the 3-question audit
Six big numbers, one small source
Here are the six you’re meant to believe: 65% fewer lost sales; >80% of logistics firms on AI/ML by 2030; 15% lower logistics costs; 20–50% fewer forecasting errors; $1.3 trillion added to the global economy by 2026; $41 billion saved a year by 2030. Impressive — and every one traces back to a single 2024 vendor blog. That isn’t evidence; it’s marketing.
Who measured it?
A vendor selling the tool — or an independent study?
On how many cases?
One hopeful pilot — or seventeen real ones?
Compared to what?
Better than nothing — or better than the old way?
Company-reported projections are useful signposts, not proof. The reading’s cost and lead-time counts are smaller, duller — and far sturdier. That’s what evidence actually looks like.
Why it isn’t everywhere yet
The textbook version
The walls, and Deloitte’s map through them
If it’s so good, why isn’t it everywhere? Because of real walls: training and running cost; system complexity across a global chain; AI-specific risks (bad data, over-trust, security and privacy); data trapped in silos and ageing systems that won’t talk; and people — trust, and resistance to change.
Deloitte’s five steps through them: 1) strategy & roadmap; 2) change management & value tracking; 3) tech-stack enablement; 4) data preparation (where NLP earns its keep); 5) use-case activation — business-led, self-funded, built with the people who’ll use it.
The other side — the real blocker
The barrier is rarely the model — it’s the data underneath. McKinsey (2026): more than two-thirds of high-performing companies say data is the primary obstacle to AI. And the honest flip side of “automate everything”: not every process should be. Sometimes a human in the loop isn’t a barrier to engineer away — it’s the feature that keeps you safe when the model is confidently wrong.
Carry this
Run the three questions on any AI stat; name one well-evidenced benefit (cost) and one weak one (sustainability); and remember the real barrier is data you don’t own, not the algorithm.
Quick recall — without looking back
Test yourself on this case
Question 1 of 3
What are the three questions to ask of any AI statistic?
Who measured it (a vendor, or an independent study)? On how many cases (one pilot, or many)? Compared to what (nothing, or the old way)? Two of the module’s six headline numbers are 2026 projections — already testable against reality.
Question 2 of 3
List Deloitte’s five steps to AI adoption.
Strategy & roadmap; change management & value tracking; tech-stack enablement; data preparation; use-case activation (business-led, built with the users).
Question 3 of 3
Name one well-evidenced benefit and one weak one, and the single biggest barrier to adoption.
Well-evidenced: cost savings (cited 15× in the reading). Weak: sustainability (once). The biggest barrier is data — messy, siloed, un-owned — not the model.