The Threshold Test
Five questions that separate AI value from AI activity
88% of organisations use AI in at least one function. 39% report any EBIT contribution at all — and most of those put it below five percent of EBIT. Filter for material impact and McKinsey lands on 6% as high performers. BCG, measuring the same gap on its own scale, lands on 5%.
Adoption is easy to count. Value needs a threshold — and the threshold has to be set before the work starts. That is the whole method. Five questions, answered in writing, dated, before the first pilot is funded.
📄 Download the Threshold Test diagnostic sheet (PDF) — one page. Print it, fill it in, put it in front of the portfolio review.
This test is the opening diagnostic from my book Platform Economies — the sheet is labelled PE-01 for a reason.
The five questions
01 — What is the threshold, in writing, before you start?
McKinsey’s own line for a high performer is more than five percent of EBIT attributed to AI. Pick your number. Write it down. Date it.
Catches: success declared against a bar invented after the results arrived.
02 — Whose P&L?
Enterprise level, or the unit where the system runs. The 39% figure is enterprise level — and the gap between unit-level wins and group-level P&L is where most business cases quietly fail.
Catches: a unit-level win that never reaches the group P&L.
03 — What does the money replace?
Headcount not hired. Contracts not renewed. Cycle time taken out of a process someone owns. If nothing gets smaller, nothing was saved.
Catches: productivity gains with no line item that moves.
04 — What breaks when it scales?
Every system that reaches production inherits a data boundary, an audit obligation, and a failure mode with a named owner. Where no one owns the failure, the system does not leave the pilot — and the value never arrives.
Catches: the pilot that cannot pass its own risk review.
05 — What is the kill condition?
Which pilots stop, on what evidence, by what date. Decided in advance, by someone with the authority to stop them.
Catches: a portfolio that only ever grows.
Scoring
- Five in writing today — you are measuring value.
- Three or four — you have a business case and no verdict.
- Two or fewer — you are counting activity.
Why a threshold, and not a business case
A business case argues forward. A threshold judges backward. The 88/39/6 pattern exists because organisations happily do the first and never build the second — adoption gets announced, value gets asserted, and nobody is ever on record having said what number would count.
Five questions, one page, ten minutes per pilot. The expensive part is not answering them. It is discovering, a year in, that nobody did.
📚 Where this comes from. The Threshold Test is diagnostic sheet PE-01 from Platform Economies: Survival in the Era of Platform Compression — the book on what AI does to business models when it stops being a feature and starts being a rule shift. If the score above made you uncomfortable, that is the book working: platformeconomies.com.
Related tool on this site: the Platform Compression Scorecard — score your organisation on the five compression signals from Part Six of the book. Background reading: What Is a Platform Economy?
About the author
Mohammed Brückner is a Senior Enterprise Architect and Digital Transformation Strategist based in Munich, working across large-scale European and international organisations. He writes about platform strategy, API operations, and what AI does to both. His other books are listed on the publications page.
Figures: McKinsey, The State of AI, 2025. BCG, The Widening AI Value Gap, September 2025. Full citations on the diagnostic sheet.