--- title: "Flywheel Locator — AI Maturity Test" url: https://mkelango.com/diagnostics/flywheel-locator/ description: "The Flywheel Locator reads your AI capability across four stages plus DRIVE and CORE — scored out of 30, benchmark 13. Six questions, about three minutes." site: mkelango.com author: M. K. Elango updated: 2026-08-26 evidence_standard: "Every figure carries a grade [A] verified / [B] claimed / [C] estimate and an as-at date. See https://mkelango.com/evidence/" --- Free · no email to start From The AI Flywheel # The Flywheel Locator Mark your four stages and two layers. Leave with one sentence: we are strongest at ___ and our weakest part is ___. - **Length:** 6 questions · about 3 minutes - **Scored out of:** 30 - **Benchmark:** 13 What it reads Where your AI capability actually sits on the powertrain — and which part is holding the wheel. From [The AI Flywheel](https://mkelango.com/books/ai-flywheel/) · ENGINE In short Updated 2026-08-26 The Flywheel Locator is a free six-question instrument that places an organisation's AI capability on four stages — Foundation, Generative, Agentic and Physical — plus DRIVE, the hub, and CORE, the rim. It scores out of 30 against a benchmark of 13, graded [C] as at 26 August 2026, and returns one sentence naming your strongest and weakest part. Why this one ## Own the road. Rent the car. Foundation, Generative, Agentic, Physical. Plus DRIVE — the hub — and CORE — the rim. Six readings, one sentence. The sentence is the deliverable. If you cannot say where you are strongest and where you are weakest in one line, you cannot brief a board, and you certainly cannot sequence ninety days of work. The book behind it The AI Flywheel Compounding advantage — because implementation discipline is the only moat left when the model is rented. Read the argument [The AI Flywheel](https://mkelango.com/books/ai-flywheel/) Where this gets fixed AI Flywheel: The 90-Day Start One workflow into production, measured honestly, with a named human owner. The programme [AI Flywheel: The 90-Day Start](https://mkelango.com/programs/ai-flywheel/) Method ## The benchmark, and how it was built. A score is only as good as the thing it is compared against. So the benchmark carries a grade and a method, exactly like every other figure on this site. The benchmark 13 /30 [C: Mean self-reported reading across organisations with a named AI budget. estimate from workshop sampling, as at 26 August 2026. As at 26 August 2026] Mean self-reported reading across organisations with a named AI budget. Grade [C] — estimate from workshop sampling, as at 26 August 2026. [A] Verified Traced to a primary source — a filing, an audited statement, a contract, a dataset held by us — and the source is named. [B] Claimed Reported by a credible party, including by us, but not independently traceable to a primary document. [C] Estimate Modelled, sampled or directionally derived. Method stated. Treat as a shape, not a number. [The Evidence Log](https://mkelango.com/evidence/) What we do with your result The score is computed in your browser. Nothing is transmitted unless you ask for the detailed report, and in that case only the score, the trait breakdown and your email address are stored. Your result is never published, sold, shared or attached to your name in any material. [Privacy](https://mkelango.com/privacy/) The questions ## All 6, published in full. Nothing is hidden behind the interaction. If you would rather read the instrument than run it — or run it on paper, in a room, with other people — here it is. **01 **Foundation AI** — How much of your operational data is clean, connected and queryable?** - 1 Almost none - 2 One domain - 3 Two or three domains - 4 Most domains, with known gaps - 5 A governed layer everything reads from **02 **Generative AI** — Are any generative use cases producing measured value in production?** - 1 No, pilots only - 2 One, value unmeasured - 3 One, value measured - 4 Several, measured - 5 Several, measured, and funding the next stage **03 **Agentic AI** — Do you run agents that take actions, not just produce text?** - 1 No - 2 Prototypes only - 3 One in production, tightly scoped - 4 Several in production with human gates - 5 Agents are an operating model with named owners **04 **Physical AI** — Does operational data from production systems train anything physical or embedded?** Answer 1 without hesitation if this is not your sector. It is not a deficiency. - 1 Not applicable or not started - 2 Exploring - 3 A pilot exists - 4 One production system - 5 A closed loop, operating data to physical system **05 **DRIVE · the hub** — Before a build starts, is success written down and is there a data gate?** - 1 Neither - 2 Sometimes, informally - 3 Success is written; no data gate - 4 Both, on the important builds - 5 Both, always, and a build has been stopped by the gate **06 **CORE · the rim** — Is governance treated as an unlock or as a brake?** - 1 It is a brake and everyone routes around it - 2 A brake, formally complied with - 3 Neutral — a checklist - 4 An enabler on some workstreams - 5 The unlock. Governance is why we can ship to regulated customers Scored out of 30. Published benchmark 13. Method and grade above. The bands ## What each range means. Published before you take the test, so the reading cannot be tuned to flatter afterwards. 0–11 The wheel is not turning There are parts, but no powertrain — one stage’s output is not another stage’s input. That is not a failure of ambition; it is a sequencing problem, and it is the most fixable state on this instrument. 12–19 Turning, slipping Something compounds and something else leaks. Typically a strong generative layer sitting on a weak foundation, which caps how far the agentic stage can go. Name the weakest part and give it the next ninety days. 20–25 Compounding The wheel turns and each stage is feeding the next. At this reading the binding constraint usually moves from technology to governance — which is the CORE argument, and it is good news. 26–30 The road is yours A rare reading. The engine is model-invariant, which means the next model release is a fuel swap rather than a roadmap reset. Re-score with a sceptic and then go and standardise the win. FAQ ## Questions about The Flywheel Locator Answered plainly, with the real figures. If something here is wrong, [tell us and it goes in the log](https://mkelango.com/evidence/). ### What are the four stages of the AI Flywheel? Foundation AI, Generative AI, Agentic AI and Physical AI — the four stages of the powertrain. Two further readings sit around them: DRIVE, the hub, which asks whether success is written down and a data gate exists before a build starts; and CORE, the rim, which asks whether governance is treated as an unlock or as a brake. ### How do I know whether my AI investment is actually compounding? The wheel is turning when one stage's output is the next stage's input. The published bands read: 0–11, the wheel is not turning, because parts exist but no powertrain does; 12–19, turning but slipping, typically a strong generative layer on a weak foundation; 20–25, compounding, where the binding constraint moves from technology to governance; 26–30, a rare reading. ### Physical AI does not apply to my sector. Does that penalise my score? The instrument tells you to answer one without hesitation if physical or embedded systems are not your sector, and states plainly that this is not a deficiency. The reading out of 30 is a locator, not a league table — its deliverable is one sentence naming where you are strongest and where you are weakest. ### What is the difference between the Flywheel Locator and the AQ Score? The Flywheel Locator reads AI capability at the ENGINE altitude — what compounds over ninety days to five years — and scores out of 30 against a benchmark of 13. The AQ Score reads organisational architecture at the ORGANISATION altitude, three to five years, across twelve traits out of 60 against a benchmark of 27. Different instruments, different books. ### What happens after the diagnostic? The next ninety days belong to the weakest part; everything else waits. That is the programme AI Flywheel: The 90-Day Start — one workflow into production, measured honestly, with a named human owner — priced at ₹12L – ₹75L and banded by workflow complexity and whether data instrumentation is in scope. Starts are rolling rather than cohort-based. The other instruments ## Five more, one per altitude. [All six](https://mkelango.com/diagnostics/) Free The AQ Score Where your organisation sits on the twelve properties that let a machine improve a million-fold in a decade. 12 questions · about 4 minutes [The AQ Score](https://mkelango.com/diagnostics/aq-score/) Free Do You Actually Have a Moonshot? Whether the thing at the top of your ladder is an objective you could be held to, or a sentence that sounds good in a deck. 7 questions · about 3 minutes [Do You Actually Have a Moonshot?](https://mkelango.com/diagnostics/moonshot-test/) Free The Coherence Matrix, lite Whether the things your group holds build anything the objective needs — or merely make money. 5 questions · about 3 minutes [The Coherence Matrix, lite](https://mkelango.com/diagnostics/coherence-matrix/)