--- title: Organisational AI Competency Framework source: https://steadman.ai/newsletters/david/ai-maturity-diagnostic.html published: 2026-04-05 summary: A five-phase organisational AI maturity framework with a 5x4 matrix, 20-question diagnostic quiz, and result prescriptions across Mindset, Strategy, Building, and Accountability. --- # Where is your organisation on the AI journey? A five-phase map of what good looks like when a business absorbs AI — from the first person drafting an email with a chatbot to an organisation that works better, quicker, and happier because of it. Most AI frameworks measure individuals. That is useful for hiring. But the question readers keep asking us is different: *where is my organisation, and what should the next phase actually look like?* This framework answers that. It uses four components — Mindset, Strategy, Building, Accountability — and hangs them on the five phases we see organisations move through as AI stops being a novelty and starts to reshape what the business is. It measures organisational maturity, not which tools people use — for that, see the [AI Usage Spectrum](/newsletters/david/ai-usage-spectrum.html). --- ## Four things to look at, in every phase A phase is not a checkbox. At every stage, ask the same four questions. Weakness in any one of them is what stops organisations moving to the next. ### i. Mindset Do people treat AI as a tool, a threat, or a colleague? What does leadership model? Is experimentation rewarded or punished? ### ii. Strategy Is there a deliberate view of where AI belongs in the business — or a hundred tactical decisions nobody has joined up? ### iii. Building Can the organisation actually ship things with AI? Prompts, workflows, agents, products — what gets built, how fast, by whom? ### iv. Accountability Who owns the output when AI did most of the work? How is quality checked? Where is the judgment that stops bad things shipping? --- ## The framework, at a glance Rows are the five phases of the journey. Columns are the four things to look at inside every phase. Read across a row to understand a phase; read down a column to understand how one concern evolves as an organisation matures. ### Phases x Components Matrix | Phase | Mindset | Strategy | Building | Accountability | |-------|---------|----------|----------|----------------| | **1. Individual Productivity** — *The hundred small things* | **Curiosity permitted.** Senior people visibly use the tools. Nobody hides it. | **Tools bought and blessed.** Chosen stack, paid licences, data rules stated. | **Personal fluency.** People can name three tasks they now do differently. | **Self-check.** People verify their own outputs; mistakes surfaced, not buried. | | **2. Team Standards** — *Conversation to architecture* | **Sharing is default.** Building a prompt for the team is a recognised contribution. | **Reuse as a metric.** Highest-leverage tasks identified. Someone owns the library. | **Plural shared assets.** Custom GPTs, prompt packs, versioned — not pasted in Slack. | **Asset owners.** Each asset has a maintainer. Bad outputs trace back to a prompt. | | **3. Process Orchestration** — *"AI does the process"* | **Think in processes.** "The agent did it" is an acceptable answer — with receipts. | **Budget for plumbing.** Processes chosen on payoff. Real systems, not demos. | **End-to-end workflows.** Named flows in production. Handoffs are designed, not improvised. | **Named owners + evals.** Every workflow has a human owner. Failures caught within the hour. | | **4. Role And Team Redesign** — *Different shape, not same people faster* | **Honest about change.** Redeployment over quiet exits. Judgment becomes a job description. | **Structure on paper.** Headcount plans written around what AI now does. | **Flatter, wider.** Spans of control widened. Old roles gone, work still done. | **Rebuilt pipeline.** Fewer people own more outcomes. Junior learning path deliberately rebuilt. | | **5. New Revenue** — *Do different things, not the same things faster* | **Willing to cannibalise.** "What are we selling?" is a live question at the top table. | **New economics.** Pricing reflects software margins, not hourly rates. | **Shipped product.** Live AI-native revenue line. No humans behind the curtain. | **Own the apology.** Quality/safety owned by named people. Firm — not vendor — answers for errors. | *Read across: what this phase looks like. Read down: how this concern matures. Slope matters more than position.* --- ## Part I — Transforming the existing business (Extraction) ### Phase 1: Individual Productivity **Tagline:** The hundred small things. Everyone starts here. People use AI for their own work — drafting, summarising, researching, thinking out loud. It is episodic, personal, and largely invisible to colleagues. The organisation benefits by aggregation, not design. **Mindset:** * Curiosity is permitted; nobody is punished for trying * Senior people visibly use the tools themselves * "I used AI for this" is said out loud, not hidden **Strategy:** * A chosen toolset, licensed and paid for * Clear guidance on what data can go in * No illusion that tool access equals transformation **Building:** * Individuals can describe three tasks they now do differently * Prompts are still conversational, not reusable * Time-saved stories circulate informally **Accountability:** * People check their own outputs before using them * Sensitive data handling has been spelt out once, loudly * Mistakes are surfaced, not buried **Snapshot signal:** Self-reported usage above 60%. License utilisation matches license count. **Slope signal:** People are talking about different tools this month than last month. Vocabulary is drifting forward. **Stuck at Phase 1 looks like:** High adoption, no aggregation. Everyone is faster in private; the business looks the same in public. Productivity gains get absorbed — people do better work, but the organisation doesn't capture the value. --- ### Phase 2: Team Standards **Tagline:** From conversation to architecture. Best practice gets encoded. The individual who figured out a better way now has a custom GPT, a prompt library, or a shared instruction set that lets twenty colleagues work the same way. The shift is from private craft to shared assets. **Mindset:** * "I built this for the team" is a recognised contribution * Sharing prompts is default, not generous * Copying what works is faster than inventing again **Strategy:** * Someone owns the shared asset library * Tasks with highest leverage are identified, not all tasks * Reuse is a metric, not an accident **Building:** * Custom GPTs, projects, or agents exist in plural * Prompt libraries are versioned, not pasted into Slack * New joiners inherit a working toolkit on day one **Accountability:** * Assets have an owner who checks they still work * There is a review rhythm for the shared library * Bad outputs can be traced back to the prompt that produced them **Snapshot signal:** At least one shared asset per team, with measurable reuse. **Slope signal:** The library is growing and — critically — things are being retired from it. **Stuck at Phase 2 looks like:** The "plz fix" law firm partner. Standardised inputs, unchanged process around them. --- ### Phase 3: Process Orchestration **Tagline:** From "AI helps me" to "AI does the process." Whole workflows run end to end with AI in the loop — multi-step, multi-tool, sometimes multi-agent. Humans shift from doing the work to specifying, reviewing, and intervening. This is where most organisations discover that their data, their systems, and their decision rights were never written down. **Mindset:** * People think in processes, not tasks * "The agent did it" is an acceptable answer — with the receipts * Leaders stop romanticising the manual version **Strategy:** * Processes are chosen because the payoff justifies the rebuild * Integration with real systems, not demos * Budget for plumbing, not just for licences **Building:** * Named end-to-end workflows running in production * Handoffs between humans and agents are designed, not improvised * Versioning, monitoring, and rollback exist **Accountability:** * Every workflow has a named human owner * Evaluation runs on every change, not only at launch * When the workflow fails, someone knows within the hour **Snapshot signal:** Three or more multi-step workflows running reliably with agents in them. **Slope signal:** Workflows are being retired because something better replaced them — not because they broke. **Stuck at Phase 3 looks like:** Impressive demos, PowerPoint diagrams, nothing in production. Or production systems nobody trusts. --- ### Phase 4: Role And Team Redesign **Tagline:** Different shape, not same people faster. The organisation chart changes. Layers flatten, roles merge or disappear, new roles appear. This is the phase most organisations flinch at, because it is the first one where the answer is not "more of our people, faster" but "fewer people, differently arranged." It is also where the real value sits. **Mindset:** * Leadership is honest about what is changing and why * People are redeployed, not quietly managed out * "Judgment" becomes a real job description, not a euphemism **Strategy:** * Headcount plans are written around what AI now does * The junior pipeline is redesigned, not abandoned * Career paths make sense in a smaller, flatter shape **Building:** * Teams are reorganised around workflows, not departments * Spans of control have materially widened * Roles that existed a year ago don't exist anymore — and the work is still done **Accountability:** * Fewer people own more outcomes, and know which ones * Quality has held or improved through the transition * How the junior generation learns judgment has been deliberately rebuilt **Snapshot signal:** Headcount-to-output ratio has moved in a way the board can see. **Slope signal:** New roles are being invented faster than old ones are being deleted. **Stuck at Phase 4 looks like:** Efficient teams, hollowed-out junior ranks, nobody being trained into the judgment the seniors are now selling. --- ## Part II — Changing what the business offers (Expansion) ### Phase 5: New Revenue **Tagline:** Do different things, not the same things faster. The organisation uses AI to make things it could not have made before — or to sell what it already makes in a fundamentally different shape. Services become platforms. Bespoke becomes scaled. A professional-services firm starts shipping software; a retailer starts designing. Revenue lines appear that have no equivalent in last year's accounts. **Mindset:** * Leadership is willing to cannibalise existing revenue * "What are we actually selling?" is an active question, not a settled one * Product thinking sits next to service thinking at the top table **Strategy:** * A clear view of which offerings translate into products and which do not * Pricing and packaging reflect the new economics, not the old hourly rate * Distribution has been thought about as carefully as the build **Building:** * At least one product is live and generating revenue from AI-native capability * Customers are using it without the firm doing the work by hand behind the curtain * The product is improving on a cadence, not frozen at launch **Accountability:** * Product quality, safety, and support are owned by named people with real authority * The firm can answer questions about how the model was trained and what it does * When the product is wrong, the firm — not the vendor — owns the apology **Snapshot signal:** A revenue line from an AI-native product that didn't exist twelve months ago. **Slope signal:** The product pipeline is getting longer, and the firm's self-description is changing. **Stuck at Phase 5 looks like:** A "product" that is really a consulting engagement wearing a software hat. Or a launch that is still, eighteen months later, the only launch. --- ## Slope beats snapshot. The most useful thing any maturity framework can do is assess *trajectory*, not current state. The same is true for organisations. A business sitting solidly at Phase 2 and experimenting into Phase 3 is a better bet than one nominally at Phase 4 but which stopped moving eight months ago. When you use this framework, mark where you are — then mark where you were six months ago. The gap between those two marks is the signal. Stillness at any phase is a warning; motion, even from a low base, is the thing worth backing. --- ## How to use this Three entry points, depending on what you actually want to do with it. **Diagnose.** For each phase, score your organisation 0-3 on each of the four components. Weakness in any column tells you where the next unlock is. Don't average the scores; the lowest one is the one that's holding you back. **Calibrate.** Compare your scores to where you were six months ago. The delta matters more than the absolute. Flat scores at any phase mean you're stalling. Moving scores at any phase mean you're on the journey, even if you're not where you'd like to be. **Decide.** Do not try to move on every phase at once. Pick the next phase, pick the weakest component within it, and invest there. The organisations that move fastest are the ones that refuse to spread their effort. --- ## The diagnostic quiz Twenty questions, about eight minutes. Four per phase — one for each component. Answer honestly for the organisation, not for yourself. Pick the statement that best matches your organisation today. If two feel true, pick the lower one — be harder on yourself than you'd like. The result is only useful if the input is honest. ### Phase 1: Individual Productivity **Q1 (Mindset): How do senior leaders in your organisation talk about using AI themselves?** * (0) They don't use it, and it shows * (1) They say it's important; unclear if they use it personally * (2) Most of them use it and say so openly * (3) They demo their own use in team meetings and it's contagious **Q2 (Strategy): Which of these best describes your AI tooling situation?** * (0) Ad hoc; people bring their own, sensitive data goes God-knows-where * (1) A tool has been licensed, but there's no guidance on use * (2) Chosen stack, paid licences, and clear rules on what data can go in * (3) All of the above, plus visible tracking of who actually uses what **Q3 (Building): If you asked a random employee 'what are three things AI has changed in how you work?', you'd most likely get...** * (0) A blank stare or 'I don't really use it' * (1) One generic example ('I use it for emails sometimes') * (2) Three specific examples with genuine craft behind them * (3) Three examples plus what they tried and abandoned **Q4 (Accountability): When an employee uses AI to produce something and it turns out to be wrong, what happens?** * (0) Nobody notices, or the person is blamed without learning * (1) There's awkward silence and informal 'be careful' mutterings * (2) The person checks their own outputs; mistakes are surfaced openly * (3) It's treated as valuable feedback and the lesson circulates ### Phase 2: Team Standards **Q5 (Mindset): When someone on your team builds a useful prompt, custom GPT, or instruction set, what's the norm?** * (0) They keep it to themselves — no reason to share * (1) They share if asked, but nobody knows what exists * (2) Sharing is the default; teams expect it * (3) Building reusable assets is a recognised contribution in reviews **Q6 (Strategy): Does your organisation know which tasks have the highest leverage for AI standardisation?** * (0) No — it's random what gets built and what doesn't * (1) Somebody has a rough view, not written down * (2) Yes — a deliberate list exists and drives investment * (3) Yes, and it's reviewed quarterly with a named owner **Q7 (Building): How are your shared AI assets (custom GPTs, prompts, templates) managed?** * (0) They aren't — things get pasted into Slack and lost * (1) There's a folder somewhere, patchy maintenance * (2) Plural assets exist, versioned, and new joiners inherit them * (3) All of the above, plus things get retired when they stop working **Q8 (Accountability): If an output from a shared prompt causes a problem, can you trace it back?** * (0) No — we wouldn't know which prompt produced what * (1) In theory, if we dug into it * (2) Yes — assets have owners and a review rhythm * (3) Yes, and the review rhythm has caught real issues recently ### Phase 3: Process Orchestration **Q9 (Mindset): How do your people talk about work they want to get done?** * (0) As tasks — 'I need to do X' * (1) A mix; occasionally someone frames it as a process * (2) As processes — 'how could this whole thing run?' * (3) And 'the agent did it' is a normal answer, with receipts **Q10 (Strategy): When you think about AI investment, what's the budget shape?** * (0) All licences, no plumbing * (1) Licences plus some experimentation money * (2) Explicit budget for integration and workflow build * (3) Process rebuilds are chosen on expected payoff and measured after **Q11 (Building): How many multi-step AI workflows are running end-to-end in production today?** * (0) Zero * (1) One or two pilots people don't fully trust * (2) Three or more, reliably, with real business value * (3) Several, with versioning, monitoring, and rollback in place **Q12 (Accountability): If a production AI workflow fails silently tomorrow morning, when do you find out?** * (0) When a customer or colleague complains, if ever * (1) Within a day or two, manually * (2) Within the hour — someone is on the hook for it * (3) Within minutes, via evals that run on every change ### Phase 4: Role And Team Redesign **Q13 (Mindset): Is leadership being honest with people about what AI is changing in the organisation?** * (0) No — it's 'just a tool to help you' messaging * (1) Partially — the direction is hinted at but not spelt out * (2) Yes — the direction of travel is explicit * (3) Yes, and redeployment plans are visible, not euphemistic **Q14 (Strategy): Have headcount and hiring plans been rewritten to reflect what AI now does?** * (0) No — we're planning as if nothing changed * (1) Some tweaks at the edges * (2) Yes — the plan looks materially different * (3) Yes, and career paths make sense in the new shape **Q15 (Building): Compared to a year ago, how has team structure changed?** * (0) Same people, same org chart * (1) Some reshuffling; nothing structural * (2) Spans of control have visibly widened; some roles have gone * (3) Teams are organised around workflows, not departments **Q16 (Accountability): How is the junior pipeline — how juniors learn judgment — being handled?** * (0) It isn't — we just hire fewer juniors * (1) We worry about it but haven't acted * (2) We've deliberately rebuilt how juniors learn judgment * (3) And it's working — quality is holding through the transition ### Phase 5: New Revenue **Q17 (Mindset): Is leadership willing to cannibalise existing revenue to launch something AI-native?** * (0) No — protecting the current book is the priority * (1) In conversation, not in decisions * (2) Yes — decisions have been made that hurt in the short term * (3) Yes, and product thinking sits next to service thinking at the top table **Q18 (Strategy): Have you rethought pricing and packaging around AI economics?** * (0) No — we still sell time, materials, or seats the old way * (1) We've talked about it * (2) Yes — pricing reflects the new unit economics * (3) Yes, and distribution has been thought about as carefully as the build **Q19 (Building): Do you have an AI-native product in market — one that didn't exist a year ago and isn't humans in a trench coat?** * (0) No * (1) Something that looks like one but is actually manual behind the curtain * (2) Yes — live, generating revenue, customers use it without us doing the work * (3) Yes, and the pipeline behind it is getting longer **Q20 (Accountability): If your AI product produces a harmful or embarrassing output tomorrow, who owns the response?** * (0) Unclear — probably legal or 'the vendor' * (1) Someone, but authority and process aren't well defined * (2) Named owners with real authority; the firm answers, not the vendor * (3) All of that, plus we can explain how the model was trained and what it does --- ## Scoring Each question scores 0-3. Each phase has four questions, giving a phase score out of 12. Each component appears once per phase, giving a component score out of 15. **Your position on the journey:** The highest phase where the score is 7 or above *and* every prior phase also scored 7 or above. This enforces the journey structure: an organisation cannot be "at Phase 5" if Phase 2 is broken. **Phase-score chart:** Five bars, each scored out of 12, showing where the organisation is strong and where it falls off. **Weakest-component diagnosis:** The component with the lowest cross-phase sum — the systemic under-investment — along with a tailored "next move" prescription. --- ## Result verdicts by phase **Pre-start (no phase reaches 7):** You're at the starting line. People haven't really begun, or the few who have are doing it alone. That isn't a failure — it's where everyone begins. The risk is staying here by default while the market moves. Invest in Phase 1 basics: visible leadership use, a blessed toolset, data rules, and a culture where trying things is safe. **Phase 1 — Individual Productivity:** People are using AI for their own tasks and the good ones are genuinely faster. The business, however, looks the same from the outside. That's fine as a starting point, dangerous as a destination. The next job is Phase 2 (Team Standards): turning private craft into shared assets so the gains compound across teams, not just individuals. **Phase 2 — Team Standards:** Best practice is being encoded. Prompts, custom GPTs, and shared instructions mean twenty people can now work the way your best person figured out. This is the first phase where the organisation — not just individuals — starts to look different. The next move is Phase 3: wiring those assets together into whole workflows that run end to end. **Phase 3 — Process Orchestration:** Real work is running end-to-end with AI in the loop. You're past the demo stage. This is also where most organisations stall, because Phase 4 asks the uncomfortable question they've been avoiding: if the process runs itself, what's the right shape of the team around it? That's your next move. **Phase 4 — Role And Team Redesign:** Your org chart has materially changed in response to what AI now does. Very few organisations are genuinely here. The trap at this phase is efficiency without invention: a smaller, faster version of the same business. Phase 5 asks whether you can use this moment to make something you couldn't have made before. **Phase 5 — New Revenue:** You have live, AI-native revenue that didn't exist a year ago, owned honestly, priced for the new economics. You're in the company of very few organisations. The risk now is complacency — and the drift back down the phases as the shiny first product ages. Keep the pipeline moving and keep asking what you're selling. --- ## Next-move prescriptions by weakest component **Mindset is your weakest column.** No amount of tooling fixes this — it's a leadership job. Pick three senior people and have them publicly use AI in meetings this month. Reward people who share, not hoard. **Strategy is your weakest column.** You're moving tactically without a plan. Write down the three highest-leverage processes you want AI in. Name an owner. Ignore everything else until those three work. **Building is your weakest column.** You have intent but not delivery. Find the one person who's actually shipped an AI workflow, protect their time, and have them build one more with a colleague looking over their shoulder. **Accountability is your weakest column.** This is the quietest but most dangerous gap — it's where reputational damage lives. Before building anything else, make sure every AI-touched thing has a named human owner and a way to catch it failing. --- ## Related tools Use the [AI Value Map](https://steadman.ai/newsletters/david/ai-value-map.html) to see where the value sits and how much you're capturing. Read [From What's True to What to Do](https://steadman.ai/newsletters/david/ai-from-whats-true-to-what-to-do.html) for the strategic argument behind the five phases. ## Re-use The quiz runs entirely in the browser. Submissions never leave the device. Free to use, cite, and adapt. Please credit Steadman. If you build on it, we'd like to hear about it — david@steadman.ai.