Five Firms Built Five AI Frameworks. None of Them Built a Ruler.

Put BCG's AI framework next to Gartner's. Then put Bain's next to McKinsey's. Then put NIST's next to all four. Don't read them one at a time, the way they were designed to be read, in separate decks from separate firms with separate cover pages. Overlay them. Do that and you'll find five institutions that don't talk to each other converging, independently, on the same shape. You'll also find that none of them will tell you the one thing you actually need before any of the rest of it is useful: which part of that shape you're standing on right now.

That's not a knock on the frameworks. Each one is solving a real problem. BCG's Deploy, Reshape, Invent gives you an ambition ladder. Gartner's Opportunity Radar gives you a targeting map. Bain's AI-era Operating Model tells you what has to be rebuilt. McKinsey's six dimensions give you a scorecard. NIST's Govern, Map, Measure, Manage gives you a risk cycle. Every one of them is correct about its own slice. The trouble starts when a leader tries to use one of them as if it were the whole answer, because none of them, on its own, tells you where you actually are before you try to use it.

BCG: Deploy, Reshape, Invent

BCG's ladder reads clean on a slide: automate tasks, then redesign workflows, then invent new business models. What the slide doesn't say is that each rung assumes a different level of organizational discipline underneath it, and the gap between rungs is measured in capability, not ambition. Deploy is something individuals can do with a licensed tool and a real task in front of them. Reshape requires a workflow that survives being run by a team, with validation that catches a bad step before it poisons the next one. Invent requires enough trust in that validation that multiple workstreams can run in parallel without someone babysitting every branch. The ladder describes the destination. It doesn't tell you whether the floor beneath your feet can hold the next rung's weight.

BCG: The 10-20-70 Ratio

BCG's other framework, the 10-20-70 split, says the algorithm is 10% of AI value, technology and data are 20%, and people and process are 70%. That ratio is directional, not a measurement anyone ran a study to produce, and it's worth saying so plainly. But the emphasis is exactly right, and it shows up again the moment you look at what actually gets verified as an organization matures. A signed team standards document. A human punch-out point proven to block bypass attempts. An adversarial agent's rejection log. None of those are model quality. All of them are process discipline wrapped around a model that was never going to be perfect on its own. The 70% is the mechanism that makes the AI safe to trust once it's chained into something bigger than a single prompt, built alongside the model instead of funded after the model already works.

Gartner: The Opportunity Radar

Gartner's Opportunity Radar solves a different problem, and it's worth being precise about which one. It plots where AI is worth pointing: everyday AI against game-changing AI, front-office against back-office, customer-facing against internal operations. That's a real and useful exercise. It is also, deliberately, silent on how ready you are to operate in the zone you've picked. An organization can correctly identify a game-changing, back-office opportunity and still lack the workflow discipline to run it without human hands on every step. The radar tells you where the terrain is interesting. It was never built to tell you whether you can survive walking on it yet, and treating it like a readiness assessment is where the confusion starts.

Bain: The AI-Era Operating Model

Bain's AI-era Operating Model names six things that need to change: structure and teams, management systems, leadership and culture, talent and roles, and business processes, all wrapped around what Bain calls "right work." Read that list next to the eight-stage model and a pattern jumps out. None of those six things need to exist for an individual to reach real competency with AI on their own task list. They become mandatory the instant work stops being a single person checking a single output and starts being a workflow that runs without someone watching every step. Bain is describing, in operating-model language, exactly what changes at the wall, the one stage in the eight-stage model that requires the team and IT together instead of one motivated person. The wheel is a close-up photograph of that single stage, the one where work stops being watched step by step and starts running as a chain.

McKinsey: Six Dimensions of AI Value

McKinsey's six dimensions, strategy, talent, data, technology, operating model, and adoption and scaling, read like six independent checkboxes. They aren't. Adoption and scaling is the stage axis itself, dressed up as a peer of the other five. The other five are the prerequisites that get checked at the wall: talent shows up as the technique someone taught a colleague at the earliest stage of individual competency, data shows up as a task library with real version history, technology shows up as a deterministic check running automatically on every output, operating model shows up as the standards document a team actually signed. Strategy is the only one that sits above the wall rather than inside it, because it's the portfolio decision about which workflows are worth building toward that wall in the first place. McKinsey built a real scorecard. It just left out which line item is the scale the other five get graded on.

NIST: Govern, Map, Measure, Manage

NIST's Govern, Map, Measure, Manage is the closest of the five to describing mechanism rather than ambition, and it maps almost one to one onto what happens once an organization actually clears the wall. Map is the honest self-assessment: which stage are you actually on, not which one you'd like to claim in a board deck. Govern is the adopted policy, the signed standards document, the rule that a human punch-out point cannot be bypassed. Measure is the deterministic validator and the quantified rubric that replaces "it seems better" with a number. Manage is the recovery loop: the case where a workflow branched wrong, got caught before it did damage downstream, and got corrected, logged, and fed back into the standard. NIST built the operating loop that runs inside a maturity model, the four verbs a team actually performs once it has enough discipline in place for the loop to mean anything.

Laid out this way, the five frameworks stop looking like five competing products and start looking like five witnesses describing the same event from five different rooms. BCG saw the ambition ladder and the value ratio. Gartner saw the targeting map. Bain saw the operating model that has to change. McKinsey saw the scorecard. NIST saw the governance loop. What none of them drew is the axis all five of their descriptions run along: the eight stages an organization actually has to climb, in order, each one building the muscle the next one assumes is already there.

The Math Behind the Wall

The wall exists because of math. Chain independent steps together and end-to-end success is per-step reliability raised to the power of the number of steps. A step that's right 95% of the time in isolation is genuinely good. Wire twenty of those steps into an unattended workflow and the chance of a clean run falls to roughly 36%. The model didn't get worse. The chain asked it to be right more times in a row than the math allows, and the first bad step contaminates everything downstream of it.

That curve is the mechanism behind all five outside frameworks at once. It's why Bain says the operating model has to change before a workflow can run unattended. It's why NIST's Measure and Manage exist as separate disciplines from Govern. It's why BCG's ratio tilts so hard toward people and process. Nobody had to coordinate on this. The math forces the same conclusion no matter which firm's language it shows up in.

Improving's Eight-Stage Maturity Model

The eight-stage model is what fills the gap all five frameworks leave open: a way to test, stage by stage, whether the organizational floor under an ambition is actually load-bearing. Stages one through three are individual competency, open to anyone with an approved tool and real work to point it at. Stage four is the wall, the first stage that requires the team and IT together, because deterministic validation and enforced review points only work if everyone has agreed to them. Stages five and six move from delegation to coordination, where the Stage 4 protections are enforced by adopted standards instead of one person's diligence. Stages seven and eight are named honestly as aspirational, because no one is running them in production today, and pretending otherwise would be the biggest counterfeit in the whole model.

None of the five outside frameworks are wrong. Each is a precise description of one part of the elephant, drawn by a firm that only had its hands on that one part. What they share, without ever saying so to each other, is the same underlying claim: the hard part of enterprise AI was never the model. What's missing from most boardrooms is the ruler, the honest answer to which stage of those five frameworks an organization has actually earned the right to attempt.

Sources: BCG (Deploy, Reshape, Invent; the 10-20-70 framework), Gartner (AI Opportunity Radar), Bain & Company (AI-era Operating Model), McKinsey & Company (Six Dimensions of AI Value Capture), and NIST (Govern, Map, Measure, Manage) public frameworks. Stage definitions from Improving's eight-stage AI maturity model.