Three of the world's most influential strategy firms ran different surveys, with different samples and different analytical frames, and arrived at the same uncomfortable conclusion in 2025. McKinsey, Bain, and BCG each found that the primary determinant of enterprise AI value is not the model, the vendor, or the technology stack. It is whether the organisation redesigns its workflows around AI capabilities — or simply bolts AI onto the processes it already had.
That is not a nuance. It is the difference between capturing structural value and quietly wondering why a year of AI spend has not moved the operating numbers. For the DACH Mittelstand, where budgets are finite and the temptation to "buy a copilot and call it a strategy" is strong, it is the single most consequential finding of the year.
McKinsey: redesign is the variable that dominates
In its March 2025 report, The State of AI: How Organizations Are Rewiring to Capture Value — a global survey of roughly 1,500 respondents across more than 100 nations — McKinsey isolates a cohort it calls AI high performers: organisations that attribute at least 5% of EBIT to their AI use and report significant value from it. That cohort is small, about 6% of respondents. When McKinsey tests what separates them from everyone else, one variable contributes more than almost any other: fundamental workflow redesign.
High performers are roughly 2.8 times more likely to have fundamentally redesigned their workflows — 55% of them report doing so, against around 20% of the rest. McKinsey's own framing is blunter than any consultant's usual register: capturing value from AI is "20% algorithms and 80% organisational rewiring." The model is the easy part. The work is everything around it.
The adoption data sharpens the point. Across all respondents, only about 21% say they have fundamentally redesigned at least some workflows as they deployed generative AI. The overwhelming majority have done the opposite: adopted the tools — chatbots, copilots, summarisation engines, code assistants — and dropped them into process architectures designed for humans working without AI. The tools are faster. The processes are unchanged. The value capture is a fraction of what is available.
This is not surprising; it mirrors every prior technology wave. When spreadsheets arrived, most firms replicated paper ledgers in electronic form, and only the ones that rebuilt their financial workflows — scenario models, automated reconciliation, real-time reporting — captured the order-of-magnitude gains. When email arrived, most firms replicated paper memos, and only the ones that collapsed approval chains and went asynchronous became structurally faster. AI is on the same curve. The technology is adopted broadly; the redesign that unlocks it is adopted narrowly. The scaling gap is, at root, a workflow gap.
Bain: the model is not the determinant
Bain's Technology Report 2025 states the case with unusual directness for a strategy firm: the biggest determinant of AI transformation success is not the sophistication of the models but the quality of workflow redesign and the cleanup of the surrounding data and application environment. When a firm whose business is advising on technology tells its clients that model selection is secondary to process architecture, that is worth pausing on.
Bain distils its findings into five critical actions, and two of them are governance and redesign in plain language. Action three is to "redesign entire workflows, not siloed activities or use cases." The wording is deliberate — not optimise, not enhance, but redesign entire workflows. Action two is to "charge general managers with meeting these targets, not the CIO or CTO." The logic is that workflow redesign is a business decision, not an IT project. A CIO can procure AI tools; only a general manager can decide how a function operates. Frame AI transformation as a technology initiative and you get tool procurement; frame it as an operating-model change and you get redesign. The ownership structure quietly determines the outcome — a point that lands especially hard in Mittelstand organisations where "AI" is often delegated wholesale to IT.
The economics back the framing. Bain reports that AI leaders have moved from pilots to profit by scaling AI across core workflows, with such efforts associated with EBITDA gains in the range of 10% to 25%. The mechanism matters more than the headline number: point solutions — a chatbot here, a document classifier there — produce incremental, measurable gains, while end-to-end workflow deployment, where AI carries a whole process with humans at defined decision points, is what shows up in operating margin. A mid-market firm will not see hyperscaler-scale returns, but the direction of the evidence is unambiguous: depth of redesign, not breadth of tool rollout, is where the economics live.
BCG: the value sits in the core, not the periphery
BCG comes at the same conclusion from a different direction. Instead of contrasting leaders with laggards, it asks where AI value actually concentrates — and finds that around 70% of it sits in core business functions: R&D, sales and marketing, operations and supply chain, and pricing. These are not the administrative corners where AI tidies up email and meeting notes. They are the functions that directly produce revenue, margin, and competitive position.
BCG's "future-built" companies — roughly 5% of the firms it studied — are defined by redesign, not adoption. They do not merely sprinkle AI tools across functions; they systematically rebuild how those functions work, changing how R&D generates and tests ideas, how sales qualifies and closes, how supply chains sense and respond, how pricing adapts. BCG reports that these firms expect to see about twice the revenue uplift and roughly 40% greater cost reductions than laggards in the areas where they apply AI. The distinction it draws is the same one McKinsey and Bain draw in different words: between an organisation that uses AI and one that is built around it.
That 70% concentration also explains why tool-level deployment underwhelms. If most of the value lives in core functions, and those functions are defined by complex, multi-step, cross-functional workflows, then deploying AI at the individual-task level — summarise this, draft that, classify the other — works the periphery and leaves the core untouched. A sales team using AI to write prospecting emails captures a thin slice. A sales function redesigned around AI-assisted scoring, qualification, and pipeline management captures the rest.
The anti-pattern all three describe
Strip away the different vocabularies and the three firms are describing one recurring failure mode. An enterprise procures AI tools, drops them into existing processes, measures adoption — how many people are using the licences — and reports that as progress. The workflows those tools sit inside never change: the same people do the same work in the same sequence with the same decision points, now with an assistant available at a few steps. That is what McKinsey's 80%-is-rewiring framing implies most firms are skipping; it is what Bain calls point solutions rather than workflow redesign; it is what BCG separates from being built around AI.
The pattern persists because it is organisationally comfortable. Real redesign changes how people work, which triggers resistance; redefines roles, which triggers anxiety; demands new governance, which demands executive attention; and forces cross-functional coordination, which means breaking silos. Buying a tool and adding it to an existing process requires none of that. It is easier, faster, and produces visible activity. It simply does not produce material value. This is the automation versus augmentation question at enterprise scale: automation swaps a human step for an AI step inside an unchanged process, while augmentation rebuilds the process around what AI and humans each do well. The evidence is consistent — redesign produces multiples of what insertion does.
What redesign actually looks like
Workflow redesign is concrete, and it runs in a sequence most organisations truncate. It begins by mapping the current workflow as it genuinely operates — not the tidy version in the process documentation, but the real thing, with its workarounds, exceptions, and informal handoffs. Process mining is the discipline that surfaces that reality rather than the org chart's fiction of it.
Next comes the question of where AI actually changes the economics of the work: which steps are high-volume and pattern-dense enough to automate, which need human judgment but benefit from AI-generated recommendations, and which are genuinely novel and should stay human-driven. The six dimensions of the AI Operating System give that analysis structure. Only then does redesign proper begin — and this is the step organisations skip. It means changing the sequence, the decision points, the escalation paths, the roles, and the metrics. A redesigned claims process does not look like the old one with an AI step bolted on; it looks fundamentally different, with AI handling classification, routing, and first-pass assessment as one automated flow while human adjusters concentrate on complex cases, customer contact, and quality oversight.
The final move is governance built for the redesigned flow, not inherited from the old one: who monitors the AI's decisions, which thresholds trigger human review, how exceptions are handled, what happens when model accuracy drifts. For German mid-market firms this is not optional polish — under the EU AI Act, AI used in areas such as recruitment, creditworthiness, or other high-risk contexts carries explicit human-oversight and risk-management obligations, so the governance layer is part of the workflow, not an afterthought bolted on once the system is live. The decision architecture framework works through these questions systematically.
The methodology alignment
The AI Operating System methodology rests on the same principle the three firms reached independently. It does not start with technology selection; it starts with the workflow. Its six dimensions — strategy, data, technology, organisation, governance, and operations — are the dimensions along which a workflow gets redesigned. And the Three Levels of AI integration define how deep that redesign goes: Level 1 reworks individual workflows, Level 2 reworks entire functions, Level 3 reworks the operating model itself.
This is not alignment claimed after the fact. The methodology was built out of DACH Mittelstand engagements in which the same pattern surfaced again and again: the clients who redesigned workflows captured multiples of the value of the clients who merely deployed tools. What McKinsey, Bain, and BCG have now validated across thousands of respondents is what hands-on work in the Mittelstand already showed — the workflow, not the model, is the unit of AI value.
If your organisation is running AI tools but has not redesigned the workflows they sit inside, the evidence from McKinsey, Bain, and BCG points the same way: you are capturing a fraction of what is available. A Fit Call starts with your actual workflows and pinpoints where redesign creates the most value — before a single technology decision is made.
References: McKinsey & Company, "The State of AI: How Organizations Are Rewiring to Capture Value," March 2025 (mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-how-organizations-are-rewiring-to-capture-value); Bain & Company, "Technology Report 2025," 2025 (bain.com/insights/state-of-the-art-of-agentic-ai-transformation-technology-report-2025); BCG, "AI Radar 2025: From Potential to Profit," January 2025 (bcg.com/publications/2025/closing-the-ai-impact-gap); BCG, "The Widening AI Value Gap" / "Build for the Future 2025," September 2025 (bcg.com/press/30september2025-ai-leaders-outpace-laggards-revenue-growth-cost-savings); EU Regulation 2024/1689 (Artificial Intelligence Act), Articles 9 and 14 (eur-lex.europa.eu/eli/reg/2024/1689/oj).