Last week, Accenture brought global industry analysts to Bangalore for an intense threeday session to assess whether its strategy is genuinely differentiated. For a company with the scale, breadth and depth of Accenture, I think this is a red herring. I went in with a different question: whether the world’s largest professional services firm is vulnerable to AI or adapting ahead of it. By the end of the week, it was clear that neither question really mattered. The real conversation was about what enterprises need to make AI work at scale – and why so few have it.  

Of the 12,000 generative AI and agentic initiatives Accenture has run since 2023, only 13% have scaled. Forrester’s latest State of AI survey tells a similar story: just 15% of AI decision‑makers say AI is positively impacting their firm’s earnings. I have written before about this AI productivity paradox: individual contributors are productive using AI tools, but the broader organisation fails to absorb them and generate meaningful value. 

The reason? Most operating models were never designed for agentic workflows. Organizations lack the capability to decompose processes and workflows to the level of granularity needed to decide where an agent belongs and where a human must remain. And the accumulated business logic of the firm – the knowledge locked in people’s heads, undocumented handoffs, and tribal conventions – remains tacit knowledge that not accessible and readable by machines. Until it is, agents will continue to be deployed into the wrong processes, automating the wrong tasks, and delivering pockets of local efficiency that feel impressive but never compound into enterpriselevel productivity. 

Accenture aptly calls the response to this problem “Reinvention Services”. It describes the integrated capability to redesign an enterprise’s operating model, workforce, and technology architecture at the same time – not sequentially, not in parallel workstreams that converge at a steering committee, but together, because these problems are interdependent. This is one of the reasons why Accenture reinvented its own operating model by merging strategy, technology and operations into integrated buyer-aligned units. 

The skill is the atomic unit of reinvention 

AI models are now abundant, improving, and increasingly interchangeable, and out-of-the-box capabilities of these models are ever-increasing. What is scarce is the human ingenuity and a capability to decompose work, redesign processes, and decide where intelligence should act autonomously and where human judgment must prevail. Until organisations invest as deliberately in those capabilities as they have in model access, productivity and value realization from AI will lag despite improvements in AI. 

Karalee Close, Global Lead for Talent & Organization, presented a system, internally dubbed “Talent Navigator”, that decomposes every process into tasks and every task into skills. It then classifies each skill across multiple dimensions, including human judgment, GenAI augmentation, and deterministic automation. This gives clarity to the process and capability owners to understand the current state and design the future state, controlling for current technology capability and skill availability; then generating the levers – training, automation, hiring – to close potential gaps. Accenture’s LearnVantage, backed by a $1 billion three-year investment in AI-enabled learning and reskilling capabilities, is one such lever: it moves workforce development beyond course completion by linking skill-building to assessments, workplace application, and measurable business outcomes. 

Forrester has argued this point consistently: the unit of analysis is not the job, nor the role – it is the skill. Most enterprise AI programmes start with available tools and work forward aiming to scale individual productivity gains across functions and departments. The ones that scale start with the work, decomposed to the skill level, and derive the technology investments and roadmap from what the decomposition reveals. 

The reinvention engine and the brain  

Accenture’s reinvention.ai platform organises its transformation capability around three motions: reinventing the work, reshaping the workforce, and redesigning the workbench. What underpins these is an Intelligent Digital Brain: the data ingestion, entity resolution, and knowledge architecture layer that makes organisational context machine-readable. Lan Guan, Accenture’s CAIO, gave the following example: at a pharmaceutical firm, decades of expert judgment – the decision rules, regulatory constraints, domain logic that seasoned professionals carry but rarely document – were systematically codified into structured artefacts that agents can reason over. The result was a production system that could surface in minutes evidence gaps and contradictions across drug development decisions that previously required weeks of senior expert review. 

The same architectural logic shows up in Forrester’s research on the agentic era, captured in the our upcoming research into the Cognitive Operating Model: architecture, workforce, and organisational design, bound together by codified organisational knowledge. The Reinvention Engine establishes this foundation inside the enterprise.  

People power the engine 

There is a strain of fatalism running through the AI conversation right now, much of it originating from the AI labs themselves, that treats large-scale workforce displacement as a near-certainty. The models will do what junior people do. Hiring will contract and the pyramid will become a diamond. I think that conclusion is wrong or at least, dangerously premature. The diffusion of this technology will take a long time and we live in a state of genuine uncertainty about how these capabilities will reshape work, and that uncertainty should make us skeptical of preordained outcomes, however well-credentialed the people predicting them.  

This fatalism risks becoming a self-fulfilling prophecy, though. If hiring managers believe AI replaces junior talent, they stop hiring juniors not because the technology proved it could, but because the narrative persuaded them it would. A hiring freeze driven by belief rather than evidence produces exactly the displacement it predicted. The workforce hollows out not because AI demanded it but because leaders accepted the prophecy and made decisions accordingly. 

Julie Sweet is fighting this narrative, and the scale at which she is doing so, as CEO of a 750,000+ employee firm, matters. 1- AI is a growth engine, not a cost play. Forrester’s own Accelerate Your AI Voyage research reinforces this, finding most enterprises still default to productivity use cases and struggle to translate AI investment into measurable impact. 2- Accenture is hiring more juniors, not fewer. She “strongly disagrees” with the diamond model, compressing time-to-productive of junior resources from six months to two weeks rather than eliminating the entry-level pipeline. And 3- Humans belong in the lead, not in the loop. The difference between people who decide what AI is for and people who check its outputs, between upskilling and slow-motion deskilling. Each of these positions is a deliberate intervention against a belief that, left unchallenged, produces the displacement it predicts.  

Why this matters beyond Accenture 

The AI Productivity Paradox is an organisational design problem — and solving it requires simultaneous transformation of workforce, architecture, and governance that very few enterprises can execute alone. That’s why I think professional services firms capable of integrated reinvention have never been more relevant. Not as implementers, but as partners that can hold all three dimensions together while the enterprise learns to operate differently. 

This is also why the engagement model matters as much as the methodology. The relationships that compound AI value will be the ones where trust and data access accrue over time — where transformation and operations are baked into the same engagement, and where the shift from time-and-materials to outcome-based pricing aligns incentives over the life of the relationship, not just at project close. The firms that earn that trust position themselves not as vendors but as co-owners of the outcome. 

Ultimately, the enterprises that scale AI will be the ones that build the skill foundation, adopt an integrated transformation methodology, orient toward growth rather than cost reduction, and sustain the governance to keep all three aligned as conditions change. That takes more work than deploying agents into operating models that were never designed to absorb them.