witkeepDiscover Witkeep ↗The uncomfortable numbers
In March 2025, McKinsey published its latest global survey on the state of AI. The finding was clear: more than 80% of companies see no tangible impact on EBIT from generative AI. Only 1% of executives describe their deployment as “mature”. According to LXT, using Gartner’s model (2025), 80% of organisations are still below the transformational level of AI maturity.
These figures sit alongside another fact: 78% of organisations use AI in at least one function (McKinsey, 2025). Adoption is widespread. Impact is not keeping pace. This paradox needs an explanation.
of executives describe their generative AI deployment as “mature” — despite significant investment and widespread adoption. (McKinsey Global Survey, March 2025)
The real reason: workflows, not tools
McKinsey analysed 25 organisational attributes to identify what correlates with generative AI’s impact on EBIT. The result was clear: the strongest factor is neither budget, model choice nor the technical team. It is workflow redesign.
Organisations that fundamentally rethink their processes around AI tools see a measurable impact. Those that adopt tools without changing workflows accumulate subscriptions, not results. Yet only 21% of organisations have redesigned their workflows so far. The remaining 79% fall into the second group.
Adopting an AI tool is easy. Reorganising processes around it, standardising usage and retaining methods that work is what most organisations have yet to do. This is precisely what separates the 20% that see ROI from the 80% that do not.
What the successful 20% do
Organisations that derive real value from AI share several characteristics. They have dedicated AI KPIs — tracking impact, not just usage. They involve leadership in governance: according to McKinsey, CEO oversight of AI is one of the attributes most closely correlated with EBIT impact in large organisations. Above all, they treat AI workflows as organisational assets : documented, versioned and shared.
The 80% seeing no ROI
- Individual adoption, without standardisation
- Each employee uses their own prompts
- No dedicated AI KPIs
- Tools added to unchanged workflows
- Absent or reactive governance
The 20% seeing ROI
- Workflows redesigned around AI
- Centralised prompts and methods
- Impact KPIs measured regularly
- Leadership involved in governance
- AI expertise treated as a collective asset
The link to knowledge drain
There is a direct link between the lack of AI ROI and knowledge drain. When AI use remains individual — each employee with their own prompts, methods and account — the organisation builds dependency, not shared capability. And every departure means starting again.
Prompts optimised over weeks of iteration and AI workflows refined by experts remain in personal histories, beyond the organisation’s reach. AI ROI is not just a workflow problem. It is also an institutional memory problem applied to AI. See Marc’s story for a concrete example, or the six challenges for the full framework.
What this means in practice
Moving from the 80% to the 20% achieving real ROI requires four things: usage governance (who uses what, with which data), standardised workflows (best practices become shared standards), retained prompts and methods (kept within the organisation rather than personal histories), and independence from models (the ability to change provider without rebuilding everything).
The question is not “which AI should we choose?” It is “how should we organise the business around AI so the value created belongs to the organisation — not just to the individuals who created it or the provider hosting their history?”