Research · 28 September 2026
Eighty percent say AI improved their individual productivity. Only thirty-seven percent see any impact in the P&L.
Most organisations are using AI. Very few have scaled it across the enterprise. A new paper sets out a maturity model and deployment approach for generative, agentic and traditional AI — built on the finding that deploying AI is an operating-model change, not a software rollout.
The gap between those two numbers is the subject of the paper.
Generative AI is not another software rollout. Agentic AI is not an extension of chatbot deployment. Both need an enterprise capability — strategy, portfolio, governance, data, platform, security, service management, workforce and measurable value — maturing together rather than one at a time.
So this is deliberately not a governance framework, an architecture, or a transformation plan. It is the spine those things hang from.
What is in it
- A six-phase maturity journey, from first controls to pervasive AI
- Twelve capability domains that must mature together
- A maturity matrix setting out the evidence you should be able to show before claiming to have moved a stage
- Detailed guidance per domain — from business cases to agent identity to decommissioning
- A practitioner toolkit: opportunity card, decision-rights matrix, operational-acceptance checklist, decommissioning checklist
It is a consolidation, not an invention. 112 sources, primary law and standards preferred, every dated claim checked against the source in September 2026.
The maturity map
Six phases and twelve capability domains, on one page.
One design principle runs through all 68 pages
Use probabilistic AI for judgment, generation and planning. Use deterministic mechanisms for authority, limits, invariants, accounting and shutdown.
That split is the same instinct behind our Agentic AI Control Catalogue, published earlier this month: the model decides, and something that is not a model enforces the boundary. The catalogue is the control detail; this paper is the operating model those controls sit inside.
Yours to use
The paper is published under CC BY 4.0 — yours to read, adapt and build on in your own work.
For information. Not a standard, certification scheme, legal opinion or completeness claim.
Enterprise AI · AI Governance · Agentic AI · AI Strategy
For how to compose and run the team doing the work — people and agents together — see Making Multi-Agent AI Work.
Enterprise AI at Scale · edition v8.1 · 68 pages · published 18 October 2026 · sha256 04789f130a627cb2d107f43723434cc4c08b7fee38f313a9f3467df23e367303