How AI lands in an organisation.

What this is

Most AI rollouts fail at adoption, not at technology. Giving people access to a tool is procurement, and most organisations are decent at that. Getting the tool to actually do something useful, every day, across a team, is design work. That is where rollouts come apart.

I have run that design work. Twelve years of enterprise rollouts at Shell, Roche, Philips and ING, and two years turning Memortium from a manual service into an AI-first operation. The methodology below is what I draw on, scaled to the organisation in front of me.

Two patterns that stall rollouts

1. Users treat AI like a better search engine. They prompt, they read, they close the tab. The shift to "AI works for me in the background" never happens because nobody designed the harness around it. The licence sits on the shelf.
2. The harness is missing. AI rolled out as a feature is a vendor demo, not a rollout. A real rollout is the flow it sits inside, the decisions it is allowed to make, the handoffs back to humans, the training, the cost discipline, and the governance posture. That is the part most teams forget to build.

The ninety days

Week 1, map. Discovery workshops with each department, the same ones I have run for twelve years. Where do orders come in, where do decisions get made, where do people get stuck, where do they secretly already use ChatGPT. Same week, a quick read on AI literacy and a first pass on the AI inventory.

Weeks 2 to 3, pilot. Pick the highest-leverage flow and ship it end to end. Not everything needs an LLM. Some of the biggest wins are simple workflow automation that nobody had time to set up. This is the phase I ran on Memortium in 2024 to 2025. Map the order intake, find the seams, automate what does not need AI, layer the AI on top where it earns its keep.

Week 4, embed. Training paired with the flow people just adopted, not separate from it. The internal story aligned, augmentation not replacement, said openly by leadership.

Weeks 5 to 13, scale and govern. Roll the pattern into two or three more flows. In parallel, complete the AI inventory, classify each system against the EU AI Act, design human oversight for high-risk systems as real interaction patterns rather than disclaimers, document for audit, and roll out role-based AI literacy. FinOps for AI lives here too. Token-aware routing, model selection by job, monthly cost reviews.

Non-negotiables

• Design before deploy. Nothing ships before the flow it lives in is mapped.
• Augmentation as the public framing. Roles change, that conversation happens openly.
• Training paired with use. Generic AI courses fade within a month.
• Right model for the right job. Token-aware routing is the difference between paying for itself and bleeding margin.
• Human in the loop wherever judgement matters.
• Governance from day one. The Act becomes fully applicable on 2 August 2026.
• Cross-functional by default. Engineering, product, legal, security, GTM, comms.

What ninety days delivers

A flow map of how the organisation actually operates. A baseline of AI literacy with gaps and strengths called out. One pilot flow live in production with measurable outcome. Two or three more in measurable progress. A complete AI inventory with EU AI Act classification. Designed human oversight for every high-risk system. A role-based AI literacy program anchored in real flows. A cost dashboard with routing decisions documented. A clear roadmap for the next ninety days.

What it comes down to

The technology is the cheap part. Getting it to live inside the organisation, in a way the team trusts and uses every day, while staying compliant with the Act and inside budget, is the work. That is the part I have spent a career learning to see, and the part I run now at Memortium every day.