In May 2026 we relaunched our AI work as a formal AI Transformation program. The question changed from “are people using AI?” to “where is AI changing how we win, build and serve – and can we prove it?”
Two tracks: value and enablement
We organised the program around two complementary tracks. One is pulled by business outcomes; the other builds the foundations everyone depends on.
Business Value Streams are owned by leaders in the business, not by IT. Each has a named lead and its own KPIs.
| Value stream | What it is about |
|---|---|
| Win More Deals | AI across marketing and sales, from prospect to qualified opportunity |
| Product Innovation | AI in what we ship to retailers |
| Time2Market | Faster, safer engineering from idea to production |
| Future of Services | AI-assisted support, monitoring and incident handling |
| G&A / Productivity | Everyday efficiency in finance, IT, HR and other support functions |
Enablement & Foundation working groups do the work no single stream can do alone.
| Working group | What it is about |
|---|---|
| Communication & Learning Culture | Building AI fluency through learning programs, a shared prompt library and regular knowledge sharing |
| Data Foundation | Making company data accessible, trustworthy and ready for AI use |
| Governance & Security | Guardrails, AI policies and approved tools that keep adoption safe and compliant |
| New Tool Evaluation | Assessing new AI tools and platforms before they enter the organisation |
| Tool Adoption | Rolling out approved tools and helping teams get real value from them |
This split was the most important design choice we made. Value streams keep the program honest about outcomes. Working groups stop five streams from solving the same security, data and training problems five times.
Building capabilities, not just pilots
The program is now putting real AI capabilities into the organisation.
- Engineering: a community of engineers is building a shared AI skills marketplace and a blueprint tool that scaffolds new services with the right architecture patterns. AI now writes first-pass investigation summaries for alerts, and a monitoring agent tracks e-invoicing and tax-compliance changes across 16 countries every day.
- Services: triage and incident-response agents are on the roadmap to take a growing share of monitoring events and support incidents.
- Sales and marketing: a working AI stack combining Claude with our CRM and prospecting tools supports pipeline generation, with an RFP-response agent next.
- Product: we are designing an AI Capability Layer on top of our open APIs, and a proactive “Tier-Zero” support agent that resolves issues before customers raise them.
- Everyone: an approved toolset, a prompt library and a steady rhythm of learning, so AI is part of the job rather than a side project.
Why we established an AI CoE
A program has an end date. Capabilities need a home. So we are standing up an AI Center of Excellence on a hub-and-spoke model.
The hub is small and deliberate: a Chief AI Officer role, a Head of AI CoE and AI Enablement Engineers, mostly built through internal moves rather than external hiring. It owns the platform choices, the guardrails and the reusable building blocks, such as our central MCP gateway for connecting AI securely to company data.
The spokes are the value streams and the people in them. They own the use cases and the outcomes. The CoE’s job is to make them faster and safer, not to do the work for them.
In practice the CoE nurtures capability in three ways:
- People: growing champions in every team, and turning early adopters into coaches.
- Patterns: capturing what works — skills, prompts, agent designs — once, so it can be reused everywhere.
- Proof: weekly value-stream reporting and a disciplined value ledger that separates what is planned, what is in progress and what is actually realised.
Honest about value
That last point matters. It is easy to claim AI value; it is harder to show it. Some of our metrics are moving clearly, others have not moved yet, and we report both. Credibility with the board and the business is worth more than an inflated number.
What I have learned
- Put the business in the driving seat. AI value streams led by IT become IT projects.
- Centralise the foundations, decentralise the use cases. Security, data and tooling once; ideas everywhere.
- Build a home for the capability. A CoE turns a one-off push into something that compounds.
- Measure with discipline. Separate the promise from the proof.
We are not finished, and with AI no one ever will be. But we have moved from scattered experiments to a governed, measurable operating model that keeps building capability long after the launch energy fades.