AI Implementation · UK Services
Buying an AI licence is 10% of implementation. The other 90% — governance, workflow selection, infrastructure build, training, and measurement — is where most businesses go wrong.
When most business leaders think about AI implementation, they think about the technology: which tool, which licence, how to set it up. This is natural — it is the visible part of the problem. But it is the smallest part.
The largest implementation challenges are invisible until they are not: the governance gap that becomes a GDPR incident; the training gap that becomes tool abandonment; the change management gap that becomes cultural resistance; the measurement gap that means you cannot demonstrate ROI when the renewal conversation arrives. None of these problems appear on the tool vendor's pricing page.
Genuine AI implementation addresses all five components in the right sequence. The technology is the last thing you build — after you know what you are trying to achieve, who will use it, how it will be governed, and how you will know it is working.
Before any tool goes live: AI use policy, data classification, approved tool list, prohibited uses, accountability structures, incident process. This is the foundation. Everything else rests on it. Governance built after implementation is crisis management.
Identifying which processes will be supported by AI and in what priority order. The best starting workflows are high-frequency, high-value, and low-risk — processes that cost significant time, create real value, and are not so sensitive that a governance gap would create immediate exposure. Typically 3–5 workflows for the foundation phase.
Building the AI Operating System: custom assistants for each target workflow, knowledge base populated with your processes and context, tone and voice configuration, integrations with existing tools. The infrastructure layer is what turns a generic AI into a business-specific one.
Training, change management, and the change engine. Who gets trained first, in what format, using what curriculum. How the champion network is built and activated. How sceptics are engaged. How leadership models the behaviour. The people stage is where most implementations either take root or fail to.
The proof-of-value framework: what metrics demonstrate success, how they are measured, and how the results are reported to leadership and the board. Without measurement, implementation has no feedback loop and no accountability. With it, you can demonstrate ROI and justify the next phase of investment.
Generic AI deployed into an unchanged organisation. No custom assistants. No knowledge base. No configured tone. People get generic outputs that require heavy editing to be usable. Value is low. Adoption is patchy. The tool is gradually abandoned.
Tools go live, then a governance incident happens — a data leakage concern, a client complaint, a regulator inquiry — and governance is bolted on in crisis mode. This is expensive, disruptive, and creates a record of poor governance practice that is hard to erase.
A one-day training event is run. People learn the basics. Within 60 days, habits have reverted. There is no reinforcement mechanism, no champion network, no ongoing support. The training investment produces no lasting return.
Implementation proceeds without an agreed definition of success. When renewal time comes, nobody can demonstrate ROI with confidence. The implementation is cut or scaled back — not because it was not creating value, but because that value was never measured.
Five stages: governance (policies, data classification, accountability); workflow selection; infrastructure build (custom assistants, knowledge base); people (training, change management); and measurement (proof-of-value framework, optimisation). The software is 10% of the challenge.
Governance defines what the AI is permitted to do, with what data, and under whose oversight. Implementing AI before governance means deploying into an uncontrolled environment. When governance problems emerge — and they always do — you are managing a live incident rather than a design decision.
'Tool without context' — deploying a generic AI tool into an unchanged organisation and expecting it to create value. Without a configured environment, trained people, and governance, generic AI tools create as many problems as they solve.
The Ignite AI Implementation Plan is produced as part of the SPARK discovery. It documents current state, governance gaps, workflow priorities, technology architecture, training requirements, and the 90-day proof milestone.
A foundational implementation — governance in place, first AI OS workflows live, first training cohort complete — is typically delivered within 90 days. Full implementation takes 6–12 months.
A SPARK discovery produces your AI Implementation Plan: what to do, in what order, with what governance, and how to measure success. The right start to any AI implementation.