Beyond the pilot: what it takes to make enterprise AI real
Ask a room of business leaders whether their organisation uses AI and almost every hand goes up. Ask how many have changed the way a core part of the business runs, and the room goes quiet.
The research tells the same story. Our recent Google Workspace research with over 2,000 global leaders and employees has found that while 91% of organisations now have an AI-first strategy or goal, only 42% of employees say it has been clearly translated into day-to-day ways of working.
This does not mean AI has failed. It means we’re early. But my conversations with customers have moved on. They no longer ask whether to use AI. They want to know what’s real, and where the value lies.
Productivity is the starting point
For most organisations, AI arrived through their people, who use it to summarise emails, make sense of meeting notes and speed up research. Many now can’t imagine working without it.
That’s valuable, but it mostly speeds up individual tasks. If everyone saves an hour a week but their processes stay the same, the business hasn’t really changed. Personal productivity is a good start, but it’s not the whole opportunity.
From tasks to processes
The bigger prize lies in the end-to-end processes that run the business: the ones with many steps, systems and decisions, where one delay holds up everything that follows.
Take an insurance claim. A customer reports damage and sends photos. Someone checks the policy, reviews the images, looks for fraud, books a repairer, approves payment and keeps the customer updated. Each step depends on the one before, and the information sits in different teams’ systems. Using AI to draft the customer letter helps a little. Redesigning the whole flow so the claim never waits between steps changes both its cost and the customer’s experience.
This is where AI agents come in. Because they can run 24/7, work no longer sits in an inbox overnight between steps. They’re beginning to take on the multi-step work: reading documents and images, pulling data from several systems and passing decisions to people when judgement is needed. We are already seeing this in practice across the UK public sector. Working with the UK government, Google DeepMind and delivery partner Faculty on PlanAssist, local councils such as Barnet are applying AI across the four-stage planning assessment workflow, from validating documents and checking policy constraints to synthesising consultee feedback, cutting assessment times by 58% while keeping senior planners firmly in control of every decision.
The evidence supports the shift. This survey from McKinsey found that the organisations getting the most value from AI were nearly three times as likely to have fundamentally redesigned their workflows. Single-task pilots tend to stay small; the value comes when the whole process is rebuilt around the technology.
What production really takes
A model on its own isn’t a production system. It needs what’s often called a harness: access to the right systems and tools, memory of what has already happened, the ability to act (updating a record, sending a message) and a way to learn from feedback.
Around the harness sit the foundations any enterprise system needs:
Trusted data that is accurate, governed and available where it’s needed.
Integration with core systems, so AI works inside the business rather than in a separate window.
Security and governance, with clear rules on what AI can access, what it can do on its own, who is accountable and real-time guardrails that keep agents operating within defined boundaries.
Operational resilience, backed by continuous monitoring, rigorous testing and built-in continuity controls.
Cost control, because what’s affordable in a pilot can get expensive at scale.
Human oversight wherever judgement, risk or customer impact calls for it.
Skip these and projects stall between pilot and production. In our own research, while 27% of organisations are now actively scaling AI, only 10% report that they have successfully transformed how work actually gets done. That gap comes down to the foundations around the model, not the model itself.
Start with the business problem
Technology leaders rightly focus on the platform: which models to use, how they connect to existing systems and how it all stays secure. Business leaders want to know whether it fixes something that matters, such as delays, cost, customer experience or growth. And because large transformation programmes change how the organisation works, the business usually makes the final call.
So the strongest deployments start with the problem and the outcome, then work back to the technology, with both sets of leaders involved from day one. There is a science to this, but there is an art to leadership too: it takes leaders who continuously upskill themselves, understand the true capacity and capability of their organisation, and win hearts and minds to build the right culture. They also plan for people, who need to understand what’s changing and have the skills to work alongside AI. Yet our UK AI Works research shows that while appetite for AI is high, structured workforce training and confidence still lag behind adoption. Closing that skills gap alongside technical deployment is what turns a promising tool into lasting organisational change.
Choosing a partner for the long run
When leaders choose a partner, they usually want two things at once: the most advanced AI, and the stability and security they expect from any enterprise system.
A strong partner offers both. They understand that the goal isn't just upgrading legacy IT, but building a secure system of intelligence that works across messy data, strict security requirements and cultural change. They work through the business problem alongside the customer rather than handing over a toolkit. And they provide open, interoperable foundations, respecting that some organisations want to go deep with a single provider, while others choose to work across multiple clouds and models.
Count outcomes, not pilots
The organisations that get the most from AI won’t be the ones running the most experiments, but the ones with fewer, deeper deployments, each tied to a clear business result. Before scaling any pilot, ask:
What business outcome will this change, and how will we measure it?
Can we trust the data it depends on, and can we get to it?
Who owns it, on both the business and technology side?
How will we govern, monitor and pay for it at scale?
Where do people stay in the loop, and are our teams trained and ready to use the technology?
If the answers are clear, the pilot is ready to become part of how the business runs. If they aren’t, that’s where the real work begins.
