Lessons on shipping AI in large organisations from OpenAI, Experian and Octopus
by CJ Daniel-Nield ✦ 21 October 2025 ✦ 5 Min Read
At the end of an API call, you now have access to some of the most powerful models on the planet. You can build and launch incredible new experiences for your customers.
And yet, three years into the GenAI wave, most teams still haven’t made it past the prototype.
There’s a gap between ambition and execution. A backlog of demos, POCs, and disconnected use cases.
We sat down with product leaders from OpenAI, Experian UK&I, Octopus and Lewis Silkin, who are building AI products inside large organisations and regulated industries.
Here are the top learnings they shared for leaders trying to do the same.
1. Lean into what your business does best
It’s tempting to chase whatever’s trending: agents, copilots, assistants. But the teams deploying AI successfully are the ones that start with a user problem and focus on making their core experience better.
As Matt Weaver from OpenAI explained, the companies seeing the most success are those “taking the thing that’s most core and differentiated to their business and investing there.”
For example, Virgin Atlantic are using AI to amplify the thing that already makes them great: their people. They’re now building a travel concierge into the app, pulling recommendations directly from flight crew.
For Lewis Silkin and Octopus Legacy, it wasn’t about using AI for the sake of it. It was about solving real customer friction. “The ask wasn’t ‘go build me a AI product,’” said Ines Liberato. “The ask was ‘how can we make this better?’ And the answer just happened to be AI.”
If you’re not sure where to begin, don’t start with a long list of use cases. Start with your company’s ambition. What’s the thing only your business can do? What does leadership want to be known for in three years?
“It’s easy to get lost in a spreadsheet of 100 AI use cases,” Matt added. “But when you slot them into a strategy with real vision, you can drive the whole organisation forward.”
2. Progress starts with prototypes
AI has changed the speed and cost of learning. For teams willing to get stuck in, every experiment adds value.
As Christine Foster from Experian UK&I put it, “Nobody ever caught a big wave by standing on the beach.” Once you’ve found a promising use case, the only way to validate it is to build.
For Ines, that willingness to move early turned into a strategic advantage: “We built an AI prototype and put it in front of clients. The response was: ‘This is a game changer.’ Showing value that early gave us the buy-in to put it into production.”
With models evolving overnight, it’s easy to hesitate. But the best teams lean in, even knowing some of it won’t stick. As Matt said, “You have to get comfortable with 30% of what you build being thrown away.”
What looks like rework is actually compounding advantage. Every prototype builds context. Every test brings you closer to unlocking value. And every time you ship, you build the confidence and capability to ship again.
3. Trust is built in, not bolted on
In high-stakes, high-trust environments like law and finance, people don’t want a black box. They want to see the wiring.
That means designing for confidence, not just convenience. As Christine put it, “We ended up putting friction into the product. Turns out, customers want to see the guts of the thing.”
Admin controls, off switches, and clear explainers aren’t just nice-to-haves, they’re what make AI feel usable and safe.
Internally, trust comes from exposure. In the most effective teams, leadership, legal, and compliance got early access to AI tools. They explored the tech, understood the edge cases, and helped shape the guardrails. That shift turned potential blockers into collaborators. As Matt explained, “We gave ChatGPT licenses to our legal and compliance teams. Once they’ve used the tools themselves, they understand the risks better.”
But perhaps the most critical trust mechanism sits in the process. Evals are becoming the backbone of safe, scalable AI products, helping teams measure performance, compare changes, and align on what “working” really means. As Matt put it, “The best teams treat evals as core product infrastructure, not a checklist.”
Trust doesn’t come from polish. It comes from transparency, clarity, and the systems you build in from day one.
New tools, same fundamentals
AI has raised the stakes. It’s easier than ever to ship something that looks impressive but falls flat in practice. The teams making meaningful progress are holding onto the fundamentals:
Solving problems rooted in user needs
Building quickly to test assumptions, prove value, and create internal pull
Engineering trust with transparency, control, and rigorous evaluation from day one
The teams that move fast without losing focus, who build with purpose, not panic, will be the ones who ultimately ride the biggest wave.
If you're ready to move from thinking to shipping, let’s talk at hey@planes.agency.
