A mid-market financial services company spent most of 2025 deploying AI across operations. Email summarisation tools. Automated expense categorisation. Prospect enrichment experiments. Each pilot delivered something — time savings, thousands of dollars here, reduced manual effort there.
By October, the CEO asked: “Why aren’t we getting value?”
The operations leader realised the answer wasn’t complicated. Pilots worked in isolation. Sales innovations never reached marketing. Finance governance decisions stayed silent to HR. Champions existed throughout the company. Nobody had organised them. The system treated each implementation as self-contained.
This is the pilot trap. Strategy says scale; execution requires something different.
Foundations first: why pilots stall
The OpenAI framework calls out what most organisations misunderstand — you cannot scale pilots into products without first establishing “foundations.” This isn’t delay. It’s the thing that determines whether pilots compound value or sit siloed, unable to reach the rest of the organisation.
Foundations means three categories: governance that works and evolves, executive alignment that’s genuine, and data access that doesn’t require three approval chains before movement happens. Without these, your pilot becomes a pocket of activity — useful nowhere else. With them, each experiment feeds forward.
The governance piece matters most. Organisations that built proper governance frameworks for AI scored 6.6 points higher on readiness scales. Not because governance is fun, but because it removes the slowdowns that kill momentum during scale. When you hit build phase and realise you can’t access certain datasets without new approval processes, you halt. Governance done right upfront — and continuously adjusted — prevents that halt.
That same mid-market company had executives excited about AI but no shared understanding of success. One team measured ROI as time saved; another as cost reduction; another as new feature velocity. Frameworks never aligned. When it came time to decide which pilot to invest in next, they couldn’t compare. Investment decisions happened by politics, not by evidence.
Building systems, not collecting pilots
The company’s biggest untapped resource sat everywhere: people who’d already learned to use these tools in their specific context. A financial analyst had figured out prompt structuring for accounting workflows. A customer service manager knew when AI worked and when it failed. But nobody had time to formalise that knowledge.
This wasn’t laziness. It was structural. Those people were too busy doing their normal jobs to document and share what they’d learned about doing those jobs better. The framework calls this out directly: allocate formal time for learning, make it an organisational habit, build champion networks that translate general lessons into context-specific guidance.
In 1000+ enterprise interviews, the pattern was consistent — people were too busy to learn the thing that saves them time. The solution isn’t motivational speeches. It’s carved-out blocks on the calendar, permission from management, systems to route what champions learn back into the organisation.
By January 2026, the company reframed their approach. Same pilots; different mental model. A governance working group — not to slow things down, but to create clarity around data access and approval pathways. Executives got into the tools themselves, not as spectators. Two hours per week per team: structured learning with champion specialists facilitating. Documentation came next: what worked? What patterns repeat? Where could code, prompts, orchestration flows, or data assets serve multiple teams?
The shift wasn’t philosophical. It was operational. Pilots remained pilots, but they were designed to feed forward — into better governance, into a learning culture, into reusable components the next project could build on. Compounding ROI doesn’t mean each pilot instantly becomes profitable. It means each pilot teaches the organisation something that makes the next initiative faster, cheaper, higher-impact.
The CEO’s original question shifted. It became: “How do we design our foundational systems so that every project we do makes us better at doing the next one?” That’s the operating model that separates organisations moving into 2026 confident in their AI capability from those stuck running pilots that never compound.
Start the reframe now.
Signal42 | Beyond the hype. Into the value. |