There are two approaches to AI investment emerging in financial services. The first is the two-year bet. Large programmes. Central funding. Roadmaps built on the assumption that value will follow if the organisation commits for long enough.
The second is the 90-day pilot. Fast, focused experiments. Small teams. Clear use cases. Visible progress. Most firms have already made their choice. Large AI programmes fail for the same reason large transformation programmes always have. They require certainty up front, stable requirements, predictable outcomes & defined benefits. None of this exists with AI so they are defaulting to pilots.
And whilst this is the right approach, it isn’t without risk. 90-day pilots don’t fail, they are hard to scale.
Why 90‑day pilots work
Pilots embrace uncertainty. They allow for rapid feedback, low commitment and provide visible output. They are easier to fund because they look controllable and the budgets are far smaller. They are also easier to deliver because they avoid complexity. This is exactly why most firms are choosing them. The assumption is the pilot will scale, and this is where the problem starts.
The uncomfortable truth about pilots
It’s easy to say “we are doing something with AI” if you run a pilot.
They appear to succeed because they are protected from reality. They rely on curated data, small volumes, limited integration and often vague or absent success criteria.
When you want to scale, the constraints appear and progress stalls. Data isn’t clean. Edge cases were ignored. Downstream systems and processes can’t absorb the output. Ownership and recovery are unclear.
The technology is rarely the problem. The issue is expectation setting. Pilots typically prove the happy path, they tend to avoid alternate and exception paths. Stakeholders see the working shiny demo and expect rapid deployment. The pilot teams know the solution isn’t ready to scale so the pilot stalls or is deployed with manual workarounds. At volume, these work arounds break. Manual review queues grow. Processing slows and the process becomes more expensive than the one it was meant to replace.
Done properly, a pilot manages expectations and exposes the scale challenges that need to be solved for the solution to be sustainable.
From pilot to scale. Getting the sequence right
Pilots are not an alternative to foundation work. They are how you identify the foundation work required. If the pilot doesn’t produce reusable components, data pipelines, controls, patterns, you start again when you try to scale. That is why some pilots accelerate scaling, while others delay it. The sequence to success is straightforward. Most firms just don’t follow it.
- Start with a high‑friction use case: Not the most exciting one. The one where inefficiency is visible and measurable.
- Run a focused pilot: Prove that AI can improve the outcome. Keep the scope tight.
- Use the pilot to surface constraints: Where does it break outside the controlled environment. Data, systems, ownership, governance.
- Invest in removing those constraints and capture the patterns required to reuse the solution: This is the step most firms skip. It’s avoided because it can be expensive, slow and politically harder to fund than a visible pilot.
- Then scale incrementally, using what you have built rather than rebuilding it: Scaling isn’t simply extending the pilot. It is a separate phase that formalises, hardens and integrates what the pilot proved.
Scaling pilots incrementally won’t give you the early momentum everyone wants, but it reduces the risk of failure when you move to production.
Most firms know how to run pilots. Scaling them is where things stall. If you want help closing that gap, come and talk to us at Simplify Consulting.
Chris Moore
Head of Solution Architecture
