Two years ago, we started TechRock with a simple premise: that the gap between AI experiments and AI delivery was a delivery problem, and that solving delivery problems was something we knew how to do.
We were right about the premise. We were less right about almost everything else.
We underestimated how much of the work would be human rather than technical. We expected to spend most of our time on architecture decisions, model selection, and testing infrastructure. Those things matter. But the engagements that succeeded or failed did so almost entirely on the basis of stakeholder alignment, process clarity, and whether the right people had been involved at the right moments.
We overestimated how much organisations knew about their own processes. Before you can automate something with AI, you need to understand what you're automating. That sounds obvious. It turns out that most organisations have a significant gap between how they think their processes work and how those processes actually run on a Wednesday afternoon when someone is off sick and the system has an edge case it's never seen before.
We underestimated the regulatory learning curve. Even organisations that had thought carefully about AI governance were often surprised by what regulators actually wanted. The gap between having a policy and being able to demonstrate compliance is larger than most teams realise.