Research & perspectives
What we're learning from enterprise AI engagements — strategy, product, design, and engineering, written by the people doing the work.
The gap between AI proof-of-concept and production deployment is where most initiatives quietly die. After working through dozens of enterprise AI engagements, we've identified the four patterns that separate the proje…
Most RAG failures are retrieval failures, not generation failures. The architecture decisions that separate systems that get it right from ones that confidently get it wrong.
AI features break the normal product rules — they're probabilistic, they drift, and users have no frame of reference for what "good" looks like. Here's how to ship them anyway.
The hardest part of enterprise AI is not the model — it's the person sitting in front of it. Design principles that move organisations from reluctant users to confident adopters.
Eighty percent of AI pilots never reach production. The engineering and organisational decisions that determine whether your initiative scales or quietly disappears.
AI readiness is not about having the latest tools. It's a function of data quality, leadership alignment, and the willingness to rethink processes — not just automate them.
Teams consistently budget 20% of effort for data and 80% for modelling. In practice it's the reverse — and discovering this midway through a project is expensive.