Insights
Working perspectives on the decisions that sit between technology, risk, governance and investment — written for the people who have to make them.
Topics
46 articles · page 5 of 6
Enterprise systems · 8 August 2026 · 7 min read
Schools do not need bespoke software. They need software they can bend to their own rules without waiting on a developer. That conviction produced an architecture of eight engines and turned most feature requests into settings.
Read the articleEnterprise systems · 5 August 2026 · 7 min read
Every school on EduFlow gets its own database. Here is why we chose the harder path over a shared table with a tenant column, and what that decision buys the schools that trust us with their data.
Read the article
Artificial intelligence · 30 July 2026 · 5 min read
Useful AI in an enterprise depends less on model choice than on scoping, data boundaries and a clear account of who is responsible for the output.
Read the articleArtificial intelligence · 21 July 2026 · 7 min read
The hard part of production machine learning is not the modelling; it is being honest about what the models do and do not know, consistently, in a hundred small places. Here are the honesty patterns that make a platform trusted.
Read the articleArtificial intelligence · 17 July 2026 · 6 min read
Scores are not decisions. A decisioning layer ranks the next best action by expected value, and knows the difference between an assumption and a learned fact rather than letting the first wear the costume of the second.
Read the article
Governance · 16 July 2026 · 5 min read
Most supplier risk programmes collect questionnaires. The useful ones know which suppliers could stop the business and what happens if they do.
Read the articleArtificial intelligence · 14 July 2026 · 7 min read
A popular-now shelf is not personalisation and a nightly batch is stale by lunchtime. Here is how an offline model that learns taste and an online contextual bandit that decides and explores fit together.
Read the articleArtificial intelligence · 10 July 2026 · 7 min read
A bare lifetime-value number sounds precise and is quietly dishonest, because it hides how sure the model is. Here is how a hurdle model and conformal prediction ship value as a calibrated range instead.
Read the article