Notes on software, AI and the sectors we build in.
Nothing matches that.
Nominet suspended a .uk domain three days after we registered it, with no warning, in an email indistinguishable from phishing. What happened, why the process is worth knowing about, and how to avoid it costing you a client.
Read the articleMost AI work fails long before the model. It fails in the reconciliation nobody wanted to do, and no amount of model selection rescues it.
Read the articleIf a model contributes to a decision someone can challenge, the explanation has to be produced by the same machinery that produced the answer.
Read the articleThe gap between a convincing pilot and a system people use is not a modelling gap. It is the work nobody scoped.
Read the articleYouth clubs hold sensitive information about children, gathered by volunteers, in whatever tool was convenient. Most of the risk is avoidable and none of it is complicated.
Read the articleMost clubs have tried a system and stopped using it. The reason is rarely the software being bad, and almost always the same thing.
Read the articleChasing subscriptions is the job that burns out club treasurers. Most of the fix is structural rather than a matter of reminding people harder.
Read the articleAn accessibility audit at the end of a project finds real problems at the most expensive possible moment. Almost all of them were avoidable weeks earlier.
Read the articleMost logging records that something happened. What gets asked later is why, by whom, and what the system knew at the time, which is a different design.
Read the articleMost automation fails at adoption rather than at build. Anything that asks people to open another system is competing with a process that already works.
Read the articleEvery supplier relationship ends. Whether that is routine or painful is decided by choices made in the first month, not the last one.
Read the articleA lapsed card is discovered at a site gate on a Monday morning. The information needed to prevent it existed for months in a system nobody planning work could see.
Read the articleMost booking systems are built as diaries. The thing operators actually need to manage is the gap between what was booked and what turned up.
Read the articleMost firms assemble evidence for an audit by going and finding it. The alternative is a system where the record accumulates as the work is done.
Read the articleHours rules cap how long someone can be on duty. They were never designed to measure whether that person is fit to drive, and the gap between the two is where incidents happen.
Read the articleA single agreement at registration is easy to build and rarely means what it claims. Real consent is granular, revocable, and has to survive contact with a live system.
Read the articleTwo teams reporting on the same outage produce different numbers, and both are correct. Until the definition is settled, every improvement figure is unfalsifiable.
Read the articleDaily vehicle checks are near-universal and largely wasted. The problem is almost never the check itself. It is what happens in the hours after one finds something.
Read the articleMedia interfaces are used habitually, for hours, by people doing something else. Friction that is acceptable on first use becomes intolerable by the hundredth.
Read the articleMost travel software is designed around the timetable. Passengers judge it almost entirely on the days the timetable stops being true.
Read the articleMost cancellations are set in motion long before anyone cancels. The window that matters is much earlier and much shorter than most operators plan for.
Read the articleOffline-first is not a feature you add to site software. It is a decision that shapes the data model, the sync strategy and the interface, and it has to be made early.
Read the articleAsk four retail systems how many of something you have and you will get four numbers. The reconciliation that follows is a symptom, not the problem.
Read the articleManaging fatigue is a legal duty in rail. Producing evidence that you manage it is a separate problem, and it is the one that catches operators out.
Read the articleMost infrastructure businesses cannot say what their plant did last month. The cost of that is larger than it looks, and it is rarely a data problem.
Read the articleThe hard part is not predicting the failure. It is producing a prediction an engineer is willing to take an asset offline for.
Read the articleRecommendation projects are scoped as machine learning work. They usually stall on a taxonomy nobody owns and a back catalogue that was tagged by whoever ingested it.
Read the articleThe plan to replace everything at once is the plan that never finishes. Most retail estates are improved by deciding what stays, not by deciding what goes.
Read the articleMost retention policies are documents. The data they describe is still there, because deletion was left to a person with more urgent work.
Read the articleReturns get managed as a warehouse cost. Most of the rate is decided much earlier, by information that was wrong or missing on the product page.
Read the articleMost data protection effort goes on securing information that never needed to exist. Minimisation is the cheapest control available and the least used.
Read the articleDiscovery investment goes to recommendation. Search gets the budget left over, despite being the thing people reach for when they know exactly what they want.
Read the articlePutting your own brand on the app members use is not vanity. It changes who the relationship belongs to, and it shows up in the numbers.
Read the articleFirms automate the process someone complained about loudest. The expensive work is usually quieter than that, and finding it takes a fortnight of watching.
Read the articleWhy the site moved from hand-written PHP to a modern static build, and what changed under the bonnet.
Read the article