Research & insights
The thinking, published before it is billable.
What we have worked out on engagements and had no reason to keep. Written for practitioners: the method is visible, the uncertainty is stated, and where something we published has aged badly we say so rather than quietly deleting it.
Latest
What the EU AI Act actually asks of a retrieval system
Most teams building retrieval-augmented systems are preparing for the wrong obligations. The Act is not primarily interested in your model — it is interested in whether you can reconstruct, months later, why a particular answer was given.
All articles
Do not give every demand forecast a human override AI demand forecasts can improve inventory decisions, but human overrides help in some contexts and harm in others. Give planners a measured exception path instead of a blanket approval step. Automate the interview script, not the hiring decision AI voice interviews can make high-volume screening more consistent, but the business case depends on who continues, who starts and who stays. Automate information collection while people retain the hiring decision. Give AI the creative queue, not the campaign budget Generative AI makes advertising variation cheap and can improve click-through. Its commercial value appears only when a controlled test follows each creative through to conversion and the outcome the business keeps. Faster junior drafting needs an apprenticeship control AI can improve a junior professional's draft before it improves their judgement. Capture the added capacity, but keep a measured part of the workflow where the person must find and repair the defects themselves. Let AI fill the product-information gap before it rewrites the catalogue AI product copy is most useful where a buyer lacks information. Find the thin listings, ground each addition in verified attributes and measure whether completing them raises conversion without creating returns. A confidence threshold belongs to the model version Replacing a model while keeping its confidence threshold preserves a number, not a decision. Rebuild the link between score and outcome before the new version controls who is reviewed, approved or refused. Put the AI copilot beside the newest service agents first A service copilot creates value by shortening the path from new starter to capable agent. Start where experience is scarce, measure resolved work and protect quality before extending it to everybody. A drift alert is not a performance alarm An input distribution can move while a model remains useful, and a model can fail while its inputs look stable. Use drift to decide where to collect labels, not whether to retrain. An approval expires when the state it describes changes An agent can execute exactly the action a person approved and still be wrong because the target changed while the request waited. Bind approval to the reviewed state, and treat a version conflict as a new decision. Synthetic data is not one training policy Studies of recursive training reach different outcomes because “uses synthetic data” hides the decisive choice: whether generated examples replace the original corpus or accumulate beside it. Specify retention and sampling before judging the risk. Every agent retry needs the same operation ID A timeout does not tell an agent whether its tool call failed. Retrying without the original operation ID can repeat the effect, so idempotency belongs in the tool contract before an agent reaches production. Your backtest knows more than your system did A chronological train/test split can still leak the future. Historical evaluation must reconstruct the information available at each decision, including late arrivals and corrections, before its score can support a deployment decision.
One note a fortnight. No newsletter voice.
New research, regulatory changes worth your attention, and the occasional admission that something we published has aged badly.
Subscribe by RSS →
Or email us to be added to the list
No tracking, no list broker, no third-party form. We hold the address and use it for this and nothing else.