How is AI used inside 20xwork, and how do we tell people?
In short
Language models write the card text and judge next steps; the facts behind them come from your organisation's own data; every AI output is labelled and a person always decides.
Last updated September 27, 2026
Where models are used. To classify which step a task belongs to, to write the card in plain words, and to judge whether a next step names who, what and when. The facts on the card, such as which desks score highest and the hours returned, are computed from your organisation's data, not generated.
Which models. Models hosted on Microsoft Azure (Azure OpenAI and Azure AI Foundry). We may change the model behind a function; we do not change what data it receives without updating this page.
Labelling. Every card and check is marked as coming from 20xwork, and the product states that recommendations are suggestions.
Human in the loop. No card writes to a system of record on its own. A person keeps, adapts or dismisses it.
Accuracy. Cards can be wrong. Every card links to the evidence it rests on so a person can check it, and dismissals are tracked so patterns of poor cards are found and fixed.