Every value your chatbot writes now says where it came from - plus dry runs and agenda stats

When software fills in your CRM for you, the obvious question is how do you know that's right? Agendas now answer it. Every value a bot learns is recorded with its provenance: whether it matched a recognised format (an email, a phone number, an amount) or was inferred by AI - and if so, which model and how confident it was - along with which record field it was written to, or why a write was refused. Open a chat conversation in the Inbox and you'll see a panel listing what the bot has gathered, with values it inferred marked AI, so an agent taking over knows which details to double-check rather than treating everything as equally solid. Fields you mark as personal data are redacted in that log - an email shows as s…@acme.com, a phone as its last four digits, free text as a character count. The real value still goes to the record where it belongs; the log keeps enough to recognise a value you already know and not enough to learn one you don't. Deliberately not a hash, either, since a hash of an email address is still a stable identifier. You can also set a retention window per agenda, after which the value payloads are cleared while the trail of what was written, when, by which model is kept - because the audit question is about the write, not the value. Two more things arrived with it. Dry runs: type a few messages a visitor might send and see exactly what the agenda would extract, which question it would ask, where each value would land, and how the person would qualify - with nothing written, no AI called and no credits spent. It also flags which turns would have needed AI, so you can see what an agenda costs before you turn it on. And stats: per-field asked / answered / written counts, completion and qualification rates, median time to complete, and the share of values that came from AI versus a free format match - a number that's both a cost signal and a trust signal. Agenda runs are also a first-class report source now, so qualification rates by agenda, by channel or over time work in the normal report builder and on dashboards, rather than being locked in a statistics panel.