// Evidence
What we know, and how we came to know it.
We have no wall of testimonials. What we have is one build published in full, a standard every figure on this site is held to, and a correction log that is public rather than quiet.
The case study
/casestudies/customer-servicing
One build, published in full: what the volume looked like, what got built, how resolution was defined, and what was hard. The definitions travel; the number does not.
// Live
Context Window
/podcast
Our podcast and newsletter on AI in insurance, with the full transcript alongside every episode.
// Per episode
Corrections
/evidence/corrections
What we got wrong, when we changed it, and what it changed to. A public log, not a quiet edit. It starts empty, which is the only honest way for a log to start.
// As needed
Field notes
/evidence/field-notes
What we saw inside the work, written by the engineer who saw it. Specific, dated, and occasionally about something that did not work.
// From launch
// How we handle a number
Attribution travels with the figure.
“AI adoption is 73%” and “AI adoption is 10%” are both true, depending on the definition. So every figure on this site carries its source and its definition, or it does not appear.
Statistics older than eighteen months are dated in the copy itself. Vendor-published research is labeled as vendor-published — several of the most useful numbers in this industry come from parties with something to sell, and they are still worth citing. They are not worth laundering.
Where two credible sources disagree, we publish both. Showing the disagreement is more persuasive than picking a side, and it is the behavior the rest of this site promises.
Our Fall cohort is already 75% full. Reclaim your operations now.
Bring us the workflow you’ve already tried to fix.
A decade of running an AI-native carrier, pointed at one of your workflows for four weeks.