Evidence
Client stories
Specific notes from programming consulting engagements on AI-assisted internal tools — including where we pushed back.
Short notes
“They refused to bolt a model onto our invoice triage until we fixed the status codes. That delay annoyed me at first, then saved a quarter of wasted engineering when the codes finally matched reality.”
“The discovery brief was denser than our last vendor proposal, in a good way. Our contractors stopped arguing about whether the chatbot was the product and started estimating the review queue.”
“The workshop forced our compliance lead and product owner into the same room. We left with a one-page risk register instead of another slide pack. I still wish we had booked a full day — the last hour felt compressed.”
“Readiness review caught that our pilot was sending free-text case notes to a third-party endpoint we had not documented. Embarrassing, necessary, and cheaper than learning it after rollout.”
Extended story: case-note drafting assist
A mid-sized Australian services firm asked us to “add AI” to case-note drafting used by thirty advisers. The existing prototype produced fluent paragraphs that quietly invented policy citations. During discovery we shadowed three advisers and found the real pain was locating the prior note template, not prose quality.
We recommended a retrieval-first assist that only drafted against approved clause banks, with mandatory human confirmation on any citation. Build guidance then covered the first two sprints while their engineers wired review states. The firm shipped a narrower tool than the original pitch — and retired the open-ended chatbot after six weeks of parallel use showed higher correction rates.
If your story resembles this, start with discovery or a readiness review if a pilot already exists.