Case studies
What we diagnosed. What we built. What changed.
Three real engagements, all run the same way: diagnose the process, design the system, build it, keep it running. In every one of them, the system we shipped was not the one the client first asked for. Client names withheld under confidentiality.
We built the system they needed, not the one they asked for.
A 30-attorney IP firm asked us to automate opportunity discovery. Our diagnostic uncovered that they were already finding opportunities — 300+ daily alerts per associate were going unopened. The bottleneck was downstream, in triage and reporting, not in discovery.
What we built
- Monitoring agent that ranks inbound alerts by client relevance rather than arrival order
- Drafting layer that assembles the quarterly client opportunity reports from ranked alerts
Report drafting, per client per quarter
Daily alerts per associate going unopened
- Client
- 30-attorney IP firm
- Timeline
- 10 weeks · 2 phases
- Engagement
- Diagnostic → two implementation phases
We shipped the foundations an AI layer could actually stand on.
A mid-sized UK university came in wanting AI essay grading. Our diagnostic established that grading wasn’t the root issue — fragmented submission intake and inconsistent faculty rubrics were. A model deployed on that foundation would have produced unreliable results, so we built the foundations first.
What we built
- Submission intake system — one structured pipeline replacing fragmented channels
- Faculty rubric system — consistent criteria applied across markers
Both foundation phases are live. AI-assisted grading is now a credible third phase rather than a failed launch — which is what it would have been had we built it first.
- Client
- ~7,000 undergraduates, essay-heavy
- Timeline
- 13 weeks · 2 of 3 phases shipped
- Engagement
- Diagnostic → two foundation phases
We built monitoring for the exceptions nobody was logging.
A 60-employee freight forwarder asked us to automate a slow exception-handling pipeline. Our diagnostic uncovered that the firm was tracking exceptions that got escalated, not exceptions that went missing entirely — and the most expensive ones were never logged at all.
What we built
- Monitoring agent across all seven carrier feeds, including exceptions that were never raised internally
- Structured triage classifier that routes inbound exceptions by type and urgency
Triage delay on routine exceptions
Carrier feeds under active monitoring
- Client
- 60-employee freight forwarder
- Timeline
- 11 weeks · 2 phases
- Engagement
- Diagnostic → two implementation phases
Something that should already be working?
A 20–30 minute fit call. We respond within one business day.