Home / Case studies / Agentic monitoring
A 26-day monitoring cycle, compressed to three
$8B+ alternative asset manager
Days, end-to-end cycle time
Effective monitoring frequency on the same headcount
Judgment decisions still made and approved by a human
- 01Documents arrive24 handoffs before
- 02Agents transportand verify
- 03Exceptions rankedby materiality
- 04Human judgmentdecision points
Transport and verification automated; judgment and accountability kept human
The situation
A core portfolio monitoring process ran 26 days end to end — not one slow step but two dozen handoffs, each waiting on a person to pull a document, check it against another, and pass it along. Because the cycle was longer than the month, the firm was structurally always looking at a stale picture of its own portfolio.
What we built
- 01A process teardown first — every handoff mapped, timed, and classified as judgment, verification, or transport
- 02An agentic workflow executing the transport and verification steps end to end, with retries and escalation
- 03Human approval preserved at the genuine judgment points — automation removed waiting, not accountability
- 04Exception queues ranked by materiality, so analysts keep the interesting 5% of the work
- 05Cycle-time instrumentation so the next bottleneck is visible the moment it becomes one
Most “slow process” problems aren’t slow steps — they’re waiting. Map the handoffs before choosing a model, and the highest-value automation is usually unglamorous: moving documents and checking them against each other, reliably, at three in the morning.