The three-week review now takes an afternoon.
Xponance, a multi-strategy investment firm, reviews the operations of every funded manager on a fixed two-year cadence. Damco built an AI-enabled due-diligence system for the team that runs those reviews: AI reads each manager's documents and drafts the questionnaire with citations, and analysts validate the drafts and keep the scoring. A review that took two to three weeks of an analyst's time now takes two to three hours.
- less analyst time per review: two to three weeks of reading down to two to three hours
- 95%less analyst time per review: two to three weeks of reading down to two to three hours
- reviews initiated in the first three weeks of production, with no added headcount
- 71reviews initiated in the first three weeks of production, with no added headcount
- delivery milestones hit on the planned date, across an eight-week build
- 9/9delivery milestones hit on the planned date, across an eight-week build
Measured against baselines drawn from the firm's own historical review data. Phase 1 was accepted through formal UAT; the production figures are from the first three weeks after go-live.
How it works
AI does the reading. Analysts do the judging.
Managers submit their documents through a secure portal, the same way every cycle. What changed is who does the reading.
Q14 · Does the manager maintain a personal-trading policy?
Code of ethics · §4.2
Scoring stays with the analyst. The AI never assigns a score. That judgment is the review.
When a cycle is finalised, the standardised reports, radar charts and trend analyses generate themselves, and every action along the way lands on an immutable audit trail.
How we did it
Automation that amplifies analysts
A due-diligence review is a compliance exercise wearing the clothes of a reading assignment. Four controls keep the judgment human while the reading gets automated.
Judgment stays with the analysts.
The split is a design rule, not a hope. The pipeline reads each manager's documents and pre-fills the questionnaire; analysts validate every draft and assign every score. Judgment stays exactly where the compliance risk is.
Every answer cites its source.
Each drafted response links back to the passage in the manager's own documents it came from. Validating a draft means checking a reference, not re-reading a compliance manual.
The risky assumptions were locked first.
The questionnaire structure, the calibration documents and the report template were agreed with Xponance before development began. The three things most likely to sink the build surfaced in week one, not week seven.
Consistency is a property of the system.
Every review runs through one workspace: same questionnaire, same scoring scale, every cycle, with an immutable audit trail behind it. Scoring no longer varies with who did the reading.
How success was measured
Gates on delivery, KPIs on impact
Delivery ran against dated go/no-go gates: specifications approved, architecture signed off, a working build through UAT with critical defects cleared. Impact is tracked the same way, through a KPI framework covering analyst time, throughput, scoring consistency, turnaround and override rates, baselined on the firm's own historical data. Every improvement is measured against the real prior state.
- reviews initiated in the first three weeks after go-live
- 71reviews initiated in the first three weeks after go-live
- in active review at the three-week mark
- 13in active review at the three-week mark
- already fully completed
- 3already fully completed
That throughput came from the same one or two analysts who previously finished one or two reviews a month, with no added headcount. Within weeks, Xponance greenlit a second phase: a 24-item enhancement backlog.
If a fixed review cadence is throttled by reading time, the first useful conversation is about which judgments must stay with your analysts.
We'll tell you when not to automate.