IllustrativeThis case study is anonymized and modeled after a real prospect conversation. No customer is identified or quoted. Real, named stories will replace these as pilots go live.
MLRO and L2 analysts went from 90-minute SAR drafts to under 30 minutes by combining the LLM composer with the LME case-build-up.
12 d → 4 d median
90 min → 28 min
30% → 8%
A retail bank running ~80,000 alerts/month had a 12-day median time-to-SAR. Analysts were copy-pasting between three systems (case manager, KYC, screening) to build the narrative, then hand-typing the goAML XML. The MLRO routinely returned 30% of drafts for rework because the narrative didn't connect the typology to the evidence on file.
The bank deployed RiskSonnar Pro across two jurisdictions. Analysts use the LME (Link Map Engine) to build the case neighbourhood, then call the SAR composer to draft a narrative grounded in the actual case events. The composer cites every claim against a hash-chained audit event, so the MLRO can verify the evidence in two clicks. goAML XML is generated from the same canonical case object — no re-keying.
Median time-to-SAR dropped from 12 days to 4. Per-SAR drafting time fell from ~90 minutes to ~28 minutes. MLRO rework rate dropped from 30% to under 8%. The bank filed 3.2× more SARs in the second quarter without growing the team — a direct response to the regulator's targeted review.
“We didn't need to retrain anyone. The composer reads the case the same way our analysts do, and every paragraph cites the audit event it came from. The MLRO trusts what's on the page because she can verify it without leaving the screen.”
The synthetic-data sandbox runs the same surfaces this story used, against deterministic test data. No customer setup needed — the session auto-expires after 60 minutes.