Echo Cancellation

Overview

Echo Cancellation was Smarsh's first AI capability inside Enterprise Supervision — the platform global banks, broker-dealers, and insurers use to review regulated communications. I owned it end to end: from the customer insight that framed the problem, through the design of a production pilot, to general availability.

The problem

Compliance review teams drown in noise. In email supervision, a single flagged message gets quoted and re-quoted as people reply and forward — so one genuine hit “echoes” into dozens of redundant alerts, re-emerging every time someone picks the thread back up. These echoes were estimated to account for roughly half of the low-value alerts in a supervision system, and no reliable method existed to separate new content from old across a thread.

What I built

As the product manager I designed the echo-detection algorithm — the rules that decide, for every new alert, whether it carries genuine risk or is an exact echo of already-reviewed content. I owned that detection logic end to end, then worked with engineering to ship it and with our financial-services clients to make it controllable at the policy level — and to hold it to three principles a regulated buyer won't compromise on:

Validation & results

We validated it in a six-week pilot inside the production system of a Tier-1 global asset-management institution — running Echo Cancellation live but leaving items open, so human reviewers' decisions could be compared directly against the model. Every conflict between the two was investigated by hand.

20%

reduction in review effort

100%

precision — zero false negatives

130K

policy triggers analyzed

6 wks

production pilot, Tier-1 global bank

Not a single determination needed reversal — 100% precision, no false negatives. The client's compliance-technology leadership endorsed rolling it out to production immediately, and the capability shipped to general availability.