MerchantRMS: fraud rate from 0.3% to 0.01%, review accuracy to 99.99%
MerchantRMS is a real-time fraud-prevention platform for card-not-present merchants. Over almost eight years I grew from Product Manager to Head of Product, leading a 40-person organization. We took merchant fraud rates from 0.3% to 0.01% and manual review accuracy to 99.99%, backed decisions with a chargeback guarantee, and packaged the tools for smaller merchants, growing revenue 40%.
- My roles
- Product Manager → Senior PM → Head of Product
- Company
- MerchantRMS Inc.
- Timeline
- Jan 2014 – Sep 2021
- Team
- 40-person organization, 8 in product
At a glance
The challenge
Card-not-present merchants need fraud decisions in milliseconds, accurate enough that manual review doesn't eat the margin, at a price smaller merchants can pay.
What I led
New detection capabilities, a faster and more accurate manual review operation, a simulator that let prospects test their own transactions, and tools packaged for every size of merchant.
Result
Fraud rate from 0.3% to 0.01%. Review accuracy from 90% to 99.99%. Manual review rate from 15% to 3%. Revenue up 40%.
Context
In card-not-present payments, the merchant carries the fraud risk. Every chargeback costs them the goods, the fee and standing with their payment processor. Decisions have to come back in milliseconds, and every transaction sent to manual review adds cost.
So a fraud platform has to do three things well: catch fraud, keep manual review small and accurate, and be priced so smaller merchants can buy it.
My role
I grew from Product Manager to Head of Product in four years.
-
Apr 2018 – Sep 2021
Head of Product
Led a 40-person organization, with an 8-person product team reporting to me. Made the build-vs-buy calls on the product's major tools, moving the company toward owning its core detection technology. Packaged the tools to sell stand-alone or bundled, which grew revenue 40%.
-
Mar 2015 – Mar 2018
Senior Product Manager
Defined and shipped new detection capabilities that took fraud rates from 0.3% to 0.01%. Raised manual review accuracy from 90% to 99.99% and cut review rates from 15% to 3%. Built the rule-engine simulator for prospects.
-
Jan 2014 – Feb 2015
Product Manager
Owned the network packet analyzer. Improved decision latency from 500ms to 350ms while adding new tools, holding a 99.9% uptime SLA.
Detection: fraud from 0.3% to 0.01%
I defined and shipped four new detection capabilities: a network packet analyzer, email reputation, user behavior analysis and fraud clusters. The packet analyzer, which I owned from my first role, examines how a transaction's packets travel across networks to reach the payment gateway and flags suspicious patterns.
Together they took merchant fraud rates from 0.3% to 0.01%, while decision latency improved from 500ms to 350ms at a 99.9% uptime SLA.
Manual review: faster, cheaper, more accurate
Transactions the models can't settle go to human reviewers. That's where cost and errors build up, so I added three tools to the review workflow: a reviewer-assistant bot, similar-transaction analysis and fraud retargeting.
| Metric | Before | After |
|---|---|---|
| Manual review accuracy | 90% | 99.99% |
| Manual review rate | 15% | 3% |
| Reviewer throughput | 20 per hour | 120 per hour |
| Operating costs across customers | — | −35% |
Turning capability into revenue
Better detection only matters if merchants buy it. Two changes moved the commercial numbers:
- A rule-engine simulator. Prospects could test their own transactions in minutes. The instant results increased sales closure by 60%.
- Packaging. As Head of Product, I packaged the tools so each could be sold stand-alone or bundled, with options for smaller merchants, and we backed decisions with a chargeback guarantee. Revenue grew 40%.
Key decisions
1. Own the core detection technology
I made the build-vs-buy calls on the product's major tools and moved the company toward owning its core detection technology.
Trade-off: building takes longer and costs more up front, in exchange for control over the capability the business is sold on.
2. Invest in manual review, not only the models
A reviewer-assistant bot, similar-transaction analysis and fraud retargeting took review accuracy to 99.99% and the review rate from 15% to 3%.
Trade-off: product time spent on internal tooling, in return for a 35% cut in operating costs across customers.
3. Let prospects test with their own data
The rule-engine simulator showed results on a prospect's own transactions in minutes. Sales closure rose 60%.
Trade-off: product effort on a sales tool, in exchange for a much shorter path from demo to decision.
How I measured it
Fraud, review operations and revenue each had their own numbers.
| Metric | Before | After |
|---|---|---|
| Merchant fraud rate | 0.3% | 0.01% |
| Decision latency (99.9% uptime SLA) | 500ms | 350ms |
| Manual review accuracy | 90% | 99.99% |
| Manual review rate | 15% | 3% |
| Sales closure (with simulator) | — | +60% |
| Revenue (packaging) | — | +40% |
What the fraud rate means for one merchant
Take a merchant processing 10,000 transactions a month, as an illustration.
At 0.3%: 30 fraudulent transactions a month. At 0.01%: 1.
Each fraudulent transaction avoided saves the merchant the lost goods, the chargeback fee and the hit to their standing with the processor.
What I'd do now
Lessons
A fraud product is the model, the review operation, the latency and the price together. Miss one and the others don't matter.
Packaging is a product decision. Selling the tools stand-alone or bundled grew revenue 40%.
Let buyers see results on their own data. The simulator moved sales closure more than any pitch.
Looking for a leader who owns detection, operations and pricing together?
That combination is what moved the fraud rate, the cost base and the revenue here.