How to prioritise Fincrime features

One of a product manager’s core tasks is to prioritise: amongst competing problems to solve and among potential solutions. The RICE framework is one method for doing so. You assess the Reach and Impact of the project, adjust by your Confidence level, and divide by the Effort required to ship it.

However, this framework runs into some difficulties when thinking about features targeting Financial Crime (Fincrime). For example, identity verification, transaction monitoring, Enhanced Due Diligence, Anti-Money-Laundering. Defining Reach and Impact is challenging for these areas, since there are components that don't convert easily to a common denominator.

Fincrime products have three key metrics they affect: conversion rate[1], fraud rate, and operational costs. Operational costs and conversion both can be reduced to a dollar figure, but even this is a moving target: a company in the growth stage might be happy to throw money at operations in order to get more users, while a company trying to streamline operations will be the opposite. Fraud is the truly challenging arm: you can do some alchemy to come up with expected costs, but it is a bit of a hand-waving exercise. For example, how do you quantify reputational damage from widespread fraud? How much should you scale up known costs to account for fraud you didn’t catch? How should you account for the threat of potential fines from the regulator, that may not materialise?

My proposed solution: make the tradeoff between these metrics visible, and track whether it is improving over time.

Specifically, show the relationship between each of the three metrics, and aim to improve the cost of increasing one in terms of the other. For example, you can always improve conversion by accepting more fraud – the product manager’s goal is to decrease the amount of additional fraud needed to achieve a fixed amount of conversion improvement.

This is similar to the economics concept of the production frontier: you want to produce at a point on the frontier, and regardless of exactly where on the frontier you choose to be, pushing the frontier outwards is always better.

How can you visualise this? It’s slightly hard: there are three metrics, plus the time element. Here are the graphs I would have on a dashboard:

1. Show each metric over time

These three graphs (with totally fake data) demonstrate that conversion, fraud and operational costs seem to be trending up over time. Perhaps not exactly what we want, but it’s hard to tell without looking more closely at the relationships.

2. Show each metric vs each other metric, with aggregated time periods as data points

This shows a clearer picture of the trend. There is a positive relationship between fraud rate and conversion rate, but this relationship is getting weaker as time passes (the dots get darker for later months) – exactly what we hope to achieve.

3. Show a 3D graph of the three metrics with aggregated time periods as data points

(I have made a complete mess of the Python dependencies on my iMac, so this graph will need to wait until I have the patience to untangle it.)

Once you have these visualisations in place, the Fincrime conversation can centre on two questions:

  1. Where on the current frontier do we want to be, as a business?
  2. Are we shipping features that push the frontier outwards? i.e. features that improve the ratio at which we can trade off the variables?

This framework is an alternative way for companies to think about tradeoffs in Fincrime products, and should lead to better decisions and more constructive conversations between stakeholders.

[1] This might be the actual conversion rate in the case of features affecting the onboarding funnel, or it could be more like the ‘customers who can continue what they were doing’ conversion rate for features that affect customers who are already active users, like transaction monitoring.