Originally posted as a Twitter thread on November 03, 2021

Many areas of financial services have “stochastic margins” per widget, but hopefully (obviously!) positive margins for the whole batch of widgets sold – unlike most manufacturers, with fixed/declining COGS at scale. This means many things when you build a “financial” business

You might make or lose money on the marginal loan, marginal insurance policy, marginal payment processed, marginal market-making trade. Apple makes the same margin on every iPhone 13 Pro it sells.

In no particular order, here are some things to think about:

Understanding adverse selection v positive selection is crucial. And the “default” (when you are providing pure MONEY) is being *overwhelmed* with adverse selection before the positive selection customers even show up. Bad loans, risky insurance, tainted properties, etc.

ONE FORM of adverse v positive selection is pull v push. In most businesses organic consumer adoption is a godsend. In risk, it is not! Compare people searching for “I need a loan” to people who get “pushed” a loan offer based on their highly desirable, prescreened credit

This is one of many reasons why postal mail works so well (surprisingly!) for lenders. It’s not just the saturation of the channel – it’s push v pull, picking your customers vs trying to pick through the sea of (possibly) adverse selection applicants

So tracking the **channel** against the cohort performance is *crucial* to understand which channels tend to have more of this adverse dynamic, even holding things like credit score or underwriting risk constant. Cohort customers by time AND channel (and other behaviors)

Be highly, highly tuned to “anomalously high” conversion rates. It might mean that you found a great channel, or it might mean you found a motherlode of desperate/bad/fake customers

A life insurance executive once told me that they found that post-midnight advertising on the History Channel was their most cost effective channel, but turned out to be netting very bad, depressed customers — which didn’t show up in medical underwriting but tracked to channel

Another form of adverse selection happens when you are buying/underwriting a subset of financial products without seeing the full set. Think lenders who sell SOME of their loans. The offered products are ipso facto riskier or worse than the retained ones.

The tail REALLY matters – cohorts need to season. If you are selling life insurance, you don’t know if you’re good or bad at underwriting until (sadly) people die. If you’re buying homes…until the VERY last homes sell.

A mistake I see many companies make is they record their early realized gains, and either assume that will continue or (equally bad) hold everything else in the tail at cost. The last trades are almost always the worst — that’s why it took so long to unload them

Promotions are their own form of adverse selection in “deal seekers.” PayPal once ran a giant promo (trying to launch their offline biz) with HomeDepot, and saw massive customer adoption – but all from SlickDeals deal seekers who saw opportunity for profit

so:
-let cohorts season before assuming anything
-understand channels
-always watch for adverse selection
-be vigilant watching for anomalously HIGH conversion rates
-push can outperform pull
-beware underwriting “subsets” of a customer’s business
-beware deal-seeking