If you saw this month's video, you know the setup. You're at a rescue, thirty dogs in the corridor, one decision, no take-backs. The rule says walk past the first 37%, then take the first dog that beats everything you've seen so far. Do that and your odds of going home with the actual best dog jump from one-in-thirty to about four-in-ten.

Mathematicians call this the Secretary Problem, and 1/e (roughly 37%) is the threshold they’ve come up with.

I want to use the rest of this issue to do two things the video couldn't fit.

First, to be honest about where the Secretary Problem maps onto pharma and where it doesn't.

Second, to talk about the behavioural failure modes that show up when pharma teams know the maths and still get it wrong.

I deliberately mentioned independent data monitoring committees (IDMCs) in the video. They're the most familiar example pharma has of a pre-committed stopping rule, and they hint at the right intuition: at some point you've seen enough.

But the maths is slightly different. The Secretary Problem is about picking one winner from a sequence of candidates you can't revisit, where ordering is random and the only thing that counts is whether you pick number one.

A trial interim analysis doesn’t follow this logic. It evaluates accumulating evidence on a single candidate using sequential testing, alpha-spending functions, conditional power, and O'Brien-Fleming¹ or Pocock² boundaries. Different problem, different statistics. The 37% threshold does not show up anywhere in a well-designed interim look, and you wouldn't want it to.

Where the 37% logic does map onto pharma is in the candidate-selection problems your organisation runs every week. In-licensing diligence. CRO evaluation. Target prioritisation across a portfolio. Site selection. Hiring. These look much more like the puppy corridor than a Phase 3 readout does: sequential, comparative, hard to revisit, and judged on relative ranking rather than statistical significance.

So treat the rest of this issue as being about those decisions, not about your interim analyses. The interim analysis machinery is its own field, and most well-run trials get it right.

What an IDMC has, and what almost no internal decision forum has, is a rule written down before the data arrives. The committee didn't pick the molecule. It didn't name the program. It has no career upside in either direction. That posture is why its stopping decisions don't get rationalised away the way internal ones do.

The Secretary Problem works for the same underlying reason. You commit to a calibration window before you see candidate one. You define what "beats the field" means in advance. Then you follow the rule.

In most candidate-selection decisions in pharma, that order gets reversed. The rule arrives after the data has already started shaping intuition, which means it isn't really a rule, it's a justification. Once the justification is doing the work, four predictable things go wrong.

Sunk-cost stickiness. A Phase 2 program misses its pre-specified bar. Not catastrophically, but clearly. The conversation in the room is never "the rule said stop." It's "we've spent eighty million dollars, the mechanism still looks interesting, let's run one more cohort." Three years ago someone named this puppy. Walking past it now feels like betraying that earlier decision, rather than honouring the rule that was set alongside it.

Anchoring on the first signal. A promising biomarker in an early translational study quietly becomes the implicit benchmark for everything that follows. Calibration stops after candidate one, which is the exact failure the 37% rule was designed to prevent. In the rescue, you'd take the first dog because she wagged her tail. In the pharma version, a platform commits to a target because one early knockout study read clean.

Loss aversion in go/no-go. Killing a program produces a visible loss. The better program you might have funded instead is invisible, because it doesn't exist yet. Kahneman put a number on the asymmetry: losses register at roughly twice the magnitude of equivalent gains. The result is that marginal assets stay alive past their futility boundaries because no one wants to be the person who killed them. An IDMC doesn't carry this weight, because it doesn't own the asset. A portfolio committee almost always does.

Winner's curse in BD. This one really is the puppy problem on the nose. You're evaluating in-licensing opportunities. Asset number two clears your criteria. You bid, and you win the auction, which by construction means you valued it more than anyone else who saw the same data package. The Secretary Problem would have told you to keep looking, because "first acceptable" is not the same thing as "best." The cost of one more diligence cycle is almost always lower than the cost of overpaying for the wrong asset.

None of this is a suggestion to apply O'Brien-Fleming to your hiring funnel. The maths doesn't carry across. But the discipline does.

A few things worth borrowing from how trials are run and pushing into the rest of your decision-making:

Write the rule before you start looking. Whether it's how many BD targets you'll review before bidding, or how many candidates you'll interview before extending an offer, that number should be on paper before you see candidate one.

Define your "walk past" criteria in writing. What would cause you to pass on the next option even if it looked great? If you can't articulate that before seeing the option, you'll rationalise after.

Separate the rule-setter from the decision-maker where you can. IDMCs are independent for a reason. Hiring committees beat solo hiring managers for the same reason. Your BD lead probably shouldn't also be your deal champion.

Filing the 37% rule under "statistics" misses the point. The interesting part is the discipline it imposes: a stopping criterion that exists before the first candidate walks in.

Pharma writes that kind of discipline into trial protocols better than any other industry. The question is what happens to it once you step outside the protocol, into BD, into licensing, into the portfolio review where someone has been working on a molecule for four years and is sitting across the table from you.

That's where the puppy you already named tends to win.

If you want the maths explained in simple terms, with a slightly aggressive amount of dog footage, the full video is on YouTube. 🐩

Thanks for reading Pharma Radar. If this resonated, forward it to whoever on your team needs to walk past a puppy this quarter.

P.S. — Pharma Radar is written by Mirko von Hein. I help pharma and biotech teams navigate HTA submissions, cost-effectiveness modelling, and market access strategy across the UK, Ireland, and Germany. After a decade across IQVIA, Parexel, and Gilead, I'm now taking on selected engagements through Von Hein Consulting.