Models Suggest, Humans Decide

Claire Monroe

Welcome back to Science of Leading, I'm Claire Monroe with Edwin Carrington. Edwin, we've spent two episodes making the case for data in people decisions. Today I want to push on the worry that comes up every time this topic gets raised, that all of this quietly dehumanizes the process.

Edwin Carrington

It's a legitimate worry, Claire, and the answer isn't to retreat from data. It's to be precise about what data is actually for. A model suggests. A human decides. The moment that order flips, you've stopped being data driven and started being data blind, just with better graphics.

Claire Monroe

Can you make that distinction concrete? Because "models suggest, humans decide" sounds good as a phrase but I want to know what it looks like in an actual room.

Edwin Carrington

Take a predictive flight-risk model flagging an employee. The model's doing exactly what it should, surfacing a pattern a manager might miss. But if that manager happens to know this employee just received a life-changing personal opportunity, something no dataset captured, that context has to be allowed to override the flag. The model did its job. The human still has to finish it.

Claire Monroe

There's a promotion example in the piece that I think captures this even more sharply, two people, one clear data-favored choice, and it doesn't go the simple way.

Edwin Carrington

An assessment strongly favored promoting one person, Person A, based on the behavioral and performance data. But Person B held critical domain expertise the data couldn't fully weigh, and was navigating a family health situation at the time. The resolution wasn't picking a side and ignoring the other. It was a phased promotion plan that honored both realities, informed by the data, but not enslaved to it.

Claire Monroe

That's such a specific example, I want to flag for anyone listening that it's illustrative, the piece doesn't name a real company here. But the underlying lesson still holds regardless.

Edwin Carrington

It does, because the lesson isn't about that specific pair of people, it's about the shape of the decision. Data should analyze. It shouldn't dictate. The moment leadership treats an assessment score as a verdict instead of an input, you've built a system that's technically sophisticated and practically worse than the gut-feel process it replaced.

Claire Monroe

So what actually keeps that line intact, practically, inside a real HR process? Because "use judgment" isn't a policy, it's a hope.

Edwin Carrington

Real safeguards, not just good intentions. Diverse review panels, so no single perspective quietly becomes the default filter. Fairness checks on the models themselves, specifically to catch algorithmic bias before it compounds at scale. And clear documentation of the actual decision criteria, so a choice can be explained and defended after the fact, not just felt to be reasonable in the moment.

Claire Monroe

Documentation feels like the piece people skip most, because it's paperwork, right up until someone needs to explain a decision six months later and can't reconstruct why it was made.

Edwin Carrington

Exactly, and that's precisely when the absence of documentation turns into a real liability, legally and organizationally. The habit of writing down why has to exist before you need it, not after.

Claire Monroe

So if someone's listening and worried that getting more data driven means getting less human, what's the actual reassurance?

Edwin Carrington

The reassurance is that the goal was never to remove the manager from the room. It was to give that manager better evidence to bring into it, and the discipline to notice when the evidence and their own judgment are pointing in different directions. That tension is the system working, not the system failing.

Claire Monroe

If you want a real behavioral starting point to bring into your own next people decision, OAD's free Top Performer Profile is about seven minutes, results in twenty-four hours, no credit card. That's O-A-D dot A-I.

Edwin Carrington

Let the data speak first. Just don't let it have the only vote.

Claire Monroe

Perfect place to close out this series. Thanks, Edwin.

Edwin Carrington

Always, Claire.