Discussion about this post

User's avatar
Drew Harry's avatar

I have at points in my career argued exactly this point, but I have come to see some serious limits in this idea. The fundamental problem is that output accountability without decision-making authority is unstable.

Consider a simple case. A marketing DS manager reviews a launch plan for a product with a push messaging strategy that is extremely abusive; the median user will receive 10 push messages on launch day based on the message trigger logic and forecasted product adoption. She argues the case with her primary marketing partner, she argues the case with the PM, and everyone shrugs. "So what?" they say -- "we'll tune the messaging rules after launch if you can show that it's causing churn."

In the past I might have argued this was a failure "a truly effective analyst would have been able to convince people that this was an issue and to change their plan. If you failed to convince them, you need to consider how to take a different approach that might work better next time." However, I would now diagnose this as a problem with PM and GTM accountability. If they don't think an abusive use of the push notification system is a problem, there is no foundation for influence. No data scientist will win this argument because the PM is accountable for hitting her launch date and adoption targets. Someone else (maybe) is accountable for the push notification system unsub rate.

In other words, holding data staff accountable for outcomes only works in a case where their partners have good accountability models that are the same as the data scientist's. Maybe you have seen that more often than I have! But in my experience, there is constant tension between what a PM cares about and what a data scientist cares about, and that is a persistent barrier to an outcome accountability model.

I have found one useful bit of synthesis between these two perspectives. There is a spectrum of accountability between output and outcome, and the higher level someone is in an org, the more outcome-centric their accountability should be. So for an IC (anything less than principal IC) I prioritize output accountability because they don't own decisions about what to work on, or the decisions made by others that result from their work. For principal and IC managers, I expect outcomes BELOW their level reliably, and AT their level sometimes. In other words, a line manager data scientist should be able to convince IC PMs of important matters. But may or may not reliably get positive outcomes from peer PM managers. And they should only be accountable for securing good outcomes when influencing UP their own reporting chain, one step. But they should not be accountable for influencing "UP and ACROSS" the org. That is the job of someone above them in their own management chain who has the necessary context and peer relationships across the org to reset the accountability of other staff to an appropriate level.

Expand full comment
3 more comments...

No posts