Hamel Husain argues that the best way to get your team to care about evals is to do them yourself first, then show the results. He recommends presenting concrete metrics like reducing error rates from 30% to 5% in six weeks, rather than asking for buy-in upfront.
So the best way to champion evals is to lead with results. And what I mean by that is don't go into a meeting and say, hey, we need to do evals. You don't want to shove it down people's throats. No one's going to listen to that. What you want to do is show up to every meeting with some data and findings and metrics. So you want to be doing the evals yourself. You want to be doing error analysis and you want to be showing up to every meeting with, hey, these are the errors that I found. These are some errors that we fixed. And these are some automated ways that we're going to find that going forward. So your slide deck shouldn't say Eval 101. It should say, my XYZ product reduced their error rate from 30% to 5% in 6 weeks. Let me show you how.
