Cast Study 2: Digital to Wet Lab (part 2)
In my last post, I described a situation in which your data team wanted your organization’s lab team to start using a predictive model to choose the different concentrations of compounds that they use in their experiments. In this post, I want to explore why this might be harder to do than it sounds. Then in the next two posts, I’ll explore two different potential approaches to a solution.
To save time, I’m going to assume you’re read the last post. As you’ll recall, when we started our story, your data team had already developed a predictive model, and even a UI that lab scientists can use. But while they’ve (hopefully) been talking to the bench scientists this project since early on, no users have actually had a chance to try it out and no one’s thought about how it will fit into the existing experiment workflow.
This is fairly common, and understandable: If you’re asking the lab team to make changes to a very expensive and complex experiment, you want to be pretty sure your model is accurate. But the side effect is that no matter how “agile” the process for developing the model was, you haven’t iterated on how this will fit into the lab teams’ workflows. So the process roll-out is going to be more waterfall than agile.
But since this is a hypothetical story, instead of talking about how we go forward from here, lets look at how we could’ve iterated on the process in parallel with the model.
The options for this will depend on an awkward question: What would you do if the model fails? What if it turns out to be worse than the way your lab team is doing it today? There are essentially two possible answers:
We’ll put the project on ice and look for some other low hanging fruit that we can address instead.
We’ll try a new model, and keep trying until we solve this because it’s an important problem that we need to address one way or the other.
Each of my next two posts will explore ways to iterate on the process for the two different answers. While you’re waiting, think about what you would do, and see if you agree with what I come up with.