I'm here with Casey Lichtendahl. Casey is a colleague of mine, a faculty member at Darden, focused on data science and analytics, and he is currently on leave at Google as a visiting researcher. Casey, thanks for joining us. Thanks for having me, Alex. You've taught a lot of general managers, a lot of MBA's about data science. What does that allow them to do that's important? Well, it allows them to understand what a data scientist does on a day-to-day basis, what models they use, what tools, techniques are in their bag of tricks, and they can better work with those people as they try to deliver value for the business. If you know a little bit about what a data scientist can do, maybe you know when to go ask them to help you unlock some opportunity. There are folks around here at Google, I've heard a manager in a meeting say, "I don't get up out of bed," and he's speaking for his team really when he doesn't find an opportunity that's going to add value of more than a $150 million. Of course, this is a Google scale. But in your own organization, you probably have some threshold for which you shouldn't get out of bed, and you shouldn't ask your data science team to work on projects that don't add value above that threshold. What are the things that a general manager can do with a data scientist to help that data scientist do better work? Well, if they know a little bit about the models, and the tools, and the techniques, and they've worked through the process, the data science process before, from ingestion of data into some kind of database, and then querying that database to get the data into some model that would be used to make some predictions. Then they take the predictions and bring that into some environment where they can make decisions from those predictions. If they've walked through that whole process, and they've really thought about the whole thing, end-to-end, and every step in between, especially the front, around issues of a data integrity. Do we have the right data to measuring the right stuff? Do we have any anomalies or do we have any just flat out errors in our data sources? If they've taken those steps, they can help the data science team tailor that whole process to the opportunity. If the opportunity is really big and it calls for a complicated model, then go at it. Spend a lot of time, build a complicated model out, and chase that big dollar opportunity. If it's a smaller dollar opportunity and yet still worth the resources of some data scientists, well, then maybe you scale back on the ambition of the model that gets built and the pipeline that supports that model. These things can all be dialed up and down, and a manager should know what's possible in the field. One of the things that you've taught me a lot about that I've found extremely useful personally is, when we're doing analytics, we need to do a good job of framing the DV, the dependent variable, the output variable to makes sure that it's actionable. Can you talk a little bit about how a manager can both better charter and better facilitate the work of a data scientist for actionability to have that impact whatever the dollar value is? Yeah, the dependent variable, the thing we're measuring and the thing we're trying to effect is simply the most important thing. It's what probably translates more closely into dollars added in terms of value for the business than anything else, you should seek out that dependent variable, and it [inaudible] a lot. I've come across it a lot in the six months I've been here at Google. Just a quick example from the capacity planning space. You can have a dependent variable where you're trying to track the demand of your customers and that can be average demand over the day. Maybe you have a minute-by-minute observations and you want to aggregate up to the daily level and try to forecast where daily demand is headed. One way to do that is with the average minute-by-minute demand. Another way to do that might be to look at the math. What was the peak minute demand during that day? That little small decision really matters for capacity planning because you're not trying to plan to the average. You're actually trying to plan to cover the most amount of demand the customer would call upon you to deliver in any one minute during the day. So that's just a little simple example of how important this dependent variable decision can be. Are there any other examples that you think would be helpful for our learners about those dependent variables? I just think that that one is so interesting and so central to a lot of the things that I've seen teams struggle with in this area. Yeah, well, I think another one is measuring demand in terms of units, and that's great. But if you're making different amounts of money off of the different products or the different units that are in your population, well, then maybe that dependent variable units is not such a good one. Because what really matters for the business would be the margin dollars you're driving and maybe that should be your dependent variable. Then you're getting super close to adding value to the business over time. It wouldn't be the kind of net present value calculation that you might like to put together if you were in finance to think about margin dollars not just today, but out into the future, year after, year after, year. But you're getting closer to the idea that you're adding value to the business by measuring something like margin dollars as opposed to units. Those are some great perspectives on the practice of data science from Casey Lichtendahl. Casey, thanks for joining us. Thank you, Alex.
