CHRIS STEVENS: Data scientists do a great job keeping it honest on the back-end. Understanding causal interaction is so important. You need discipline about the method by which you gather the data; you need to be disciplined about the way you set things up to get that causal stuff at the backend; and then you need to be disciplined about interpretation of results.
We’ve had a lot of interesting dialogues internally about the extent to which we can interpret results for business conclusions. That sort of “data cop” is one role the data scientists do.
What are the other roles?
On the other side, there’s data exploration. You take a couple terabytes worth of data and try to digest it. That’s an exercise unto itself and it involves considerable effort to distill the data into something that people can explore.
Their expertise does a really good job in terms of understanding how to visualize or explore that information. Very often in the very beginning and in the very end, the data scientists are critical, and business intuition can be misleading.
How so? Where do the problems mostly come from?
Common business dialogue is influenced by opinions from various parties representing various interests. But interests are not facts, they’re objectives. If you think your business interests are always aligned with media sellers then you probably also think your real estate agent is doing you a favor.
It’s incredibly important to be skeptical about any pitch related to data science; everybody’s got an algorithm, everybody’s got the “best” algorithm. The differences between them are often fleeting.
How do you define “intuition” when it comes to evaluating “the best algorithm?”
I would interpret it as skepticism, but other people would probably look at it in a different way. There are always business decisions that have to be made in the end too. Data science guides us and gives us a lot of insight, but it doesn’t tell us everything. We still have to make judgment calls.
The judgment is really what you're paying for in terms of management. Our managers have to make decisions in the absence of perfect information all the time.
Orbitz has a lot of first party data. Do you suffer from overflow?
We are rich with data. All e-commerce providers are rich with data. But not all of them share the same level of sophistication with respect to how they use data. Our organization is certainly getting better about data and data science. We’ve got a bunch of good people.
Is it a matter of the process or the tools evolving?
I would use the creative process as an analogy. It’s very common to get sets of creative back from a creative department, and if you ask various people which one’s going to work the best, you’ll get several very passionate opinions, most of which are wrong.
Because they’re subjective?
We always have latent bias in all of our thinking, right? It’s very difficult to predict that kind of thing. That doesn’t mean that the creative process isn’t valuable, in fact, having good creative is hugely valuable. With more machine learning, you needed more creative because the algorithms do better when they have greater diversity from which to choose.
The interplay there is, the roles may change slightly, but if anything in this case, it amplifies the objectivity of decision-making and amplifies demand for “actual” talent vs. the most impressive, and subjective, opinion. I think that that interplay will continue. It will change the way people interact surely.
Is it a matter of having less or more data scientists mixed with more creative problem solvers – those who lean more towards “intuition” – to help guide you on that? Is there a “too many cooks in the marketing department” issue?
Corporate America always has a lot of cooks. I rely on our data scientists quite a bit and I suspect I’ll rely on them more in the future.