Why good data scientists make good product managers (and why they’ll be a little uncomfortable)
When I was transitioning my career from data scientist to product manager, I solicited a lot of feedback from current data scientists and product managers about getting in touch with others who had attempted such a transition.
I was surprised by how often I heard some variant of “Hmm, I don’t know anyone who’s made this transition, and it seems a little odd to me.” I’ve always thought that the best data scientists are product-focused and have users and their needs in mind. Training models for the sake of training models isn’t really useful until they can be productized. To me, it seemed a perfectly natural transition.
I wasn’t discouraged, and I’d like to offer some perspective on what the transition is like for the benefit of others who may be thinking of either making this transition themselves or for hiring managers who are considering hiring a data scientist as a product manager.
Natural affinities between good data scientists and product managers…
Data scientists and product managers make decisions with data.
A major part of a data scientist’s job is choosing between competing options by identifying the relevant evaluation metrics, predicting the potential impact of a particular intervention, and communicating those results to stakeholders in a clear and concise fashion, pitched at the appropriate technical level.
Product management is no different. You’ll need to know “how will I know if my product is successful”, “how much of an impact do I think this new feature will have”, and you’ll need to communicate the why to both senior executives and junior engineers. Plus, data scientists already know SQL and can do their own quick analyses. They know not to overreact to variance in their metrics and know that experiments are a good remedy to reading tea leaves in a random walk.
Data scientists and product managers work cross-functionally.
Most data scientists are used to working across teams with colleagues in differing roles, from marketers to engineers to designers. They know they’ll need to think about things like how to serialize their models and how to surface the predictions of their models to users. They know they can’t just build a model and throw it over the wall to engineering to reimplement.
Product managers jump from writing and explaining acceptance criteria and specs to engineers to reporting on the performance of their products to working through wireframes and mockups with designers. They know they need to be able to be technical enough, business-oriented enough, and design-focused enough if they want to ship their product.
Data scientists and product managers choose an objective function and ruthlessly optimize for it.
Good data scientists know that optimization problems always involve tradeoffs. Want to maximize clickthrough rate for a particular part of your site? Prepare to see the clickthrough rate for another part drop off accordingly. Picking the right objective function is as important as finding the right features for your machine learning model and being explicit about the tradeoffs you’re making is the mark of a seasoned professional.
As a product manager, you’ll constantly need to be asking “does working on X help me achieve my goal of shipping Y?” Time, money, and people’s ability to work on things are finite. Every decision you make to work on something is a decision to not work on anything else.
… and some things that will make most data scientists uncomfortable.
Product managers often make decisions with incomplete or no data.
When you’re deciding as a PM to enter a new market, start work on a new feature, or many other things, you often won’t be able to have the data you’d be used to as a data scientist to make your decisions — but you’ll still have to make the decision. You can’t A/B test your solution because you can’t build two versions of the product or enter multiple markets.
Perhaps you’ll have some qualitative data, or some anecdotal data (I can feel the data scientists cringing at that phrasing), but you can’t wait to decide. This is hugely uncomfortable the first few times. The keys to managing this are: a) have a plan in place for collecting data ASAP, b) make predictions about what you think will happen if you are right, and c) be willing to admit that you were wrong and change course if things go south. The only thing worse than a bad decision is doubling down on a sunk cost.
Especially at earlier stage startups or with new products, you’ll likely either lack data or have very low quality data. Setting a path to collecting data and establishing a practice for reporting on this data can be a big first win for you as a new product manager.
I’m not going to take the easy out here and say you have to “trust your gut” but I will say that many decisions may end up being a coin flip. And that’s alright, because many decisions aren’t nearly as momentous as they feel at the time.
Product managers have very close relationships with design and UX
You’ll work very closely with your colleagues in UX and design as a product manager than you ever did as a data scientist. This requires a bit of an adjustment if you’re used to having very quantitative discussions. For some data scientists, this will require a real reckoning about how much they truly value qualitative research and user research. Minimizing these disciplines is a grave mistake, but I suspect many data scientists turned product managers will make it.
The simple solution is almost always the best choice
Data scientists turned PMs will be tempted to reach into their toolbox to apply machine learning to every problem that comes their way. They may or may not have the appropriate engineers and data scientists on their staff to even do so, much less make a reasonable decision about if this is a good idea or not. As a new product manager, I would urge you to be very, very sure that machine learning is an appropriate solution to your problem.
How far could you get with some simple heuristics or a change in the user experience? Probably pretty far. By the time you’ve spun up the infrastructure, collected the data, trained the models, and productionized them, your competition will probably have already taken the simple route. You must be comfortable saying no to machine learning except when it is warranted.
Hopefully I’ve convinced you that the leap from data scientist to product manager isn’t a huge one, for the right kind of data scientist. The career path for product managers is much better defined for data scientists, and I suspect we’ll see more people making this transition over the coming years. In the meantime, give that data scientist’s resume a chance when it lands on your desk.