Leveling Up as a Data Scientist - Why Product Management Mindset is Essential
One of the management skills that senior data scientists ought to learn: product management.
Why? Think about this.
You've worked hard for weeks or even months on a really smart computer program that can analyze lots of data and make predictions. Your tech team loves it. But when you actually try to use it for a real-world problem, it doesn't help the people it's supposed to, or it costs more money than it's worth. In the end, your smart program is like a paperweight—interesting but not useful.
Then we run into a bottleneck. We realize there is a disconnection between us and business, between our work and the actual needs that business expects.
This takes product management.
In this article, I wanted to discuss why you can benefit a product centric mindset, and what aspects that you can focus on right away.
Why Should a Data Scientist Take a Product Mindset?
Business often wants us to build a model to effectively solve their problems. Behind this request, what they are really asking for is a useful solution with cost concerns.
Data scientists start to focus on model building. But often, what they are ignoring is the reason “why” they even work on it.
I've seen this happen too often working with client data teams. Teams can build amazing, technically sound models that miss the mark. They don't solve the user's problem, and they cost a ton of money. All because the team was more excited about building the model than figuring out why they were building it.
The issue isn’t that your model is bad. It's often that the model was the focus first before building the actual product. The team is technically capable, but given the limited time decided to build first and forgot to scope. Add in the limited timeline because the scope was non-existent, and the features to work on couldn’t be estimated.
Product management helps here. It's essential to guide the business towards practical solutions, not just technologically advanced ones. I've had to tell clients too many times that a simple ML model that answers their business needs is better than a complex one that's hard to keep up.
This doesn't mean you should neglect the technical dimensions of your role and morph into a full-fledged product manager.
But you should think about how your model will actually help the business. This will show that your model is useful and worth having. It also helps you figure out how much it will cost to use the model and whether it's worth the money and effort.
In the end, we need to know what to focus on to create a good ML model that’s a product the business can use.
Ask ‘Why’ First
As much as we data scientists don’t like to admit it, not all business problems need a machine learning solution. And not all things are novel. Simple heuristics work sometimes, and can even outperform models.
We need to start with the ‘why” behind the model — the reason for its existence. Good data products start with understanding the user’s needs. This helps us find out what is possible now, possible later, or even too complex to do with our current team and resources.
Scoping and planning meetings I’ve found are very useful for this. We moan about being stuck in endless meetings. But a 2 hour scoping and product feature meetings can find really quickly what you need.
If your scoping its important to focus on:
Critical Features: These are the must-haves that allow your model to provide the basic insights or predictions you're aiming for. For example, if you're building a weather forecasting model, accurately predicting temperature is a critical feature.
Future Features: These are nice-to-haves that can wait. For instance, adding a feature to predict the likelihood of hail could enhance a basic weather forecasting model, but it isn't essential for the model's initial version.
You will also run into these types too:
Unfeasible Features: These are ideas that are too hard to implement right now because they would need a lot of time, people, or other resources. For example, integrating real-time satellite data into your weather model could be awesome but might be too complicated and expensive to do at the moment.
Wishlist Features: These are features that would be amazing to have but are not priorities and could be considered luxuries. For example, adding a voice assistant to summarize the weather forecast from your model would be cool but isn't necessary for its main function.
Critical features will keep your model going. Future features can help improve the baseline ML product you are crafting. They can also demonstrate to the business you’re thinking beyond the immediate project.
As for unfeasible or wishlist features? Save them. You’d be shocked how many times they hint at another need the business may not realize.
While you’re understanding the why, focus on their needs. Not how you will solve them. Its easy to want to discuss potential solutions, but I’ve found that will often take the meeting off track.
Check How the Business Does It
Ask how business is currently doing a process without ML. That simple trick can save you weeks and months of frustration. One of the easiest ways is asking how the SMEs currently solve the problem.
Chances are, there's already a way people are handling the issue unless you're working on a brand-new project. It's a major advantage if you’re trying to take a product focus — you have users who know the product and the best features.
So, knowing the current process helps you to build a product, not just a model. It informs your decision on what features to prioritize, what risks to mitigate, and how to present an appealing business case. You can use the following strategies to deepen your insights:
Find the Existing Baseline: Utilizing the current process as a benchmark isn't just sensible, it's essential. For example, if SMEs are taking five hours to manually analyze customer churn, your ML model should aim to do it more efficiently and accurately.
Accelerate User Onboarding: Given that you already have users familiar with existing methods, design your ML model to be easily adoptable. User experience goes a long way to a great product. If SMEs are using spreadsheets for a task, make sure your model can interface easily with this format.
Zero in on Feature Gaps: When you know what's currently being done, you can identify the gaps and build features to fill them. If the current method is time-consuming for customer service agents, for instance, automation becomes a priority for your ML model.
Preempt and Minimize Risks: If you're informed about the current process, you can better identify and avoid pitfalls. If security is a concern in the existing system, you already know to prioritize it in your new model.
Make an Irresistible Business Case: Understanding the existing costs and inefficiencies can help you make a more compelling business case for your ML model. For example, if the current manual process costs $10,000 a month, highlight how your model can do it more efficiently for a one-time setup cost.
Prioritize High-Impact Additions: With insights into the current method, you can prioritize the features that will make the most difference. If real-time data is missing in the current process, that becomes a high-impact feature for your future ML model.
By aligning your model development strategy with what's already in place, you're not just creating a machine learning model—you're solving real-world problems in a way that's integrated and focused.
This positions you to create a real product. And a product that you can qualify the impact gets buy-in from both users and decision-makers.
Final Thoughts
Taking a product-focused approach to ML models isn’t a one-time event. It's a lasting commitment that demands constant refinement.
Start by evaluating your current models and look to ways the business is currently doing manually. Look for those with both strong business impact and technical feasibility.
Learning a product focus is hard. As the landscape changes with new discoveries and tech advances, your ML model will evolve. And so will your product mindset.
Don't just make a model—elevate it into a living, breathing product.
If you found this compelling, share it with your colleagues, spread the word, or engage with us. We're always eager to hear your valuable insights.
I’d love to hear your thoughts.
Want to learn more? Here are 3 ways I could help:
Whenever you ready, there are 3 ways I can help you:
1. Work with Us: Work with our team specializing in analytics, data, and AI. We provide a systematized approach that yields tangible business results.
2: Book a Meeting: Let’s get to know each other in a coffee chat.
3. Free Daily Insights: regularly share free content on Linkedin and X.