Carla Saavedra Kochalski on building strong PM teams, and scaling B2C ML-driven products
Product State Q&A
Carla Saavedra Kochalski is a Sr. Product Director of Personalization at Capital One, leading ML-driven experiences like the Eno Intelligent Assistant. She formerly held roles at Tourneau and Samsung.
EC: What’s the key to building consumer ML products at scale?
CSK: The first step in building an ML product is no different from building any other product: You must start with the customer problem and start with what data you have. But because ML is data-heavy and expensive, the next question to ask yourself is whether you need ML to solve your problem at scale.
Some good use cases for ML include:
Predicting something for the customer - Predicting the top five questions the customer may ask when they click on the feature in the app
Making a recommendation to the customer - Recommending the best products for them based on their previous purchase history
Looking for common patterns - Understanding a customer’s intent through common language patterns to retrieve the best answer
Narrowing in on key customer problems — and then testing your hypotheses within narrow use cases — really helps you understand whether you’ve achieved product/market fit, and can scale.
Just know that scaling ML products can take years — depending on how narrow or broad the use cases are — and how good your model is. I think the biggest thing to know is that it’s currently a much longer journey to achieve success with ML products.
Other things to consider when building ML products at scale include data privacy and integrity (these should be thought of on day one — how to protect your customers), risk management, and full instrumentation of an experience to collect additional data you need (don’t just launch a thing and then make your poor Data Scientists and Analysts retrofit a model based on what is out there).
EC: How do you build a team of strong PMs at one of the world’s largest banks?
CSK: The best product managers here really understand the balance between solving problems for our end user, the business, and how to do it in a well-managed way.
We are in a highly regulated industry and need to maintain customer trust, so the first thing I always tell a new product manager is to become close with your Risk, Compliance, and Legal partners and bring them in early.
They will not only ensure you’re well managed but also make your product better.
The other thing I emphasize to my team is that diverse, cross-functional teams make better products — so don’t think you know everything.
Lean on your engineers, data scientists, designers, researchers, analysts, legal and risk teams to help understand the problems and how to best go after them.
EC: What advice would you share with customer care, CX, and marketing pros who want to move into product management?
CSK: First off, I have a journalism background, so I can confidently say that there’s not one path to product management. There are actually a ton of similarities between being a reporter and product manager:
You are tackling something new every day, you need to work with a ton of experts in the field to understand the core of your story/problem, you need to break down super complicated stuff into a simple understandable thing, and you need to convince people why they should care.
The key to being a product manager is being curious, asking questions, understanding the core problem, and translating that into how you might fix that problem.
I currently have a product apprentice on my team who spent years in customer service, so she knows the customer really well, which is a big first step.
Now she is learning how to translate those customer problems into hypotheses she can test and build with her engineering and design teams. A lot of people, especially women and people of color, are intimidated by product because they feel like they need to be super technical or business savvy or some sort of unicorn, but that’s not the case.
If you love looking at problems and how to solve them, there’s a place for you in product. That’s my number one career mission.
“Scaling ML products can take years — depending on how narrow or broad the use cases are, and how good your model is.”
Thanks for reading Product State! Subscribe for free to receive new posts and support this work.