I’ve been working with two clients recently who are building recommender systems. (Think: recommending fitness classes based on people’s goals. Not a real client example, but you get the idea.)

The instinct is to jump straight to building: What data do we have? What model should we try?

I love moving fast and getting started prototyping, but there’s one step you can’t skip. You need to determine the placement: Where in the user journey will this recommendation appear?

Why? Because this completely changes what the user is trying to accomplish, and therefore what kind of recommender is appropriate.

The placement defines model requirements:

  Browsing/Exploring Deciding/Checking out
User goal Discovery, inspiration Confidence, commitment
Optimize for Novelty, diversity, engagement Relevance, conversion

I love using the Jobs-to-Be-Done (JTBD) framework here. The quick primer: A fast-food chain once studied milkshake sales by asking morning commuters: what job are you hiring this milkshake to do? They discovered the real “job” wasn’t satisfying a sweet tooth, it was making a boring commute more tolerable while being easy to consume in a car. That insight led to completely different product decisions than if they had just focused on making a “tastier” milkshake.

The same question is relevant when building AI features:

What job is the user trying to get done at that exact moment?

That answer does two critical things:

  1. It gives you the right success metric. Are they browsing? Deciding? Comparing? Committing? Each has a totally different definition of “good.”

  2. It tells you what your ML model should optimize for. Browsing/Exploring? → optimize for discovery, novelty, engagement Deciding/Checking out? → optimize for confidence/conversion

The placement in the journey determines what you build. The job comes first, and the model comes second.