🔍 What Are Recommended Points of Interest?
Every Sunday evening, we take all of your sessions from the past week and run something called unsupervised learning on them. In plain English, that means we’re letting the data speak for itself. Instead of us telling the system what to look for, it finds natural patterns on its own.
Here’s how it works:
- 🗺️ We create an embedding plot (basically a smart map of all your sessions).
- 🤝 We run clustering on that map to group similar behaviors together.
- ⏸️ We look at the clusters with the highest “hesitation scores” (moments where users pause, stumble, or seem unsure).
- 📝 Finally, we summarize those clusters for you so you can clearly see what’s going on.
Think of these “recommended points of interest” as little signposts 🚏 from your users.
They highlight:
- ⚠️ Where hesitation is spiking.
- 🔄 Which patterns are happening often enough to matter.
- ✨ Where the clusters suggest there’s an opportunity to smooth things out.
By paying attention to these patterns week over week, you’re not just reacting — you’re proactively steering your product in the right direction 🚀.
🏆 Why This Matters?
The beauty of unsupervised learning is that it’s unbiased. There are no pre-set assumptions, no filters, no one telling it what to find. It just surfaces what’s actually happening in your product.
For product managers, this is gold 🏆:
- 📊 You can see how often an issue pops up — not just that it happened once.
- 👥 You can see how many unique users are affected.
- 🧠 You avoid falling into the trap of availability bias (overweighting the feedback you remember most, instead of what’s truly common).
In short: it gives you a clearer, data-driven view of where friction really lives in your product 🔦.
⚙️ Adopting Recommended Points of Interest
- 📂 We create a default project where we run unsupervised learning on all of your sessions weekly.
- 🎯 You can create Discovery projects with filters to hone in unsupervised learning on specific product sections or user cohorts. See more here.