Recommenders learn from feedback. Ours cannot use most of it.
A film recommender improves because millions of People rate films, and a bad recommendation costs someone two hours. A Practitioner recommender has neither property. The feedback is sparse, it arrives slowly, and a bad recommendation can cost a Person their willingness to try Therapy again.
So the design question is not which model to use. It is which signal you are willing to learn from, because that choice decides what the system quietly optimises for over years.
The signals we ruled out, and why
- Session content. The richest signal available, and the one we will never touch. What a Person says in a Session is the reason the Session is private. A system that improves by reading it has already broken the thing it is trying to serve.
- Star ratings. Asking someone to rate a Practitioner a week in confuses two different things: whether the Practitioner is good, and whether the hour was comfortable. Good Therapy is frequently uncomfortable. Optimising for a five-star hour selects for Practitioners who avoid the hard conversation.
- Engagement. Time in app, messages sent, features touched. In consumer software these proxy for value. In Care they proxy for distress. A system rewarded for engagement learns to keep People in the product, and the goal here is the opposite: a Person who needed us and no longer does is the success case.
- Practitioner outcomes as a ranking input. Ranking Clinicians on symptom improvement sounds rigorous and creates a brutal incentive: avoid the hardest Clients. Any metric that punishes taking on severe cases will, given time, produce a Network that does not.
The signal worth testing
The Person came back for a second Session with the same Practitioner.
A second Session is evidence that the Person chose to continue. It is not proof of clinical fit, but it requires no rating, no survey and nothing from inside the room. Used carefully, it can help evaluate whether a shortlist led to a relationship the Person wanted to keep.
A switch can add context. If a Person leaves one Practitioner and continues with the next, the two choices say more about the original shortlist than a star rating alone.
What that choice costs
Return visits arrive slowly and miss the Person who quietly gives up. They can also reflect convenience: a Practitioner who is affordable and available may collect second Sessions for reasons that have little to do with fit. Price, schedule and Cover therefore belong in clear eligibility filters, while the remaining signals are evaluated separately.
That slower path is worth taking. A matching system that learns quickly from the wrong signal becomes confidently wrong, and the cost can be a Person deciding that Therapy is not for them.
How the learning is kept away from the Person
Our evaluation design excludes Session content and individual clinical history. It tests aggregate match patterns, such as the constraints shown and whether a Person chose to continue, before any such signal can influence matching.
Writing that boundary into the system before data accumulates matters. Once sensitive data exists, convenience can start to look like permission.
Iris narrows and explains. A Person chooses and can choose again. This post is a companion to how the shortlist works.
