Designing AI Matching for Reservation Services

As more customers book alone, reservation services increasingly need a matching feature that connects strangers to each other. Golf "join" bookings are a good example. Success depends less on the matching algorithm itself, and more on how that algorithm is layered onto the product.

What should the matching criteria be?

Simple rule-based matching — "group people whose time slots line up" — is easy to build but leaves users unsatisfied. Combining several criteria (skill level, personality, past usage patterns) is what actually produces "someone I'd want to play with." But more criteria means fewer possible combinations, so you need a clear priority order among them.

How do you handle a failed match?

The part most often left out of a matching algorithm's design is what happens when a match fails.

• The longer someone sits in a queue, the more likely they are to churn.
• You need re-matching logic that relaxes some conditions and tries again.
• The relaxation rules (e.g., widen the skill range, adjust group size) need to be designed up front — not improvised during operations.

Explainability builds more trust than the match result itself

A match you can't explain is hard to trust, even when the outcome is good.

Showing users even a short rationale for a match (shared interests, similar skill level, etc.) has a big effect on satisfaction. A black-box match that only outputs a result leaves you with no basis to respond when a user pushes back.

Why you need to design this alongside settlement

Payment and settlement almost always follow a match. If you build matching logic and settlement logic separately, exception cases — a cancelled match, a change in group size — tend to throw settlement out of sync. You need to map out the settlement flow from the earliest stage of matching design.

ONEON, built by Nora, was designed with join matching based on time, skill, and personality alongside automatic settlement, and Nora continues to operate it today. The same approach applies beyond golf to other reservation-based services — see Golf ERP & Reservation Platforms for details.