Resy · Concept Case Study 2026

focus

Product design, Group payments, Privacy

Timeline

2026

Tools

Figma, React, Claude

the problem

Everyone said yes to dinner. So why are you still chasing people?

I designed a Resy group-booking flow where guests cover their own deposits and the organizer can see when the table is ready to book.

2× Speed
−10s
0:00 / 0:00

Where group bookings get stuck

To book a table that works for everyone, the organizer has to ask what people can spend, relay dietary requests, and sometimes put down money on the group’s behalf.

research

One organizer always burdens a group reservation.
How can Resy make booking easier, friendlier, and more intuitive to all guests.

Resy

Built for small parties. Large groups are usually sent to the restaurant by phone, email or an events form.

Gap: the host leaves the app

OpenTable

Private dining and large-party requests start as inquiries, and the details move to email.

Gap: no shared plan for guests

Partiful

Ticket sales and the 'Chip In' feature collect money for events. Sends guests to outside payment apps and can’t confirm who paid.

Gap: money without the reservation

The gap is responsibility: a yes in a group chat costs the guest nothing, while the organizer fronts the deposit and the restaurant holds the table.

product scope

How might Resy let guests take responsibility for their own part of a group reservation while giving the organizer enough certainty to book?

Each person in the booking gets their own view:

Organizer

Find a restaurant the group agrees on, see how many seats are confirmed and what’s holding up the reservation, and decide when to book.

Never sees guests’ dietary notes or budgets

Guests

Share their budget privately, pay their own share, and send dietary needs directly to the restaurant.

Details private by default

Restaurant

Gets the headcount, the deposits and each guest’s note, and answers individual guest questions.

Answers go back privately

design decisions

Three decisions about who sees what, and who decides.

01 / Private by default.

2× Speed
−10s
0:00 / 0:00

Dietary needs and special requests go straight to the restaurant, and budgets only shape suggestions. If the restaurant can’t accommodate someone, that guest chooses what the organizer hears.

02 / Covered isn’t the same as booked

Covered isn’t the same as ready

The organizer’s view separates said yes, seats covered and ready to book. Only uncovered seats block booking; an open menu question doesn’t. If there's something blocking the booking, the agent tells the organizer what to do next.

How the deposit works, in one example.


Harborline asks for $90 to hold a table for 5, so each seat is $18. When a guest pays, their $18 is held on their card, not charged. Nothing is charged until the table is booked, and it’s all refundable until Thu., Nov. 19 at 5 PM. Three guests have paid, so $54 is held. Before the 5 PM hold ends, the organizer can:

Cover the last 2 seats

$36 held on her card

Jordan and Sam can pay her back. Booked for 5.

Ask for more time

Nothing held or charged

The 5 PM hold stays unless Harborline agrees.

Book for 3

Each seat becomes $30

The $90 deposit stays. Paid guests keep their $18 held and each approves $12 more first.

03 / Every option shows who it affects.

Every option shows who it affects

Before the 5 PM hold ends, the organizer can cover the last seats, ask for more time or book for fewer. Each option says what happens to them, to each guest and to the restaurant, so nobody finds out after the fact.

flow & edge cases

When everything goes right…

User flow: organizer, guest and restaurant, from invite to booked

and when something changes… Integrating agent-models.

The agent has one job: when something changes. My first version had AI plan the whole dinner. It made the agent the hero and skipped the real problem: the gap between a yes and a paid seat. So I narrowed it to the moment plans change. It reads the restaurant’s terms, works out the new share and drafts a request. The organizer decides whether to send it, and every paid guest approves any change to their share.


Here, a party of 5 drops to 3 after two guests cancel:

The agent has one job: when something changes. My first version had AI plan the whole dinner. It made the agent the hero and skipped the real problem: the gap between a yes and a paid seat. So I narrowed it to the moment plans change. It reads the restaurant’s terms, works out the new share and drafts a request. The organizer decides whether to send it, and every paid guest approves any change to their share.


Here, a party of 5 drops to 3 after two guests cancel:

User flow: two guests drop out, the restaurant accepts, guests approve, booked

Edge cases the design has to handle:

What I prototyped: the agent reading Harborline’s terms, working out the $30 share and drafting the request, the organizer sending it, and guests approving. What I’d build next: testing it against real restaurant terms.

A guest never answers

Their seat shows as not covered. The organizer can cover it, ask for more time or book for fewer.

Nobody is chased in the group chat

A payment fails

Only that guest is asked to retry. Their seat stays held until the deadline.

No one else pays more

The restaurant can’t accommodate a guest

The guest already chose what the organizer hears: that they can’t come, the reason, or nothing.

Privacy holds when it fails

The hold runs out

The organizer gets the decision before 5 PM, not after. If nothing is chosen, the table is released and nobody is charged.

No one pays for a lost hold

testing

How I’d test it. What I would measure:

Covered vs. said yes

Do organizers book sooner when they see covered seats instead of RSVPs?

Measure: time from first yes to booked

Say yes first, pay later

Do more guests pay once they see the cost and the deposit, compared with paying when they RSVP?

Measure: RSVP-to-paid rate

Private by default

Do guests share dietary needs more often when only the restaurant sees them?

Measure: notes sent per booking

What changed while I designed it:

Revision 1

From pay-first to RSVP-first. Guests first had to pay to say yes. Now they RSVP with a private budget and decide on money only once the group picks a spot and they know the cost.

Revision 2

From minimum eight for this feature -> any group. I first designed for a big booking, like a venue. However, the same gap shows up at three people, so nothing in the flow assumes a large party.

Revision 3

Privacy protects the organizer too. Keeping budgets and allergies private means the organizer never has to ask about them in front of everyone.

next steps

Where Resy earns from it, and what I’d build next:

Deposit processing

A small fee on deposits collected through Resy. Especially relevant for bigger bookings and larger deposit sums made directly in-app.

Ex. 3% of $60 deposit * 12 people = $21.60

Bigger tables, booked in app

Group dinners that start in a group chat end as Resy bookings.

Value to restaurants

Fewer empty seats

A deposit per seat cuts no-shows on group bookings.

Value to restaurants

What I’d build next!

Restaurant side

Design Restaurant’s view: headcount, deposits and dietary notes in one place.

Test the agent on real changes

I prototyped the agent for one moment: guests dropping out. Next I’d test it with real restaurant terms and measure how often organizers send its drafts without edits.

© 2026 April Kwon

Play with the stickers!

© 2026 April Kwon

Play with the stickers!

© 2026 April Kwon

Play with the stickers!