Back to work

Loyalty app for a hospitality group · 2026

AI-assisted Product Discovery

A hospitality group wanted to connect several restaurant brands through one loyalty programme.

I led the discovery and initial product direction, using two AI-assisted experiments to challenge and strengthen the concept. Claude worked from the project research, strategy and constraints, while I evaluated what was useful and what changed.

Client details have been anonymised. The process and decisions haven’t.

Project overview

Role

Senior Product Designer

{{ m.k }}

{{ m.v }}

My AI-assisted workflow

  1. {{ w.n }}

    {{ w.title }}

    {{ w.body }}

    2.
Using Claude to synthesise the project evidence, then reviewing the outputs against research, strategy and technical constraints.

Discovery

I brought the project material and my research into a dedicated Claude project, using it to organise the evidence and surface recurring needs and opportunities.

I then reviewed the outputs against the source material and with the wider project team, using that judgement to shape the design principles.

Product opportunities

From the research, AI-assisted synthesis and team input, I defined five principles to guide the concept.

  • {{ op.n }}

    {{ op.title }}

    •

Together, they grounded the concept in customer needs, business goals and technical reality.

Starting concept

From those principles and the requirements established with the team, I created the starting concept: shared membership, cross-brand discovery and distinct brand experiences.

Original membership home wireframe: member overview with points, tier progress, next reward, benefits and ecosystem brands. Original brand page wireframe: Wave masthead, location, menu, booking, events, earning promotion and a make a booking call to action.
My starting wireframes: membership home and individual brand experience. Brand names are fictionalised throughout.

AI exploration

Experiment 1 — Challenge the concept

Claude had the same research, principles and technical constraints I had worked from, but my wireframes were deliberately withheld so the exploration wasn’t anchored to my solution. It worked through Claude Code and Figma MCP to explore alternative structures.

The useful finding was that I had been treating two states as one: a true cold start with no customer context, and an experience where some context was already known. It also suggested a lightweight optional prompt — “What brings you here today?” — to create useful context without guessing.

Experiment 1 cold start exploration: three concepts covering guided orientation, single-path first action and progressive disclosure onboarding, each annotated with the principle tested, assumptions and risks.
Cold start: three structures, each annotated with the principle it tested and the data it would depend on.

My evaluation

Evaluating the exploration

I checked each suggestion against the customer evidence, the strategy, the business need and what was technically feasible. Two ideas from the same exploration went opposite ways.

Carry forward

Lightweight optional preference prompt.

It creates a real, customer-supplied signal without pretending we already know the customer.

Not carrying forward

Inferring a starting brand from acquisition data.

The available project evidence did not establish that this data was reliable enough to design around.

Earn context rather than guess it.

AI critique

Experiment 2 — Critique my concept

I then introduced my wireframes and asked Claude to critique them against the same evidence and constraints. This time the goal was to identify where the concept could be improved.

The critique reinforced two things: a cold start should establish context before leading with loyalty, and recommendations should only appear when there is a reliable signal behind them.

Experiment 2 refined cold start exploration: context prompt, seeded experience and skip fallback, annotated with what came from the human concept, what came from experiment one and what changed.
Critique of my cold start: each annotation separates what was reused from my concept, what came from the first experiment and what it argued should change.

My decision

What changed

The useful ideas I chose to take forward after weighing the critique against the original evidence.

  • {{ c.n }}

    {{ c.body }}

    •

Refined direction

The core concept stayed intact. What changed was the hierarchy: context first, a relevant starting point next, with membership still accessible but secondary.

Refined direction wireframe: Club member shell, Looking for dinner? with Here's a place to start, Wave as the contextual starting point with Explore Wave, Explore the ecosystem across Wave, Meadow, Ember and Nomadic, and Club membership as a secondary access point.
The refined cold start home I designed after evaluating both experiments.

Starting concept

Loyalty status → rewards → benefits → ecosystem

Refined concept

Customer context → relevant starting point → ecosystem discovery → membership secondary

What I learned

AI was most useful as a challenger and second pair of eyes. With enough project context, it could explore alternatives quickly, surface assumptions and help me check what I might have missed.

The value was having more options to evaluate — not accepting its answers.

What this means for you

This is the workflow I’d carry into future projects: give AI the right context, use it to explore and challenge the work, then apply design judgement to what comes back.

AI can accelerate the process. The designer still owns the decision.

Previous project

Shell

Next project

Fashion & Play