We work out what people will do. Then we design it, build it and test it with them.
Product strategy, customer research and behavioural design for financial services and consumer products where trust decides the outcome. We build working software ourselves, so you get a real prototype in weeks, not a slide deck.
Available for consulting, contract and interim product leadership
Neon Lighthouse is a new studio. This work was led in-house and in our own ventures. It is the evidence behind the method, not a client list.
The practice
Every engagement starts with one question: what will people actually do, and why?
Product strategy, including AI
Deciding where a product, or AI inside it, earns its place and where it is decoration. Discovery, working prototypes, and the question underneath: will people trust this enough to act on it?
Customer research and validation
Interviews, behavioural analysis and small, sharp experiments that lead to a direction you can defend.
Customer experience and service design
Onboarding, engagement and outcomes. Includes inclusive design for vulnerable customers.
Event and community platforms
Membership, engagement and network effects, for products where the value comes from other people showing up.
Build
We design it, and we build it.
Most consultancies hand you a document. We hand you software people can use. We built Bachata Connect, a live platform, on our own, with AI coding agents. We wrote the prompts, the rules and the design.
It runs a real event
Participants, door and desk staff, teachers and admins use it live. It has 122 routes and 142 database migrations. One person can hold several roles, for example a dancer who also teaches.
The stack
Next.js, React, strict TypeScript, Tailwind, Supabase (Postgres with row-level security) and Vercel. Python for data work: loading spreadsheets, cleaning and merging duplicate records, and syncing with ticketing and video. Video runs on Cloudflare Stream and email on Resend.
How we keep it safe to ship
The whole app is strict TypeScript, about 560 files. CI runs a typecheck on every pull request and blocks the change if it fails. Playwright tests run the main user journeys. Accessibility checks and colour contrast tests also run in CI, so a change that breaks them can't ship.
A design system, not a set of pages
Three brands run on one component library. A theme is data: colours are named by job, never by hex value, so adding an artist means filling in one table. One of our brand colours failed text contrast at 2.6 to 1. We split it into two tokens, and a test now fails any text colour that drops below WCAG AA. A missing translation key fails CI too, instead of showing a blank label.
AI where it earns its place
Claude translated 325 expert reviews for under €1 and wrote 16 artist feedback digests. Class recommendations use plain rules, not a model, and each one shows its reason, such as "Your reviewer flagged musicality as your biggest area to grow." Recommended classes sold at twice the rate, and 65% of people bought add-ons.
Privacy built in
GDPR set up from the start: data controller, EU representative, privacy policy and a record of processing.
Production code is private because it holds participant data.
Work
Three examples, with real numbers.
Problem. An international event with about 1,500 applications a year for about 350 places, run on spreadsheets, email and desk queues.
What we did. We designed and built the whole platform, then turned it into a year-round platform for artists to sell their own courses. Paying artist clients are live.
Result. 340 of 344 arrivals linked a wristband at check-in, with no manual payment overrides. Recommended classes sold at twice the rate. 65% bought add-ons.
Problem. A platform migration in 2018 left customers locked out. The app was rated 1.8 stars.
What we did. We put the first in-house designer in place. The customer response was designed to reassure, listen to and update customers while fixes happened behind the scenes. Then the design function and design systems were built from zero, hiring the team, and working directly with the engineers rebuilding the app.
Result. App rating 1.8 to 4.6. Digital sales up almost 20%.
What we did. We led a team of 15 and owned CX strategy for products serving 1.7M+ customers. We co-authored the Vulnerable Customer Policy and designed an inclusive design framework. We led early-intervention strategy for credit cards and vehicle finance, redesigning journeys for 109,500+ customers. For Moneybarn, we led the team that mapped the journeys for its new platform, then worked with delivery teams on the user stories.
Result. Journeys redesigned for 109,500+ customers. The new Moneybarn platform is being implemented.
Earlier work
Amplifon hearing-aid app: research with users and audiologists across Italy, six international awards. Pepper, Israel's first digital bank: service design and product prioritisation. SAS EuroBonus, Telenor and Cisco at Designit. Before that, C++ and UI development at Opera Software, building browsers for Nintendo Wii and DSi and Motorola phones.
Wearables
A band that found its real job.
Most wearable projects fail on behaviour, not hardware. We learned that first hand.
At Bachata Week 2026, 340 of 344 arrivals linked a wristband to their profile, with no manual payment overrides. But only 38% of holders tapped another person. Taps fell from 139 on day one to 35 on day seven. After the event it was 2 taps against 56 saves in the directory.
So we stopped asking the band to be a social habit. It became the moment you arrive. A tap still means you met face to face. Finding people again moved to the directory.
That is why the band is worth taking to other events. Every festival already buys entry wristbands. If yours is the entry band, hundreds of people are one tap from the app.
Talk about a wearable pilot
Method
A clear path from question to direction.
Understand the problem
Frame the real question, not the surface symptom.
Understand the people
Interview, observe, and map the human context.
Find the behavioural barriers
Pinpoint what blocks action, not just opinion.
Test assumptions
Run focused experiments with real users, in working software.
Design the next experiment
Turn signal into the smallest useful test.
Create a clearer path forward
A prioritised, defensible product direction.
Engagements
Four ways to work together.
Exploration call
You describe the problem. We tell you whether it is a research, product or behaviour problem, and what we would do first.
Ends with a written next stepOpportunity review
A fast read of an existing product, journey or AI concept against evidence of what customers do.
Prioritised opportunities and risksValidation sprint
Interviews, behavioural analysis and a working prototype, built in code and tested with real users.
A direction you can fund or stop with confidenceFractional and interim leadership
Embedded with a founding team or innovation group. Setting direction, running discovery and shipping alongside your engineers.
Senior capability without a permanent hireLet's explore the problem before building the solution.
Understanding human behaviour. Designing better products. Building stronger communities.