Scoping an AI Concierge from App to WhatsApp

Via.ai is a travel-tech startup that helps people turn their loyalty points into trips. It gives banks and travel retailers an AI concierge that recommends curated travel services to their existing customers, opening a new revenue stream.

I joined as the sole designer and product strategy advisor, working directly with the two founders and one engineer. That meant I wasn't handed a spec. I helped decide what the product was: which market to enter first, what the AI should and shouldn't do, what we built to prove the idea, and eventually what we threw away to ship it.

Our primary users had loyalty points scattered across companies. Credit cards, hotels, airlines… and no easy way to see what those points were actually worth. Getting the best value meant jumping between sites, comparing redemption rates, and piecing together a trip by hand. Most people gave up and let the points sit.

Our first scoping decision was who to ignore. Rather than build for travelers everywhere, we picked Israel as the beachhead market: high outbound travel, strong loyalty-program adoption, and daily habits already suited to a conversational product. Narrow enough to actually win, and a decision that paid off later in a way we didn't expect.

The market backed the bet:

The core challenge was deciding what the AI actually was: a search box with better manners, or someone you come back to.

We named it Daniel, and the name did real work. Giving the agent a name forced us to answer product questions we'd otherwise have left vague: what Daniel knows about you, what he's allowed to recommend, when he pushes and when he backs off. It also gave our partners something to picture, which mattered more in a pitch meeting than any screen did.

I scoped the conversation around three questions the team had to answer before anything could be built:

• How does Daniel respond to vague or evolving requests?
• How does he handle mid-journey changes, like a canceled flight or a sold-out hotel?
• What relationship should someone build with him over time?

Answering those three set the boundaries of v1 and, just as usefully, told us what to leave out.

At a pre-seed startup, the scarcest resource is belief. We bet that the fastest way to buy belief was to make the whole vision feel real before a line of it existed, so I designed a fully interactive, end-to-end app in Figma:

• Personalized trip flows
• Smart destination suggestions
• Retail-point redemption
• Dynamic offers and upgrades

The bet paid. The prototype became the centerpiece of investor and partner conversations, showing concretely how Via.ai could unlock new revenue and deepen loyalty. It was used in the company's fundraising and in partner pitches to national banks, travel retailers, and an airline, which signed on as a partner.

Then it worked too well.

Partner interest created a deadline the roadmap couldn't meet. One engineer, real clients waiting, and a native app that would take months we didn't have. Keeping Plan A meant showing up to those conversations with the same Figma file for another two quarters, and watching the interest cool.

After assessing engineering capacity, I recommended we kill the app and keep the concierge, and led the rescope with the founders. The prototype had already done its job. Holding onto it as the shipping plan would have cost us the momentum with our new partners.

This is where the beachhead decision paid off. Because we had scoped to Israel, we had a channel almost everyone in the market already used every day. Moving Daniel into WhatsApp cut the install step out of the funnel entirely and shrank the build from months to something the team could actually ship.

• Used daily by 95%+ of Israelis
• Already trusted for personal and business communication
• Purpose-built for conversational AI

What we gave up: the rich visual surfaces, the upsell moments, and most of the brand control our partners liked in the app. What we got: zero adoption friction, and a version of Daniel that one engineer could actually deliver to real partners. For a company that needed proof more than polish, that was the right trade.

I wrapped my engagement as the WhatsApp build got underway.

WhatsApp gave me a fixed set of parts: text bubbles, images, and reply buttons. No custom components, no screens of my own, no back button. Every product decision had to be expressed in those three things, which moved the design work out of layout and into sequencing and language.

A working version of the WhatsApp experience, rebuilt in English. Tap the replies and plan a football trip with Daniel, from the first message to a booked package.

The most valuable thing I did on Via was agree to throw the app away.

The prototype was my best work, and it was still right to cut it. Its job was to win belief, and it had already done that. Knowing when a piece of work has finished its job is the contribution I'd point to here.

What I'd do differently: Building the prototype was the right call. Letting partners see it as the only shipping plan was the mistake. Next time I'd show the full vision and a leaner first release side by side, so a pivot reads as the plan all along.