Capi — Tuali Growth Agent

Capi is a coach-mascot AI agent inside Tuali that turns a shopkeeper's goal into a game plan using Tuali's own tools, tracks the win, and learns — voice-first, so even non-tech users can grow.

Comment

Inspiration

Tuali (by Arca Continental, a Coca-Cola bottler) already gives Mexican shopkeepers powerful tools — promotions, loyalty points ("Gana"), suggested orders, and coupons. But as Arca's own brief told us: these tools create value individually, yet there's no way to use them together to reach an actual business goal.

We also dug into Arca's protopersonas and user journeys. Two things jumped out: shopkeepers like Raúl (63, only 15% digital skill, relies on a sales rep) can't use a dashboard at all — and Fernanda's journey literally lists "reminders on order day" as a missing feature. The opportunity wasn't another chatbot. It was an agent that orchestrates what Tuali already has, toward a goal the shopkeeper chooses — and that even non-tech users can actually use.

And a small spark: Tuali's "Gana" branding already features a soccer ball. With the 2026 World Cup hosted in Mexico, the metaphor wrote itself. In Spanish, meta means both "goal" (business objective) and "goal" (the net you score in). So we built a coach.

Our design principle: fewest effective clicks

This is the rule we held everything to. These users are busy — Fernanda runs a store and a household; Raúl avoids anything that feels like learning a new system. They open the app to do one thing and leave. So Capi has no fluff, no menus to dig through, no steps that exist just to look smart. Every tap moves you closer to the goal: state it, get the play, act, see the result. The whole loop is reachable from a single floating button, and for users who don't even want to tap, voice does it all. We designed for get in, grow, get out — because that's how real shopkeepers use the app.

What it does

Capi is a growth-agent mascot — a soccer-coach ball — that lives as a floating button inside the Tuali app and closes a full loop:

  1. Set your goal ("your next gol") — sell more, or raise average ticket. One tap or one spoken sentence.
  2. The play — Capi builds a 3–4 step plan, each step drawn from a tool Tuali already has (Gana reto, suggested order, promo, coupon), with a reason personalized to that store ("water sells a lot in your area but you rarely order it — add it"). Each action activates in place — no hunting through the app.
  3. Track the win — average ticket moves from $180 to $202 (+12%), shown as progress toward the net. Score equals goal.
  4. Learn and adjust — Capi tells you what worked, what didn't, and proposes the next goal.

It's voice-first (ElevenLabs) so users with low digital skills can grow hands-free, and it can send reminders ("order day", "goal progress") via WhatsApp / push.

How we built it

  • Front end: React Native + Expo — because the agent had to live inside the mobile app. We replicated Tuali's real UI (lavender background, Coca-Cola red, the Gana-centered tab bar) and layered Capi on top as a floating assistant.
  • Backend and data: MongoDB Atlas stores products, encrypted behavior data (no PII, per the brief), goals, and the action-to-outcome log that powers the "learn and adjust" step. Its flexible schema was ideal for that loop record.
  • Brain: Gemini API parses the shopkeeper's goal in natural language and reasons over store context to assemble the play.
  • Voice: ElevenLabs gives Capi a warm, natural Mexican-Spanish voice.

Challenges we ran into

  • Staying faithful to a real app we don't own — we worked from screenshots and the brand kit to match Tuali's look so judges would feel it's their app with a layer on top, not a clone.
  • Making the loop visible, not just claimed. Most teams stop at "recommend products." The hard part — and our focus — was showing the second half: the number moving and the agent learning. We simulated a time fast-forward so the impact is felt live in the demo.
  • Cutting, not adding. The hardest design discipline was saying no. Every time we wanted to add a screen or an option, we asked: does a busy shopkeeper need this to reach their goal? If not, it was gone. Fewest effective clicks meant constantly resisting our own feature ideas.
  • Designing for Raúl, not for us. Choosing voice over a dashboard, big buttons, and reminders through WhatsApp (where shopkeepers already live) all came from the personas, not our preferences.
  • Orchestration over invention — the discipline of using only Tuali's four existing tools, instead of dreaming up new ones.

What we learned

That the winning insight wasn't a new feature — it was connecting what already existed, and then removing everything that got in the way. We also learned how much a character changes adoption: a coach who cheers you on feels like a teammate, not software, which is exactly what an intimidated shopkeeper needs.

What's next for Capi

Proactive nudges ("you haven't ordered water in 3 days — your area is selling out"), a "trophy room" of completed goals, real WhatsApp Business API integration, and multi-goal tracking.

Built With

Share this project:

Updates