B2B Commerce
at LATAM scale.
BEES is AB InBev's marketplace platform — the digital infrastructure that replaced the sales rep visit for hundreds of thousands of small retailers across Latin America. I worked embedded with the product and VP team across multiple simultaneous initiatives: onboarding and activating new buyers, shipping features that drove retention, and proposing the future direction of BEES.ai.
At BEES I didn't own a single product area — I was brought in where the thinking was hardest. That meant shipping a real feature one week and presenting vision demos to VP leadership the next. The work below reflects that range: four shipped initiatives spanning the full user lifecycle — attract, onboard, activate, retain — and a future-vision proposal that became the foundation for BEES.ai.
The BEES homepage — featured brands, partner stores, AI-powered suggestions, loyalty rewards, geo-personalized trending products. A full commerce platform, not just an ordering app.
Attract & Onboard
Closing the gap from invite to first order — by redesigning registration, login, and credit onboarding as one connected activation system.
Registration abandonment was the platform's biggest growth leak. Retailers were being invited to BEES — via WhatsApp, SMS, and email — but a significant portion never completed registration. Of those who did, many never placed a first order. Login friction compounded it: forgotten passwords created drop-off even for users who'd registered successfully. A credit product sat downstream, adding a second trust barrier before retailers could unlock the platform's full value.
The activation barrier wasn't awareness or motivation. It was friction in the first 60 seconds. Simplify registration. Introduce trust signals before asking for commitment. Remove passwords as a login barrier. Give users a visible path to their first transaction — and measure success there, not at signup.
Redesigned the full user registration flow — reducing steps, adding clearer progress signals, and building market-specific paths (Mexico, Brazil, and Ecuador each had different legal and data requirements). Introduced OTP as a passwordless login method that became the primary authentication path. Designed the invite-to-activate journey across WhatsApp, SMS, and email so the first experience felt personal, not generic. Designed the credit onboarding flow — making a financial product feel trustworthy in a market where informal credit had always been the default.
We stopped measuring success at registration completion. The real signal was % of new users placing a first order within 7 days. That single shift changed every onboarding decision — copy, step count, confirmation states — because we were now designing for the moment of transaction, not the moment of signup.
Optimize for invite channel earlier. WhatsApp dramatically outperformed email in Brazil — but that signal took too long to surface and prioritize. The channel is part of the onboarding experience, not a distribution detail.
Invite to activate
WhatsApp message pre-populates account context. The first tap opens a registration flow that already knows who you are.
Simplified registration
Steps reduced and sequenced by market. Mobile phone as primary identifier — 91% of registrations happened via phone, not email.
OTP over password
A one-time code replaces the password step. Fewer failed attempts, faster completion, and a login method that stays frictionless on return visits.
Account activated
"Welcome to BEES — you can now shop products from Nestlé." Confirmation routes directly to the homepage, not a generic success screen.
Track My Order
Delivery visibility that reduced order refusals by 14% — by reframing a logistics problem as a communication problem.
48% of orders were being cancelled because the business was closed at delivery time. The only status BEES showed was a single line of text. About 30% of all customer support tickets were delivery-related.
The cancellation wasn't a logistics problem — it was a communication problem. If retailers received proactive updates with enough advance notice, they could plan around it. The solution wasn't a better truck tracker. It was better information architecture around the delivery arc.
Led discovery across the Deliver and Customer teams. Mapped the full communication arc from order placed to delivery confirmed. Ran a preference test with retailers on two directions — horizontal vs vertical status indicator, with and without a live map. Designed the MVP and handoff.
I initially argued against the live map in the MVP — small sample size in testing, added complexity. Stakeholders pushed for it. They were right. The map became the feature's most cited element in post-launch feedback.
Test in Brazil specifically before shipping. Internal testing gave us false confidence on notification timing — optimal lead time varied significantly by neighborhood density in São Paulo.
Push notification
Order out for delivery with arrival window — giving retailers time to prepare.
Order details
"Track delivery" surfaced prominently — one tap to live tracking.
Live map
Minute-by-minute truck location with estimated arrival time.
Delivered
Three-stage progress bar resolved. Order confirmed with timestamp.
Add New Vendor
Turning a 3% vendor request rate into a scalable activation flow — by solving for placement and timing, not just discoverability.
Only 3% of monthly active retailers were requesting access to new vendors — despite 16,783 potentially eligible accounts. The entry point existed. Nobody found it. And when they did, the contracts were fully manual.
Placement mattered more than the feature itself. A retailer who just reordered Ambev products was primed to discover Nestlé. The moment of category relevance was the moment of action.
Designed contextual discovery placement — a partner banner surfaced inline at the right category moment. Built the Shopping Options drawer showing locked vs active vendors. Designed the category interest selection flow so the backend could process signals without requiring full manual contracts upfront.
Legal flagged that being explicit about vendor matching created liability. We reframed from "we'll match you with sellers" to "tell us what you're interested in" — a subtle copy change that cascaded into every screen in the flow and required rethinking the entire confirmation architecture.
Contextual entry point
The Nestlé banner surfaces inline on the homepage — at the partner stores section, not buried in settings.
Locked vs active vendors
The Shopping Options drawer shows the full vendor landscape. The lock icon makes the activation path legible without explanation.
Category interest
Retailers select the categories they want to carry. This signal routes to the backend — faster for retailers, scalable for ops.
Submitted
Warm confirmation state sets expectation for the verification step without creating anxiety around timing.
New Buyer Registration
Building a personalized homepage from a retailer's first session — using category preference signals to unlock a tailored commerce experience.
New retailers landing in BEES saw a generic homepage — the same brands and offers regardless of whether they ran a bar in São Paulo or a corner store in Quito. Engagement was low because relevance was low.
Capture category intent in the first session — even just 2–3 taps — and the homepage feels built for that specific store. The personalization didn't need to be perfect. It needed to feel personal.
Designed the category selection flow with rich product photography to make categories concrete. Built the animated "building your store" moment — the bee mascot progressing through each personalization step — to make the AI logic visible and earned.
The animation was a late addition — originally we cut straight to the homepage. But user testing showed people didn't trust the personalization was real without seeing it happen. The animation made the intelligence legible.
Market-specific onboarding length. Retailers in Ecuador were more willing to answer additional questions than those in Brazil, who preferred to start browsing immediately. One-size onboarding left value on the table in both markets.
Rich product photography makes categories concrete. Sticky "Continue" CTA activates once a selection is made.
The bee mascot progresses along a yellow bar while the homepage builds in real time — brands populating, partner stores unlocking, products appearing. The animation makes AI personalization visible and earned.
"Special suggestions for you — Powered by AI" and "BEES customized suggestions" surfacing products matched to the retailer's stated categories.
The homepage is the first impression BEES makes every session. A generic homepage says "we don't know you." A personalized one says "we built this for your store." Retailers who completed onboarding had significantly higher 30-day retention than those who skipped it.
Meet Buzz —
your AI business partner.
In parallel with shipped features, I collaborated with VP leadership on the future direction of BEES.ai. The question: what does it look like when BEES doesn't just process orders, but actively helps a small business owner run their business?
The richness of the BEES homepage — the foundation Buzz sits on top of.
Buzz handles a broken refrigerator — from photo to technician booked in one conversation.
A tendero notices their Guaraná Antarctica fridge is broken. They tell Buzz. Buzz opens a support ticket, books a technician for the next morning, notifies the sales rep, and proactively suggests an inventory recovery plan — because a broken fridge means lost sales, and Buzz knows that.
Every e-commerce platform has a shopping assistant. Nobody has an AI that handles operations, equipment, support, and inventory in a single conversational interface. This is BEES.ai as an intelligent business partner — not a smarter search bar.
Voice ordering · Image recognition for equipment issues · Support ticket creation · Technician booking · Inventory recovery planning · Order status tracking · Product returns and exchanges · Multi-turn conversation with full context.
One trigger, all modes — the search bar is the single entry point for voice and text. The bee mascot appears only during active AI states, never idles. Response surfaces scale to task complexity: compact card for simple answers, full sheet for task completion.
What BEES taught me
about design at scale.
Reframe before you design
Every initiative I worked on was misdiagnosed at the start. The best work began with asking "is this actually the problem?" before opening Figma. That question saved months of building the wrong thing.
Measure at the moment that matters
Onboarding success isn't registration — it's the first transaction. Shifting the KPI from "completed signup" to "placed first order within 7 days" changed every design decision downstream. The metric you pick is the experience you build.
Legal and ops are design constraints
The vendor activation copy change wasn't a PM decision — it was a legal constraint that cascaded into every screen in the flow. At scale, the most important design partners are often outside the product team.
Make the intelligence visible
The NBR animation, the Buzz mascot during listening, the personalized homepage reveal — all were about making AI logic legible. Users don't trust what they can't see happening. Design the trust, not just the output.