Shopping Chat on WhatsApp

The pilot was already live when I joined. My job was to understand what worked and what had to be fixed before scaling it to another country.

Result

23.3% of customer conversations ended in an approved order, against ~12% in previous campaigns

Role

Field research, Conversational design, Metrics definition

Challenge

Deciding what to fix in a live pilot before scaling it

Key decision

Moving the MVP to Peru to observe usage in person

Context

Consultants were already selling over WhatsApp: they shared the catalog PDF, their customers marked what they wanted on the image and sent the screenshot back. The bet was to move the order into that same channel, with a native WhatsApp cart tied to each consultant's code, and capture demand that until then got lost in the chats.

The pilot launched in a few regions of Chile before I joined the project.

The full journey, from picking a brand to confirming the order, without leaving WhatsApp. Six screens meant to replace the PDF and the hand-marked screenshot.

Finding

Interviews with consultants from the Chile pilot, field studies in Peru, and two focus groups.

Almost nobody got as far as sending their order. In a focus group, only 1 of 8 consultants completed the purchase. The navigation was understood; the wait broke it: the first menu took 22 seconds, and six out of six believed their order had already been sent when they chose “That's all”.

Announcing it by message wasn't enough. A cart inside WhatsApp was so new that nobody understood it until they tried it.

It served whoever already knows what they want. Compared against the Digital Catalog and the PDF, the chat won on speed and on sharing, and lost on product detail. And none of them reached the customer who needs in-person guidance.

Decisions

I proposed moving the MVP to Peru. If watching usage up close was what gave us the answers, we had to be able to attend the launch conferences and run tests in person.

I trimmed the flow before scaling. Less text per message, numbered steps at checkout, and clear notices about what state the order was in.

We focused the communication on those who already know which product they want, instead of forcing the flow to serve browsers too.

The pilot's greeting explained the tool before letting anyone use it. We cut it to one line and put the brands right there: less to read before the first step.

Confirming the name and accepting the terms were two loose moments. We turned them into numbered steps, so it's clear how much is left before the order goes out.

“That's all” made people believe the order had already been sent. We replaced it with “Confirm order” and, since the stock check took time, added a notice that the order was being reviewed.

On iOS the list didn't show each lipstick's shade and we couldn't change that. We added an image of the shades and their names before the list: not the ideal solution, but it let people choose without leaving the chat.

Measurement

I built the campaign-by-campaign tracking of the funnel. Only a third of the conversations came from customers: the rest were consultants testing it.
Among real customers, approved orders went from ~12% of conversations to 23.3% in the best campaign. Without the promotion, the next one dropped to 16.1%: below the peak, above the starting point.

The customer funnel campaign by campaign, from checkout to approved order. The C12 peak and the C13 drop tell the same story: what the promotion lifted, the channel couldn't hold on its own yet.

Looking for someone who understands the business,
designs with evidence and ships it working?

Let's talk.

/aldo-huapaya