ListoApp

Guide · Real case study

How We Built a WhatsApp AI Assistant for Our Own Business

Chef Toto is the AI assistant that answers WhatsApp messages inside our own company. It's not a demo — it's in production. Here's what we learned building it.

Written by the ListoApp team · August 2026 · 8 min read

Our business makes food products. The team works over WhatsApp — not some slick CRM, WhatsApp. And the same kind of question kept showing up every day: how much a recipe yields, how much a kilo costs, what goes into a product's spec sheet.

Chef Toto came out of that. Here's the actual order we built it in.

Week 1 — Document, don't code

We didn't write a single line of code the first week. We sat down and listed the real questions that came in, and wrote out the correct answer to each one. Three categories emerged:

• Recipes and yields — calculations with fixed rules.
• Costs — these depend on prices that change, so they had to come from the system, not a document.
• Processes — how something gets done, who approves it.

That exercise revealed something uncomfortable: on several questions, the team didn't agree on the correct answer. That's not an AI problem, it's a business problem — and it was better to find that out this way.

Weeks 2 and 3 — Connect the knowledge, not just the model

The difference between a generic chatbot and a useful assistant is where it gets its information. Chef Toto doesn't "know" things from the internet: it queries the business's knowledge base and the live data in the inventory system.

Two decisions we still stand by today:

1. One single source of truth. If the cost changes in the system, it changes in the answer. No copying data into a separate document.
2. Memory of the business, not of the chat. The knowledge lives documented and versioned, so every conversation picks up where the last one left off instead of starting from zero.

Week 4 — Production, with the handbrake on

We launched it first for the internal team only, not for customers. For two weeks, every response was reviewed. We found three flaws:

• It answered too confidently when it didn't actually know. We fixed that by forcing it to say "I don't have that information" and hand off to a person.
• It answered too long. Nobody reads eight lines on WhatsApp. We adjusted the tone: short answer first, detail only if asked.
• It had no personality. It sounded like a form. That's where the Chef Toto character was born — his own name, voice, and tone. Sounds silly, but it's what got the team to actually start using it.

Adoption wasn't a technical problem. It was about trust and tone.

What we'd do differently

• Set up analytics from day zero. We added it late and lost our baseline — we can't compare before and after with hard numbers. Don't make that mistake.
• Assign a content owner before launch. One person responsible for making sure the answers are still accurate six months from now.
• Start with fewer questions. Ten well-answered questions build more trust than fifty half-answered ones.

What it takes to replicate this in another business

You don't need a tech team. You need:

• A channel where your people already are (almost always WhatsApp, or SMS/email for a US-based team).
• Twenty frequently asked questions with an agreed-on answer.
• A reliable data source for whatever changes (prices, inventory, order status).
• One internal owner and a month of patience.

If you want to see Chef Toto answering live, it's on the homepage, unscripted (the demo runs in Spanish, its working language).

Frequently asked questions

How long does it take to build a WhatsApp AI assistant?

With the business's knowledge already documented, 3 to 6 weeks until it's up and running. Without that groundwork, the timeline doubles, because documentation is the real bottleneck.

Does the assistant replace a team member?

Not in our case. It replaces repetition: the questions that have already been answered a thousand times. The team is freed up for what actually needs judgment.

What happens when the assistant doesn't know something?

It has to say it doesn't know and hand the conversation to a person. An assistant that makes things up does more harm than one that recognizes its limits.