AI Showreel the independent observatory of AI in marketing
← The index
Proof A Live confirmed

Magalu

end-to-end genAI sales agent in a messaging app, from advice to payment

IndustryRetail & e-commerceLeverActivation / conversionFamilyConversationImplementationHybridStagepurchase
Pattern proven in 9 industries still untouched in Banking, insurance & fintech, CPG & D2C, Tech & SaaS +3 See the pattern map
x3
Channel conversion rate, compared with standard search in the Magalu app
"três vezes superior à busca tradicional do aplicativo e um NPS de 84 pontos" S1

Launched in November 2025, Magalu's WhatsApp da Lu keeps the entire purchase inside WhatsApp through more than 20 orchestrated AI agents; according to Magalu's 2Q26 earnings release, the channel has passed 18 million conversations, converts three times better than standard app search, posts an NPS of 84, about 20% repurchase and more than 100 million reais in cumulative sales.

Key points

  • Magalu sells in WhatsApp through Lu, an AI that advises, takes payment and tracks the order.
  • More than 20 orchestrated AI agents, co-developed by Luizalabs with Meta, Google Cloud and Bain.
  • Conversion three times higher than standard app search, about 20% repurchase.
  • More than 100 million reais in cumulative sales, figures included in the 2Q26 earnings release.

Objective

Open a sales channel where Brazilian customers already spend their time, WhatsApp, and have Lu, the brand's virtual influencer, carry the entire purchase journey there, to convert better than the app and the website.

The deployment

WhatsApp da Lu is a sales channel Magalu opened in November 2025, first for a group of 1 million active customers, then for all consumers before the end of 2025. The customer writes to Lu, the brand's virtual character, by text, voice message or by sending a photo. The assistant asks clarifying questions (style, budget, use), proposes a selection from the first-party (1P) catalog and from that of the roughly 300,000 marketplace sellers (3P), then takes payment within the conversation, by Pix or card, with no redirect. Order tracking and after-sales service go through the same thread. According to Bain's case study, the system orchestrates more than 20 AI agents that split product discovery, recommendation, search interpretation, customer service and the transaction, on an architecture that mixes proprietary and third-party models. Development is led by Luizalabs, Magalu's technology division, with Meta, Google Cloud and Bain & Company. On the WhatsApp side, Meta says Magalu relies on card carousels, WhatsApp Flows and payments in WhatsApp, and that marketing messages are the main entry point. In its second quarter 2026 release, Magalu makes the channel the core of its new strategic cycle: more than 18 million conversations since launch, conversion three times higher than standard app search, an NPS of 84, about 20% repurchase among customers who have already bought, and more than 100 million reais in cumulative sales reached in July 2026. The comparison base for the conversion figure varies by source: app search in the release, the app in the Bain study, existing sales channels in the press release relayed by the press. At group scale, the channel remains small: the same release reports 9.3 billion reais in e-commerce sales for the quarter alone, down 11.9% year on year. In 2026, Magalu added a Mother's Day operation (personalized AI-generated video from a voice message and a photo), a first TV campaign dedicated to the channel in June, and the system won a bronze Lion at Cannes 2026.

Results Proof A

x3
Channel conversion rate, compared with standard search in the Magalu app
"três vezes superior à busca tradicional do aplicativo e um NPS de 84 pontos" S1
R$ 100M+
Cumulative channel sales since the November 2025 launch, threshold passed in July 2026
"já totalizaram mais de 100 milhões de reais" S1
environ 20%
Share of customers who had already bought with Lu who bought again through the channel
"dos clientes que já compraram com a Lu, cerca de 20% voltaram a comprar" S1
18M+
Cumulative conversations since launch, as of the 2Q26 release
"superou a marca de 18 milhões de conversas desde o seu lançamento" S1
7,7M
Unique users over the first eight months of operation, according to the company
"alcançou 7,7 milhões de usuários únicos em oito meses de operação" S3

The 3x conversion, NPS, repurchase of about 20%, 18 million conversations and 100 million reais in sales appear in the 2Q26 earnings release published by Magalu on its investor relations site; they are consistent with the Bain case study, the WhatsApp Business customer story and the press release relayed by Central do Varejo.

How it works

Inferred typical approach

The internal detail is not public. Here is a proven approach that leads to the same result, to adapt to your stack.

recommandations en carrouselpreferences memorisees, boucle d'apprentissage Client Magalu surWhatsApp WhatsApp da Lu WhatsApp Business Platform (carrousels, Flows, paiements) Orchestration de plus de20 agents IA (modelesproprietaires et tiers) Catalogue 1P et 3P,historique d'interactionsclient Paiement Pix ou carte,suivi de commande

The stack in detail

  • plateforme WhatsApp Business Platform Card carousels, WhatsApp Flows for product pages, payments in WhatsApp integrated with Pix, marketing messages as the entry point.
  • infra Google Cloud Development partner cited by Bain; the services used are not publicly detailed.
  • plateforme Systeme multi-agents Luizalabs More than 20 orchestrated AI agents, proprietary models trained on Magalu data and unnamed third-party models.

How it runs, concretely

For ops teams
CadenceReal time for each conversation. WhatsApp marketing message campaigns and seasonal operations (Mother's Day, TV campaign) drive traffic in waves.
Operated byLuizalabs for the platform and the agents, Magalu's customer experience team for service quality, marketing for the campaigns that bring people into the channel.
  1. 1
    Entry into the channel Marketing

    The customer arrives in the conversation, most often through a WhatsApp marketing message or a campaign promoting the channel.

  2. 2
    Understanding the request AI

    Lu interprets a text, a voice message or a photo and asks clarifying questions about style, budget and use.

  3. 3
    Selection and presentation AI

    The search and recommendation agents draw on the 1P and 3P catalog and display a selection in a carousel, with structured product pages.

  4. 4
    Payment in the conversation Customer and payment platform

    The customer pays by Pix or card without leaving WhatsApp.

  5. 5
    Tracking and after-sales AI, with the customer experience team

    Lu sends shipping updates and handles after-sales requests in the same thread.

  6. 6
    Measurement Customer experience team and senior management

    Magalu tracks conversations, conversion, NPS, repurchase and cumulative sales, and reports them in its financial communication.

The signal that drives it

The conversations themselves and the interaction history, used to learn each customer's preferences, plus the 1P and 3P catalog. Bain says each interaction is tracked with performance metrics and quality checks: without a current catalog (price, stock, sellers), the agent recommends products it cannot sell.

How your customers perceive this type of use

Sourced studies

Les consommateurs n'acceptent pas les chatbots par defaut : 64% prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (Gartner, 2024) et pres d'un utilisateur sur cinq du service client par IA n'en retire aucun benefice (Qualtrics, 2025). L'acceptation se construit sur trois conditions mesurees par Salesforce : savoir qu'on parle a une IA, pouvoir escalader vers un humain, comprendre la logique de l'agent.

64%
Consommateurs qui prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (2024)
53%
Consommateurs qui envisageraient de passer a un concurrent s'ils apprenaient que l'entreprise prevoit d'utiliser l'IA pour le service client (2024)
pres de 75%
Consommateurs qui veulent savoir s'ils communiquent avec un agent IA (2024)

Acceptance conditions

  • Etre informe qu'on parle a une IA et non a un humain (pres de 75% le demandent, Salesforce 2024)
  • Un chemin d'escalade clair vers un agent humain (45% plus enclins a utiliser l'agent IA, Salesforce 2024)
  • Une logique de l'agent clairement expliquee (44% plus enclins, Salesforce 2024)

Red lines

  • Rendre l'humain injoignable : c'est la premiere inquietude des consommateurs sur l'IA dans le service client (Gartner 2024) et 50% craignent que l'IA les coupe du contact humain (Qualtrics 2025)
  • Remplacer le service client par l'IA sans alternative : 53% envisageraient de partir chez un concurrent (Gartner 2024)

Sources: Salesforce 2024 · Gartner 2024 · Qualtrics 2025

See full acceptance: by country, by use, by generation

How to replicate

Inference, not sourced

Data prerequisites

  • structured product catalog kept current on price, stock and seller, queryable in real time
  • customer base reachable on WhatsApp with opt-in for marketing messages
  • interaction history tied to the customer to remember their preferences
  • order tracking exposed through an API to answer within the conversation

Org prerequisites

  • a brand character or a voice customers already know, which gives them a reason to write to the channel
  • a product team that owns the channel end to end, from conversation to payment
  • a payment method native to the messaging app in the target market
  • a quality control process for conversations before and after going to production

Possible stack

  • WhatsApp Business Platform through a BSP provider
  • LLM from a cloud provider, with agent orchestration and tools connected to the catalog
  • existing product search and recommendation engine exposed as a tool to the agents
  • payment solution integrated with the messaging app, or a payment link
Team to operateA channel product owner, AI engineers for agent orchestration and conversation evaluation, integration developers (catalog, payment, orders), the customer service team for cases the AI does not resolve, and CRM marketing for WhatsApp campaigns.

The plan, step by step

  1. Step 1
    Map the purchase journey to fit into the conversation (search, clarifying the need, recommendation, payment, tracking, after-sales) and decide what stays outside the channel at first.Deliverable: Functional scope of the first version.
  2. Step 2
    Split the work into specialized agents (search, recommendation, transaction, service) connected to the catalog and order APIs, rather than a single model that does everything.Deliverable: Agent architecture and exposed tools.
  3. Step 3
    Integrate payment into the conversation with the dominant local payment method.Deliverable: Order paid end to end with no redirect.
  4. Step 4
    Launch with a group of active customers, track conversion, NPS and repurchase against app search, then open to everyone.Deliverable: Comparative read and decision to open.
  5. Step 5
    Promote the channel through marketing messages and seasonal operations, and track the share of new users they bring.Deliverable: Channel activation calendar.

First step: Open the channel to a limited group of active customers, on a narrow set of categories, and compare its conversion with app search for those same customers.

Sources

  1. S1 Magazine Luiza - Divulgacao de Resultados 2T26 (release de resultats) Primary ri.magazineluiza.com.br · Aout 2026 (teleconference de resultats du 7 aout 2026) · accessed 2026-10-05 archive pending
  2. S2 Magalu e pioneira em compras com inteligencia artificial no WhatsApp (Bain & Company, client results) Interested party bain.com · accessed 2026-10-05 archive pending
  3. S3 Magalu ultrapassa R$ 100 milhoes em vendas com canal de compras no WhatsApp (Central do Varejo) Secondary centraldovarejo.com.br · 2026-07-09 · accessed 2026-10-05 archive pending
  4. S4 Magalu | WhatsApp for Business (success story) Interested party whatsappbusiness.com · accessed 2026-10-05 archive pending