One of Brazil's largest mobility companies
Car subscription sales, from discovery to after-sales, are now handled by an AI agent on WhatsApp.
For one of Brazil's largest mobility companies, Taqtile built an agent that explains, qualifies and sells car subscriptions on WhatsApp, and handles after-sales: it collects the monthly payment and schedules maintenance.
- 3 days
- from the start of the work to the first prototype, tested with real customers on WhatsApp
- 5 subscriptions
- closed by the agent in two days of prototype testing
- 6 stages
- of the journey run by agents, from discovery to maintenance
- +90%
- agreement between the agent's lead qualification and the customer qualification team's, in the first version
- +20%
- in billing collections after the agent joined the process
The story at a glance
Before, more than 70% of leads needed a salesperson. Car subscription is a product the market does not yet know, and it has to be explained at every step, from plans to payment and delivery.
A prototype that sold before going to production. Ready in three days, it closed five real subscriptions in two days of testing, still in the first week of work.
One agent runs the whole journey on WhatsApp, without the team growing in the same proportion. It answers questions, checks prerequisites, builds the subscription and sends the payment link. After the sale, it handles billing, explains damage charges and schedules maintenance.
AI billing collects more. Collections rose more than 20% after the agent joined the billing process. The sales path follows fixed rules, the agent does not make up prices, and a person takes over the conversation when needed.
The pain
The challenge
Car subscription has existed in Brazil for years, with an audience that grows every day. Like any new service model entering a market, it still brings challenges of understanding. Prospects who come in do not know how it works: how to sign up, how much it costs, which plans exist, how to pay, how long the car takes to arrive. More than 70% of leads only felt confident enough to move forward after talking to a salesperson on WhatsApp. The questions do not stop at signing. After that come delivery, maintenance, fines, insurance and damage.
This makes every sale a long conversation, and every customer a sequence of explanations until the end of the contract. Prospects arrive on WhatsApp faster than a team can answer, and the sale goes cold while it waits. Growing the product meant growing the number of people explaining it at the same rate. It is the situation of any company that sells a new product through a digital channel: interest scales, and explanation does not.
The person who called Taqtile was the general manager of the unit, new to the role, who needed to grow the product and wanted to show results with AI, not a pilot.
What we built
The agent
The solution in one sentence: an AI agent that guides the car subscription customer on WhatsApp, from discovery to after-sales, with a person following along and ready to take over.
What the agent does, in the order of the journey:
Discovery
The pre-sales agent (SDR) answers the questions of people getting to know the product, validates purchase intent and checks the prerequisites
Sale
Builds the subscription with the customer, from model and term to protection plans, and sends the payment link
Credit
Explains why documents are needed, receives them and forwards the analysis with a ready recommendation; the decision stays with a person
Billing
Notifies the customer of the monthly payment due, details each item on the invoice and offers card or instant bank transfer (PIX, the Brazilian instant payment system)
Damage
Explains the coverage and the customer's share, and offers payment in installments
Maintenance
Schedules the service, warns of delays and tells the customer when the car is ready
How it decides.
The sales path follows fixed rules, so the agent never makes up a price, a term or a condition. Generative AI steps in where it adds value: explaining, answering objections and conversing in each customer's tone.
The human role.
The client's team follows the conversations in real time. When a conversation calls for a person, the service rep steps in, resolves it and hands it back to the agent. Supervision has been in place since the prototype: it was through supervision that the client's team corrected the agent while it was being built. Autonomy grew in stages, as the agent proved it got things right.
Evolution.
The agent started with sales. Discovery, credit, billing, damage and maintenance came later, each with its own prototype and validation.
What stayed with the client.
All the intellectual property: the code, the specifications and the architecture.
How the client's team operates today.
The client's team brought the platform in-house and evolves it on its own. The safeguards that stayed (the deterministic path in the critical steps and the orchestrator) make it easy to adjust and extend the agents without depending on Taqtile.
Inside the technology
Read detailsHide details+
Internally, each agent is a graph of smaller agents with defined roles: one understands the need, another looks up price and inventory, another explains the contract, another keeps the memory of the conversation, another gives the final answer in the same tone, another hands the conversation to a person. An orchestrator distributes the conversation among them and works as a safeguard.
This graph is combined with deterministic flows. In the steps where a mistake costs money or trust, such as price, term and payment conditions, the answer comes from a rule, not from the model. This combination is what brings hallucination down to a level the operation accepts. Off-the-shelf tools were tested, and in the critical steps, controlling the flow in-house won out.
The agent runs on WhatsApp Business and integrates with the client's systems through the interfaces they already expose (APIs), with a knowledge base of rules, contracts, models, prices and inventory. The orchestration uses open tools, with no proprietary platform.
What already existed stayed. The client's contract, credit, billing and inventory systems remained in place. The agent connected to the APIs that already existed; nothing was replaced.
Results
The results
Sales began before production. In two days of prototype testing, with real customers, the agent closed five subscriptions.
In just two months in production, the agent's service capacity grew 4.5 times, without growing the team in the same proportion. Already in the first version, released within a few weeks, the customer qualification team agreed with the agent's decisions in more than 90% of cases, and the salesperson began receiving better-qualified customers ready to sign. In credit analysis, the credit team began receiving a ready-made recommendation and still makes the decision. In billing, collections rose more than 20% after the agent joined the process: it explains each item on the invoice, offers a way to pay and upholds the company's policy with empathy.
The fear of anyone who puts AI in customer service is losing control of the conversation and losing sales. Here the agent sold on the second day of testing, multiplied service capacity by four and a half, and raised billing collections by more than 20% after joining the process. It did this without making up a price or a rule. People stayed in control, now with the capacity to handle a journey the team could not cover before.
Recognition
The unit was voted the best in its category by the Brazilian consumer complaints platform (Reclame Aqui) and earned that platform's customer service seal (RA1000). The company entered the 100+ Most Innovative Companies in the Use of IT ranking (As 100+ Inovadoras no Uso de TI), by the Brazilian technology outlet IT Forum.
How Taqtile worked
How we worked
The work began with a one-week immersion and prototyping engagement (the AI Sprint, the engagement clients start with at Taqtile) with the sales, marketing, customer success and customer service areas, to map the whole journey, find the pain points at each step and choose where to start. Each prioritized product became its own project. Sales was the first. The prototype was ready in three days and was tested with real customers on WhatsApp, before any build contract. Then each step of the journey became a cycle with the same method: immersion with the team responsible for the step, the incoming questions turned into rules and conversation, prototype, validation and a gradual increase in autonomy. With the agents in production, Taqtile set up the metrics structure and supported the product. Taqtile designed, built and monitored the agent's operation from 2024 to 2026. Today the platform belongs to the client's team, which adjusts and evolves it on its own.
A similar challenge?
In your operation, which conversations can only a person handle today: selling, collecting payments, analyzing credit? Within days, we show you how an agent would handle them.
Questions this case answers
How long does it take to have an AI agent selling on WhatsApp?
In this project, the prototype was ready in three days and was tested with real customers in the first week. The full build, step by step, came later, each step with its own prototype and validation.
Can an AI agent collect monthly payments without annoying the customer?
Yes, when it explains each item on the invoice, offers a way to pay and upholds the company's policy with empathy. In this project, collections rose more than 20% after the agent joined the billing process.
How do you keep the AI from making up a price or a condition?
The sales path follows fixed rules: price, term and conditions come from rules, not from the model. The generative model steps in to explain and converse. And a person follows the conversations and takes over when needed.