Insights
Artificial intelligence applied to real operations: customer service, integrations, deployment and governance.
For the first time, an electric car made it onto Brazil's sales podium. Behind the number is an automaker that treats data and artificial intelligence as part of the product — and a buyer who arrives at the dealership expecting the same from customer service.
6 min read
More dealerships competing for the same buyer, consumers choosing a new Chinese car over a late-model used one, and used-car prices under pressure. The expansion of Chinese networks affects parts of the operation that traditional dealerships do not always associate with competition.
Buyers send the same message to several dealerships, often at night. Whoever replies first with real information and keeps the conversation going over the following days gets the sale. That is where AI makes a difference.
Over two days in September, three of the leading AI labs launched cheaper models. The cost of the model went down — but it is rarely what determines whether a project is viable.
Not every system has an API. The new generation of agents navigates the system the way an employee would — which removes a long-standing bottleneck, but brings risks of its own.
In the same week that a case involving OpenAI's agents became public, Dataiku launched a tool to inventory enterprise agents. Both stories point to the same problem.
The question that defines the scope of an AI project is not which model to use, but which data the agent will have access to. Without integration, the scope changes — and it is better to know that up front.
The cost of an AI project is almost never where sales presentations put it. Understanding the real breakdown avoids both surprises midway and return expectations that never materialize.
RAG is the technique that lets a model answer based on your company's documents instead of what it learned in training. But it is not the answer to every internal question.
An AI agent at a dealership is not an FAQ chatbot. It checks inventory, qualifies interest and books visits — as long as it is connected to the right systems.
Automating WhatsApp is easy. Automating it without irritating the person on the other end requires deciding carefully what the machine answers and when it steps aside.
The question is not whether the model will make mistakes, but what happens when it does. Governance is the set of answers you prepare before that.
