Don't Sell AI, Sell It On Top of Something They Already Pay For

The audio and video are in Spanish. These notes are a translation.

Today I had three guests who do not sell artificial intelligence. They use it to remake their industries.

Rafael León is CEO of Mirlo, the first telco in Mexico that sells mobile plans with voice agents included. Anabella Guimarey is CEO of El Cero KM, the first 100% online marketplace for new cars in Argentina, run by six people. Jorge Jiménez is CEO of Conectamos, an agents startup that works with some of the largest retailers in Mexico.

The pattern they share is the same: their customer does not buy AI, they buy a service. AI is how they deliver it, and that is why the price gets compared against something else.

We talked about tickets that triple, about an agent that converts twice as well as a human salesperson, and about why small businesses are adopting this faster than corporates.

Structure and Key Highlights

Why Sell It Packaged and Not Loose

• Mirlo started in late 2024 as a digital telco for businesses. The insight was that buying and managing phone plans for a business is a horrible process: going in person, calling for every activation and every cancellation, canceling is almost impossible, and there is zero visibility into how much you pay
• During 2025 they started putting AI inside the plan. Today they sell data, social networks, minutes, and roaming, plus the minutes of an agent that answers the phone with the business context and gets created in less than three minutes
• Their example case: a real estate broker in Michoacán with his cell number stuck on 100 properties, who needs that exact number and cannot change it. He had been with a traditional telco for 10 years
• Rafael’s explicit advantage: I am packaging it for you inside something you are already paying for, and that you have paid for your whole life
• Jorge’s argument for why not to sell it loose: if you sell AI itself, in the consumer’s head you compete with Microsoft, OpenAI, and Google
• And his customer does not ask for AI. They ask to know where their merchandise is

El Cero KM and an Industry Without APIs

• The starting point: buying a new car should be an important and pleasant experience, and it was a poor one. There was no price transparency, and the salesperson knew less about the car than the buyer
• On the other side, the dealership complains that the people who show up are not coming to buy
• Why they could not do it without AI: dealerships handle an enormous amount of information on models, versions, and prices, and they do not have APIs. The information arrives in WhatsApp messages, spreadsheets, and voice notes
• Anabella started doing the data entry herself, and the alternative was to outsource it to a cheap country
• The second finding: an agent sells better than a traditional salesperson because it has patience, it compares across models, it knows every specification, it apologizes when it gets something wrong, and it does not lie
• The knock-on benefit for the dealership: the buyer arrives already knowing which car they want and at what price, and they only need to pay for it

Conectamos and the Choice that Chose Them

• Their reading of the moment: building only software demands a lot of investment with a late return, and the venture capital environment was still complicated
• So they started with the service, with clients who already needed logistics and with an industry they knew
• The differentiator is direct: they need much less headcount, so they can offer better rates
• And that operation gives them the data they use to keep improving the model they now sell separately

The Efficiency Numbers

• Mirlo: the average ticket for the mobile plan alone was 12 to 15 dollars, with margins similar to software. With AI voice minutes it rose to between 30 and 40. Adding WhatsApp, it gets close to 50
• What is interesting for Rafael: they end up with a ticket similar to traditional telephony in a certain segment, but with superpowers
• El Cero KM: they are six people, and Anabella estimates that without AI they would need 30, and they are still not at full capacity
• Their most important metric: whoever talks to Tomás, their agent, is twice as likely to buy as whoever talks to a traditional salesperson. In another cut of the experiments the difference reaches 5.5 times
• And the NPS is better with the agent. She reads comments from buyers thanking Tomás for how well he served them, without realizing he is a bot
• Conectamos: the volume they move today with two or three people is what he used to move with about 30 people five years ago, at another startup in the same field
• The structural consequence: in industries that needed enormous density to be profitable, you can now be profitable with a contained business. They are funding their technology team with that profitability

Where to Start

• Rafael’s filter for the owner of a small business: look for the processes that are repetitive, urgent, and that make you gain or lose money
• At Mirlo those processes were the after-hours calls, the support calls, and the sales calls
• The second factor they analyzed was the economics. Telco is a giant market led by a few companies with a lot of pricing power, and putting technology in let them make the price comparison stop existing
• Jorge’s strategy with corporates: start with something adjacent, not with the core. With a large Chilean retailer they started with one part of the process, it worked, and from there they scaled to the rest
• With the small ones it is the reverse: the more rudimentary their operation is (WhatsApp groups, Excel sheets), the easier it is to structure a workflow and implement it in a couple of weeks
• His recommendation to small businesses: you just have to dare

Where the AI Goes and Where the Human Stays

• Anabella’s criterion: AI has to help where the most mistakes are made and where the dealership invests the most money. That is the salesperson, not the test drive
• How they decide it: experiments with 5, 10, or 15% of the conversations handled by the agent and the rest by salespeople, and then statistics
• Rafael describes the change at Mirlo: at the beginning they sinned by wanting to do everything with AI. Today the agent resolves 90 to 95% of the tickets and the team steps in within seconds when it is needed
• What he calls magic is not only the savings. It is the visibility into everything that is happening
• Anabella calls it human in the loop: not putting a person in the middle, but giving the agent the piece of context that only a human knows and handing the conversation back
• They have seen conversations where Tomás talks for 95% of the time and the customer asks for a call only at the end, on cars of 60 or 70 thousand dollars
• And a new task that appeared: keeping the agent educated every day

Mistakes They Have Already Made

• Conectamos started with a very open and conversational model in its control tower. The operators treated it like their buddy, with very Mexican slang, and they ended up chatting with it at night. They narrowed it with clearer guidelines
• El Cero KM has caught it inventing things: recommending cars it should not recommend, or talking about used cars when they only sell new ones
• Their architecture lesson: a single model did not solve everything. Orchestrating several agents specialized by topic worked better, because the best model for choosing a car is not the best model for answering about financing
• And the recurring moment of humility: days when the agent answers strangely and they do not control the model
• Conectamos’s CTO does not call sprints sprints. He calls them bets, because they do not know if they are going to work. Adapting their logistics model to distribution in wholesale markets seemed far off, and it turned out very well
• Rafael says an agent identified a bug a customer had reported, in seconds, something that would have taken someone on the team hours. He calls it playing on easy mode

How You Control What You Do Not Control

• El Cero KM’s operations center is Slack. Summaries of every conversation and the NPS land there
• Their trick: another agent analyzes each conversation and classifies whether it had high purchase intent or whether it ended up talking about anything else
• Anabella sums it up as AI on top of AI, quoting Guillermo Rauch: the solution to the problems of AI is more AI
• At Mirlo the control is by preprompt. They have templates by industry (personal assistants, health, real estate) with guidelines so the agent does not confront the customer and adjusts the tone
• The little box, as Rafael calls it: Mirlo’s instructions, which the customer then personalizes with their own
• And they are already rolling out integrations with EasyBroker and Shopify, so the catalog and the inventory become the source of truth automatically, instead of an Excel

Interfaces

• Jorge calls it a game changer: many ideas were limited by the rigidity of traditional apps, which also create friction with the person who actually uses them, the operator or the courier
• Their approach is to be agnostic: pull the data from wherever the customer already has it, whether a legacy system or WhatsApp, and put in as little friction as possible
• El Cero KM splits the two halves: choosing the car is visual and needs a marketplace, with photos, specifications, and videos. Buying it is WhatsApp centric
• They have never tried handing out the WhatsApp directly. You enter through the marketplace
• Mirlo noticed the opposite of what they expected: since they launched agents, use of their interface grew a lot, because that is where the knowledge base lives
• Jorge’s clarification about WhatsApp: it is not that it is phenomenal. It is the interface everyone already knows how to use. A corporate wanted to give Microsoft Teams to every truck driver, and he anticipated that it was not going to work

Why Small Businesses Move Faster Than Corporates

• Jorge’s contrast: a month and a half to see a corporate contract, another month for the pricing, and two more to implement. With a small business they have implemented from start to finish in two weeks
• His clients at the central de abastos are small only in structure. They bill hundreds of millions of pesos while operating with WhatsApp and notebooks
• The counterintuitive advantage: they are skipping SaaS, so they do not bring the bad habits of someone who demands a dashboard
• The corporate has budget but needs a champion, and very often the person you talk to is someone in the middle who fears that this is their own job
• Anabella cites the meme of the AI software whose real usage was extremely low, and that still got expanded because they asked the AI to put together a report justifying the usage
• At a small business you talk to the owner, the one who answers the phone at midnight, and for whom every peso hurts
• Her favorite analogy, taken from Mark Cuban: implementing AI in small businesses looks like when computers arrived, with someone knocking on the door to explain what they are for

What the Best Customers Do

• Rafael insists on the onboarding. Today they dedicate one session to each customer, and he recommends that the owner of the business join, with genuine curiosity
• Anabella recommends grabbing a small problem, probably in support or in repetitive tasks, and going all the way to the end. It is easy to throw in the towel the first time
• And putting real budget behind it. Her example is the CEO of a small business who gave his team two dedicated weeks to learn an automation tool, with the obligation of coming out knowing how to use it
• Jorge on his favorites, a pair of cousins who distribute at the central de abastos: they simply dared, and they got involved in every implementation. They ended up helping build the product
• The challenge Rafael sees on the other side: the customer’s expectation grows exponentially by day five, and with AI it is easy to lose focus building a million things made to order

How They See Their Industries in Five Years

• Logistics, according to Jorge: companies are already shutting down. He expects a reconfiguration between those who own the assets and those who own the operational intelligence, plus more fragmentation, with smaller local players that have capabilities only the large ones have today
• Cars, according to Anabella: buying a car is the second most important financial decision after a house, and a lot of people end up buying the wrong car. She sees agents as personal shoppers that make the fit between car and buyer, and a more transparent industry as a consequence
• Telco, according to Rafael: the price per gigabyte is going to keep falling and the mobile plan is going to become more commoditized, but enormous opportunities open up to monetize other services on top. Whoever has the best technology plus a large distribution network wins

Silicon Valley Glossary

People, Companies and References

People Mentioned

• Rafael León: CEO of Mirlo, a Mexican telco with voice agents included in the plan
• Germán: his cofounder at Mirlo
• Anabella Guimarey: CEO of El Cero KM, a 100% online marketplace for new cars in Argentina
• Jorge Jiménez: CEO of Conectamos, AI agents for logistics and retail in Mexico
• Tomás: El Cero KM’s sales agent, whom buyers thank by name
• Guillermo Rauch: of Vercel, cited for the line that the solution to the problems of AI is more AI
• Mark Cuban: cited for the analogy between adopting AI and the arrival of computers at small businesses
• Enzo Cavalie: host of the podcast

Companies and Tools

• Mirlo: mobile plans in Mexico with a voice agent and WhatsApp included
• El Cero KM: an Argentine marketplace for new cars, run by six people
• Conectamos: logistics with agents, profitable at a small volume
• WhatsApp: the shared interface of the three, because it is the one everyone already knows how to use
• Slack: the operations and visibility center for El Cero KM and for Mirlo
• EasyBroker and Shopify: Mirlo’s first integrations, used to feed the agent
• Central de Abastos de Ciudad de México: the distributors Jorge points to as his best customers