Before Juan Casian built software to hold thousands of customer conversations, he was the person asking a product team to squeeze in a few more. At Atrato, the Mexican financing company he co-founded in college, the refrain was plain: “Talk to more customers. Get more insights.” The team sent surveys, watched net promoter scores and scheduled interviews. On a good week, perhaps three or four people made it onto the calendar. A number came back. An explanation usually did not.
Eight out of ten is crisp enough for a slide and vague enough to be dangerous. Why not ten? What happened before the customer clicked? Which irritation never appeared in the form? Casian later described those product decisions as “guessing with slightly better odds.” There was data, certainly. There was rarely the whole sentence.
That irritation became the seed of Boom AI, the San Francisco company Casian now leads with former Atrato colleagues Sergio Garcia and José Carlos Toscano. Boom’s agents contact customers over WhatsApp, email, SMS and voice. They can chase an abandoned onboarding, interview former users or qualify leads. The pitch is scale, but the more interesting idea is attention: make the customer’s reply as operational as the company’s spreadsheet.
“Every product decision felt like we were guessing with slightly better odds.”Juan Casian on the limits of fragmented customer research
The advantages of not knowing
Casian’s first company began with a problem close enough to touch. He and his Atrato co-founders were students who had been frustrated by their own experiences with banks and credit cards. In Mexico, shoppers without a conventional credit card often had to pay cash for a large purchase, postpone it or accept expensive store financing. Atrato set out to let a shopper apply on a phone, receive an offer and pay a merchant purchase over installments.
The founders had to learn underwriting, fraud, credit scoring, debt, sales and product design. They built the application flow and servicing infrastructure while acquiring merchants. At launch in 2021, the company said it was working with roughly 110 merchants in Mexico. The list of required competencies reads less like a startup plan than an entire financial institution folded into a browser tab.
Casian was 20 when the company started. He has argued that the missing experience was an advantage. Had the founders understood in advance how difficult it would be to raise both capital and debt, or how many calls stood between them and a first customer, the project might have looked politely impossible. Instead, ignorance rationed the difficulty. Each obstacle arrived when it was time to meet it.
This is not a hymn to being unprepared. Atrato handled real money and real repayment decisions, terrain where cheerful improvisation has a short shelf life. The useful naivety came earlier, at the point of permission. The founders began before they possessed every credential the problem appeared to demand, then recruited knowledge as the company required it. Edgar Castillo helped with the risk model, became an investor and later joined the team. Early supporters supplied connections, capital and debt. The company was not built by pretending expertise was unnecessary. It was built by refusing to treat expertise as something all three founders had to own before day one.
Do not confuse a complete map with the next useful move. Large, regulated problems still get solved one merchant, model and application at a time.
The third application
Atrato’s route into Y Combinator was equally unsentimental. The founders applied, interviewed and were rejected. They tried the next batch and were rejected again. On the third attempt, Atrato joined Winter 2021. Casian tells the sequence with a laugh, but his chosen lesson is consistency. He likes the idea that you cannot lose if you keep trying. It is not literally true in every arena, of course. It is very useful when the cost of another application is smaller than the cost of abandoning a working company.
The company’s path also forced Casian out of his original box. He started as CTO and became CEO. The change meant learning how to sell, lead and scale, verbs that contain fewer semicolons and more moods. His public trace from the earlier period still carries the engineer’s sense of humor. The biography on his Stack Overflow account reads: “I code, sometimes it works, sometimes it doesn’t.” It is difficult to improve on that as a description of software or company building.
By his account, Atrato reached about $20 million in annualized revenue before it was acquired. In 2024, he celebrated 48 consecutive months of month-over-month revenue growth and credited the team for the company’s place on LinkedIn’s Top Startups list in Mexico. The clean milestones conceal the messier education: two YC noes, repeated sales calls, rejected assumptions and the move from overseeing a system to overseeing the people around it.
Conversation as infrastructure
Boom reunited Casian, Garcia and Toscano around a problem they had already felt from the buyer’s side. An online business can collect oceans of behavioral data while knowing remarkably little about the person who left. Clickstream analytics show the exit. A conversation can reveal the hesitation. Boom’s current positioning reaches beyond e-commerce support into proactive customer engagement: recover churned users, re-engage drop-offs and gather information before the relationship disappears.
One early episode sharpened the distinction. A Boom customer commissioned a churn study. The conversations suggested that some clients had not simply drifted away; they were angry about alleged misconduct by one of the company’s advisors. The company knew advisor quality could be uneven but had not understood the severity. It began using the tool to evaluate advisor quality on a recurring basis. A survey might have caught the low score. The exchange found the story underneath it.
A metric can locate the smoke. A well-timed question may find the fire.The operating thesis behind Boom’s customer conversations
There is a necessary tension here. Conversation feels human because it is contingent. The next question depends on the last answer. Software at scale prefers repetition. Boom’s bet is that contemporary models can preserve enough of the former while delivering the reach of the latter. The company says an agent can conduct outreach at a thousand times the reach of a manual effort. The important test will not be whether an agent can produce words. It will be whether the exchange returns context a team can trust and act upon.
Casian’s two companies look different at first glance. One financed purchases. The other orchestrates dialogue. Both, however, live inside missing information. Atrato had to decide whether a customer could repay when the conventional credit-card path did not serve them. Boom tries to learn why a customer stalled before that silence becomes lost revenue. In each case the company must make uncertainty legible without pretending it has disappeared.
The shortcut is another person
Casian is a repeat Y Combinator founder who still speaks about the network as a reason to join. That is less redundant than it sounds. Other founders have already solved problems his team is only discovering. One conversation over coffee, he wrote after visiting Campus Founders in Heilbronn, can save weeks spent building the wrong thing.
He also stripped the claim of its velvet rope. The community does not need to be YC or an accelerator. A small group of founders meeting regularly can work. That detail suits the rest of his story. Institutions matter, but the useful unit is still the exchange: a customer explaining why the score was eight, an investor taking a chance on three young founders, a peer warning that the road ahead contains a hole.
There are other small continuities. Before the companies, Casian worked through algorithms, systems and computer graphics at 42 Silicon Valley. He built a desktop app for an investment fund with Electron and Firestore. He helped lead a volunteer project that raised more than 25,000 Mexican pesos for earthquake relief. Each project is a modest machine for coordinating decisions, money or people. The scale changed. The instinct did not.
Boom is young, and its final shape is still being negotiated in those customer conversations. Casian’s own trajectory suggests he is comfortable with revision. Engineer became executive. A health-financing idea widened into buy now, pay later. A team rejected twice entered YC, built for years and sold. Former colleagues assembled around a new blank page and went through YC again.
The comic lesson is that a founder can spend years automating work only to discover that the valuable part was asking someone what happened. The serious lesson is close beside it. Companies do not suffer from a shortage of dashboards. They suffer when the dashboard becomes permission to stop listening. Casian is building Boom around the possibility that listening itself can become software without becoming empty. The next answer belongs, appropriately, to the customer.