$100k to $10M in 5 Months: The Startup That Out-Sold Brazil's Biggest Banks

Domu's Hyper-Growth Playbook

Domu Technology Inc. grew from $100,000 to $10 million in annual recurring revenue across five months. Romero's LinkedIn notes the San Francisco company, with a growing São Paulo presence, sells AI agents that handle debt recovery, customer onboarding, and collections for eight of the world's 20 largest financial institutions. Their agents complete verifiable outcomes — payments, originations, renewals, resolutions — in conversations a regulator can audit line by line.

A three-person sales floor runs a volume playbook that looks more like a call center than a SaaS startup. The team makes 750 dials, sends 250 cold emails, and fires 200 LinkedIn messages every week. They have physically relocated five times in six months, parking themselves next to prospect offices to close contracts. Their first sales development representative doubled his compensation in two months.

Manuel Romero, a GTM engineer who scaled the company from $1 million to $6 million ARR, described the approach as "AI-enabled, automated operating models in high-growth, regulated environments." hardreset.info reports Romero scaled the company from $1 million to $6 million ARR. The company's own LinkedIn updates frame it as a 10x ARR expansion in eight months. GetLatka's estimate puts revenue at $870,000 as of June 2024, with headcount between 5 and 33 depending on the source, a range that signals how fast the org chart is melting and reforming.

Chief Operating Officer Maria Alejandra Pulido built the operational infrastructure that lets the sales motion scale without collapsing delivery. She implemented a real-time supervisor system that flags AI hallucinations on live client calls instantly, letting quality teams intervene before a conversation goes off rails. A generative AI ops dashboard monitors voice, WhatsApp, and chat channels simultaneously. Customized bots handling two-way messaging in Spanish and English lifted engagement 50 percent. At Klarna, her team trained models on transaction histories to predict delinquency and automate personalized nudges, recovering more funds. This approach has cut deployment times for BPO giants like Teleperformance and Alliance One.

The client list reads like a financial services index: Chubb, Amwins, Ascent, Okinus, Starmark, Real Time Resolution. Domu's agents handle millions of live interactions daily for Fortune 500 banks and insurers end-to-end.

The hiring board reflects the strain. Open roles span AE (Cash + Equity + Commission) in São Paulo at $45K–55K, Technical Account Manager - Brazil at $30K–45K, Business Analyst Brazil at $40K–60K, and Software Engineer Brazil at $70K–90K. San Francisco roles run higher: Senior Software Engineer at $150K–210K, UI/UX Product Designer at $120K–160K, Senior Operations Leader at $90K–170K. The board's median salary band sits at $120K across nine salaried roles.

The clearest signal is the Technical Account Manager - Brazil role posted on the Zero G Talent board. Based in São Paulo, remote across the state, it carries a base of USD 30K–45K plus 0.03–0.05% equity, modest by San Francisco standards, but the LinkedIn listing drew 162 applications in its first month. The board shows an 85% response rate and two-hour average reply time. Domu isn't waiting for talent to appear; it's hunting for a specific hybrid: someone who speaks Portuguese fluently, understands Brazilian banking regulation, and can deploy real-time voice AI into legacy telephony stacks. The job description makes the constraint explicit — "real-time voice AI at scale, legacy banking infrastructure, enterprise telephony, and regulated markets across the US and LATAM."

The board's median across nine salaried roles sits at $120K; the band spans $38K–$178K. That spread reflects a team that needs senior engineers and a localized sales motion simultaneously.

The Latin America AI SDR market, valued at $231.2M in 2025 and projected to hit $907.7M by 2030 (31.5% CAGR), is expanding faster than the talent pool can supply. Brazil's slice grows at 30.3% annually. Domu's founders, Nick Diaz and Camila Zancanella, are effectively paying a scarcity premium for reps who can handle Banco Central compliance conversations in Portuguese while debugging SIP trunk latency.

For incumbents — Itaú, Bradesco, Nubank — the message is clear. They're not just competing for engineers anymore. They're bidding against a 50-person startup that moves its sales floor into prospect lobbies.

Pulido emphasizes sustainable scaling, keeping quality while expanding into telecommunications and beyond. But the market isn't waiting. Brazil's banks allocated $8.7 billion to technology in 2024, up 13 percent year over year. Ninety-six percent already use AI; 54 percent have generative AI in production. The applied AI market in Brazil is projected to hit $65 billion by 2035. Domu's sprint isn't an outlier. It's the leading edge of a procurement shift that has incumbents scrambling.

The Latam Financial AI Gold Rush: Why Brazilian Banks Are Prime Targets

Brazil's banking sector spends like a tech company because, increasingly, it is one. The Febraban Banking Technology Survey, conducted by Deloitte across 25 banks representing 85 percent of industry assets, Febraban-Deloitte data shows the 2024 technology budget at R$47.8 billion (US$8.7 billion), a 13 percent jump over 2023. Deloitte's Febraban survey projects that figure to hit R$50.4 billion in 2026, marking a 58 percent climb in five years and a 104 percent surge since 2015. For context, Deloitte's survey found Brazilian banks grew tech spend 97 percent from 2018 to 2023 while global peers averaged 35 percent.

The money flows into infrastructure most markets are still building. Digital channels handle 82 to 83 percent of all banking transactions; mobile alone accounts for 78 percent. Deloitte's figures put Pix users at 150 million (28 percent heavy users) with 240.8 billion transactions processed in 2025, and Deloitte's survey found over 110 million Brazilians conduct more than 80 percent of their financial life on a phone. Pix, the Central Bank's instant-payment rail, counts 150 million users (28 percent of them heavy users) and processed 240.8 billion transactions in 2025. Open Finance adoption, while still modest, has crept from 6 percent of customers in 2022 to an expected 9 percent in 2024, with a third of banks targeting 6–20 percent penetration this year.

Regulators didn't just allow this; they engineered it. The Central Bank mandated Pix, built the Open Finance framework, and is piloting Drex, a wholesale CBDC that 75 percent of surveyed banks expect will unlock smart-contract automation. "With the need to seek an increasingly relevant position towards differentiation, banks can use the Open Finance movement to expand data and knowledge about their customers," said Sergio Biagini, Deloitte's lead partner for financial services, in April 2024. "With the use of technologies such as AI and GenIA, information can be explored more efficiently and accurately to personalize customer relationships."

Banks have the data pipes, the customer habits, and the regulatory green light. They also have the talent base: 45,300 IT professionals, up 22 percent year over year, and a dedicated AI/GenAI upskilling budget of R$29.4 million, a 31 percent increase. The Febraban survey found 96 percent of banks already run AI technologies; 54 percent have deployed generative AI. Early adopters report an 11 percent average efficiency lift across processes, 17 percent in back-office operations, and 12 percent in both customer service and application development. Bradesco's internal coding assistant, BIA Tech, lifted developer productivity 46 percent. Santander cut a 40-hour program conversion to seven minutes. Banco Inter quintupled monthly deployments on Azure AI.

This isn't experimental budget. It's operating budget with measured returns. The Brazil AI in Finance market, valued at $507 million in 2024, is projected to hit $2 billion by 2029. The broader Applied AI market could reach $65 billion by 2035. Vertical agents that plug into Pix, Open Finance, and the mobile-first customer base don't need to sell digital transformation; they sell the next layer of automation on top of the stack Brazilian banks have already paid for.

The kicker: every major bank has named customer experience and security as its top competitive differentiators, 80 percent of institutions, per Febraban. Vertical agents that resolve conversations end-to-end, with line-by-line audit trails, hit both targets at once. That is why Domu's five-month sprint found buyers so fast. The infrastructure was waiting. The budget was allocated. The mandate was live.

Incumbents Strike Back: Nubank, Itaú, and Bradesco's AI Counter-Moves

Brazil's three largest financial institutions are not waiting for vertical AI agents to eat their lunch. Each has moved decisively — Nubank by acquiring a Silicon Valley foundation-model startup, Itaú by embedding generative AI directly into its merchant-acquiring hardware, and Bradesco by building a governed, multi-agent platform that now resolves more than 80 percent of customer interactions without human handoff.

Nubank buys the model layer

In June 2024, Nubank acquired Hyperplane, a two-year-old data-intelligence company co-founded by Daniel Silva, Felipe Lamounier, Rohan Ramanath, and Felipe Meneses. The deal (Hyperplane's $6 million seed round had backed Lachy Groom, SV Angel, Clocktower Technology Ventures, Liquid2 Ventures, Crestone VC, Soma Capital, Latitud, and Atman Capital) gave Nubank a team that had already built proprietary foundation models for financial services. "Nubank's mission since its founding is to fight complexity and empower fanatical customers," founder and chief executive David Vélez said. "Our early investments in AI, coupled with the impressive infrastructure and talent that Hyperplane team has been able to put together, will accelerate our mission."

The strategic logic was data. Nubank processes petabytes daily across more than 100 million customers in Brazil, Mexico, and Colombia — a dataset Hyperplane co-founder Felipe Lamounier called "the perfect home to accelerate our vision of hyper-personalized consumer banking because of the maturity of their data stack and a culture of innovation." CTO Vitor Olivier added that Hyperplane's systems would "seamlessly integrate into our systems, enhancing our ability to analyze vast datasets and personalize our services on a granular level."

Nubank has since moved beyond integration to creation. In a 2026 technical deep-dive, the company revealed it had published its own foundation-model research — the first financial-services firm to do so — and that its v2 model runs on over 500 terabytes of data with four times the context window and four times the training data of v1. The payoff is measurable: "hundreds of millions of NPV being created by new generations of these models because suddenly customers that looked not creditworthy start to look creditworthy under that curve."

The incumbents' advantage isn't just capital — it's the proprietary data moat that lets them train foundation models no vertical-agent startup can replicate.

Itaú puts a co-pilot in every merchant's hand

Itaú Unibanco, Brazil's largest bank with roughly US$500 billion in assets and a 17,000-person engineering organization, took a different route. Instead of buying a model lab, it embedded generative AI into the "Laranjinha+", the orange point-of-sale terminal already ubiquitous at Brazilian merchants. The update, rolled out to all terminal owners in June 2024, lets entrepreneurs trigger voice-command payments without a phone or secondary device. "With faster sales and operational gains, we save entrepreneurs time and improve their operations, allowing them to sell more and boosting the growth of their company," said Angelo Russomanno, director of payments for legal entities. Itaú brands the embedded layer "Itaú Intelligence," positioning the terminal as a sales co-pilot rather than a mere card reader.

The move exploits Itaú's distribution: every new merchant acquiring relationship becomes an AI touchpoint. It also sidesteps the chatbot paradigm entirely — no separate app, no WhatsApp flow, just the device the merchant already touches dozens of times a day.

Bradesco builds the governed multi-agent backbone

Banco Bradesco chose to own the orchestration layer. In November 2023, the bank, working with Microsoft and Avanade, launched Bridge, a multi-agent, technology-agnostic generative AI platform built on Azure AI Foundry, Azure Red Hat OpenShift, and Azure API Management. Bridge powers BIA (Bradesco Artificial Intelligence), the virtual assistant the bank introduced in 2016, across phone, chat, and WhatsApp for both retail and corporate clients. The result: an 80-percent-plus resolution rate on generative-AI-handled conversations.

Bridge's architecture is deliberately modular. Low-code interfaces via Microsoft Power Platform let business teams deploy new agents without engineering tickets, while governance (standardized through Azure Red Hat OpenShift) satisfies the Central Bank's audit requirements. Bradesco's 2016-era BIA becomes just one agent among many, each specialized, each auditable, all orchestrated on a platform the bank controls end to end.

The pattern: own the data, own the surface, own the orchestration

Three distinct plays, one underlying reality: the incumbents are converting their scale into structural AI advantages that pure-play agent startups cannot match. Nubank owns the consumer-data moat and now the model weights. Itaú owns the merchant-acquiring surface and is turning every terminal into an inference endpoint. Bradesco owns the orchestration fabric that lets it swap models, enforce governance, and deploy agents at the pace of regulation rather than the pace of a vendor's roadmap.

For vertical-agent companies like Domu, the signal is clear: the wedge cannot be a better model alone. It must be a workflow the incumbents have not yet absorbed, or a compliance layer they cannot easily replicate.

Beyond the CRM: Auditable AI Agents and the Compliance Gauntlet

Traditional CRM systems were built as digital filing cabinets, places where sales representatives manually logged calls, updated contact information, and moved deal stages. In regulated financial servicing, that architecture creates a structural problem: the record of what happened lives apart from the conversation itself. A collector notes the outcome after the call ends. The audio sits in one system, the disposition in another, the compliance review in a third. When a regulator asks for the complete trail (what was said, when, under which policy, with what result), the institution stitches fragments together after the fact. That gap is where violations hide.

Domu Technology Inc. [/ai-companies/domu-technology-inc] took a different approach. Compliance is not a checkbox, it is the architecture, the company states. Every agent is certified before it speaks to a customer, watched on every word, and audited after every call, with a full evidence trail ready for any regulator. The platform monitors 100% of supported conversations so servicing teams can identify risk, coach operations, preserve evidence, and act while an outcome can still be influenced. Traditional quality assurance reviews only about 1% to 2% of customer conversations, leaving a 98% blind spot. Human-only operations cannot consistently inspect every interaction in real time or turn that evidence into immediate action.

In financial services, getting AI to work isn't the hard part, but proving what it did is. Every conversation can become a compliance question: What happened? Why did it respond that way? Can you show the record? Most AI companies add compliance after the product is built. Compliance for us, is part of the architecture from day one. So when the question comes, you don't scramble to reconstruct what happened. You already have the proof.

The audit trail is granular. Each call produces an exam package: original audio retained, word-level transcript with every speaker and timestamp, the policy applied (Reg F v4, approved rule source), the model action explained and sourced, and a signed export (transcript, source, decision, score) available in one click. The On Air panel shows every active call and every critical flag from the last 30 minutes. Conversations are labeled with the event that followed: payment completed, promise to pay kept, account brought current, application finished. Performance is measured against real outcomes, not call counts.

This matters acutely in Brazil, where the central bank and consumer protection agencies demand traceability for every customer interaction. Nubank, the region's largest digital bank, has processed 73 million AI calls through Domu with 2,400× growth in AI call volume. The platform's financial-servicing controls support monitoring and operational alignment with CFPB expectations, TCPA, FDCPA, and Regulation F, U.S. frameworks that Brazilian institutions mirror when serving diaspora portfolios or partnering with American capital markets. Security certifications and regulatory alignment are distinct: certifications validate defined controls, while regulatory tooling helps teams monitor policies, disclosures, consent, interaction quality, and evidence across supported workflows. Domu does not replace a customer's legal or compliance judgment.

The competitive set underscores the shift. Legacy dialer vendors GetLatka reports TCN at $41.5M revenue, GetLatka's figures put Skit at $22.4M, and GetLatka found Prodigal at $15M. GetLatka reports TCN ($41.5M revenue, 377 employees, founded 1999), Skit ($22.4M revenue, 500 employees, founded 2016), and Prodigal ($15M revenue, 59 employees, founded 2018) bolt AI onto existing call infrastructure. Domu's agentic platform replaces the dialer, the QA layer, the compliance dashboard, and the CRM connector with a single system where the conversation, the evidence, and the regulatory proof are inseparable. "Financial institutions are looking for practical, compliant ways to scale AI," said Nick Diaz [/people/nick-diaz], CEO at Domu. "What matters most is that every conversation operates within policy and is fully auditable."

No human collections floor gets every required legal disclosure right on every call. AI personas can. They negotiate within set limits, take payments on the spot, and handle all calls without a human handoff. The conversation ends with a payment confirmation (paid, confirmation 84-2211) and the record is immutable, timestamped, replayable. For Brazil's banking giants racing to match Nubank's automation, that completeness is the new baseline. The CRM screen is no longer the system of record. The conversation is.

Five months ago, Domu was a $100k ARR startup with a three-person sales floor making 750 dials a week. Today its agents handle millions of conversations for those eight institutions, each call leaving a trail a regulator can read line by line. The incumbents have bought model labs, embedded AI in terminals, and built orchestration platforms of their own. But the wedge that let Domu in — compliance baked into the conversation, not bolted on after — remains the one layer the giants have not yet absorbed. The next time a Brazilian bank pays $10 million for an AI agent, it won't be buying a better model. It will be buying the proof that the conversation happened exactly as the law requires.

Working in AI? Zero G Talent tracks the openings: see every open Domu Technology Inc. role, browse AI jobs, the companies hiring, and the people building the field.