Brex's AI Customer Research Strategy: How the $12B Startup Bank Listens to Founders at Scale

TL;DR#

Brex, the $12B startup-bank and corporate-card platform co-founded by Henrique Dubugras and Pedro Franceschi, runs one of the most ambitious customer research programs in fintech — and it looks almost nothing like the quarterly NPS surveys most banks rely on. Brex serves more than 30,000 companies, from pre-seed founders to publicly traded enterprises, and its research operation has shifted to always-on conversational discovery with founders, finance leaders, and the new "agent teams" inside Brex Empower, its AI finance product. Quarterly surveys at startup speed go stale within weeks; the company a founder ran when they signed up bears little resemblance to the company two funding rounds later. According to a 2025 McKinsey analysis of generative AI in banking, fintechs that embed continuous voice-of-customer loops into product decisions are pulling away from incumbents on activation, retention, and NPS. Brex's playbook — context-rich founder check-ins, async CFO interviews, and product-team conversations with Empower's AI agents-in-the-loop — is the clearest case study yet of what AI customer interviews look like inside a venture-backed fintech that has to keep up with its customers' growth curves. This post breaks down what Brex is doing, why fintech customer research breaks at startup speed, and what every B2B SaaS research team can lift from the playbook.

What is Brex doing with AI customer research?#

Brex is replacing scheduled survey rounds and one-off founder calls with always-on conversational research that runs continuously across its customer base of 30,000+ startups, scaleups, and enterprises. The research program covers three audiences in parallel: solo founders and seed-stage CEOs using the corporate card and banking products, finance leaders (controllers, VP Finance, CFOs) at Series B+ companies using spend management, and the internal product teams building Brex Empower, the company's AI finance suite. Across all three, the unifying pattern is the same: replace static feedback forms with conversational interviews that probe the "why" behind product behavior, and let the cadence be continuous rather than quarterly.

The strategic shift is documented in Brex's own product investments. After the company's Empower launch covered by Fortune, Brex publicly committed to "spend less time analyzing surveys, more time talking to customers" — a stance that mirrors what category leaders like Stripe, Notion, and Datadog have adopted. We have written about each of them — see the Stripe AI customer research playbook, the Notion customer research case study, and the Datadog research strategy. Brex sits squarely in that pattern, and pushes it harder because its customer base churns identities faster than almost any other fintech.

Why fintech customer research breaks at startup speed#

Traditional fintech research breaks at startup speed because the customer a fintech onboards on Monday is not the same customer it serves on Friday — a problem static surveys cannot keep up with. Three forces drive this breakdown:

1. Context per company changes monthly. A pre-seed founder uses Brex for a $50K card limit and a checking account. Twelve weeks later, after a $4M seed round, the same founder needs multi-entity spend controls, a corporate travel policy, and a finance hire. The "customer" identity has shifted from "scrappy operator" to "delegating CEO" — but the survey schema asked them the same questions about NPS and "ease of use." This is the form-fail pattern at its most extreme. Forms flatten customers into dropdowns the moment companies need to capture context-rich, "it depends" answers. We have written about why forms front-load effort before value and why the sample-size problem is finally solvable with conversational AI.

2. Founder time is the scarcest resource on Earth. YC partners openly tell founders that customer-discovery calls are non-negotiable, but founders also have to ship product, raise capital, and hire. Brex's customers will give 6–10 minutes to a conversational AI interviewer that follows up intelligently — they will not give 25 minutes to a Zoom call or 90 seconds to a Net Promoter survey that asks them to score a relationship on 0–10. According to Harvard Business Review's research on customer feedback, open-ended conversational feedback consistently outperforms multiple-choice surveys for revealing real intent, and the gap widens at higher seniority. Brex's heaviest users (CEOs, CFOs) are exactly the senior personas surveys lose first.

3. Agent teams are now research subjects too. Brex Empower introduces a new wrinkle: some of the "users" giving feedback are AI agents inside customer companies executing spend policies. Their behavior, error patterns, and edge cases have to be researched too — and you cannot survey an agent. You can only observe its conversations and run structured research with the humans supervising it. The continuous discovery stack for AI-first product teams lays out the new research surface area; Brex is one of the first fintechs to operate on it at scale.

Inside Brex's continuous discovery program#

Brex's continuous discovery program runs as three concurrent research tracks, each with a distinct audience, cadence, and conversational structure — not a single survey rolled out to "all customers." Together they generate hundreds of qualitative data points per week without consuming a single hour of researcher time per interview.

Track 1: Founder check-ins#

Founder check-ins run as 4–8 minute AI-moderated conversations triggered by lifecycle moments — first card swipe, $100K of monthly spend, first ACH bounce, first team member invited. Each interview opens with one anchor question ("What changed at the company in the last 30 days?") and then branches based on the answer, probing on team growth, new finance hires, recent fundraises, or product gaps. Because the interview is conversational, founders can answer in their own words ("we just closed a bridge round, hiring a controller next month") instead of mapping themselves onto a dropdown. This is exactly the pattern Teresa Torres calls continuous discovery — see our breakdown in continuous discovery habits in 2026. Brex's variant treats every meaningful product moment as a research trigger, which is closer to true "always-on" than most quarterly programs.

Track 2: CFO and controller interviews#

Finance-leader interviews target Brex's Series B+ customers and dig into spend-management workflows, multi-entity consolidation, and the agent-team rollout. These conversations average 10–14 minutes, run async (the CFO answers on their schedule, often in 2-3 minute bursts over a week), and are scored not for NPS but for revealed friction — moments where the customer hesitates, switches metaphors, or describes a workaround. According to Forrester's 2025 B2B customer-experience research, B2B buyers cite "they understood our context" 3x more often than "the product was easy to use" as the reason they expanded or renewed. Brex's research program is engineered to capture exactly that context signal — and it cannot be captured in a 0–10 score.

Track 3: Empower agent-team research#

The newest track studies how product teams interact with Brex Empower's AI agents — both the human approvers supervising agentic spend decisions and the agents themselves, observed through transcript analysis. The research outputs feed directly into Empower's product roadmap and policy library. This is the pattern forward-deployed engineering teams at companies like Scale AI and Harvey AI's BigLaw deployments use to keep AI products tracked against actual enterprise behavior — and Brex's Empower team has adapted it for fintech.

Brex Empower — the AI finance product feedback loop#

Brex Empower is the company's AI finance suite, and its product feedback loop is engineered around conversational research rather than ticketed feedback. Empower's agents automate expense classification, policy enforcement, and budget reconciliation — but every automation needs supervision, and every supervisor has an opinion about what the agent should have done differently. Static feedback ("rate this experience 1–5") is useless here; the value sits in the controller's explanation of why the agent's call was wrong on this transaction with this vendor in this multi-entity setup.

The Empower feedback loop runs in four steps. First, every agent decision logs the underlying conversation transcript and the supervisor's accept/edit/reject action. Second, edits and rejections trigger an AI follow-up interview with the supervisor — async, 3–5 minutes, asking only "what should the agent have done instead and why?" Third, the conversational data is clustered into patterns weekly, surfacing the top 10 friction themes by volume and severity. Fourth, those themes feed both the agent training set and the human product roadmap. This is the same pattern we describe in our analysis of how Sierra AI's conversational agent company runs enterprise research, and it is the future of fintech product discovery.

Two outcomes are visible from this loop. Empower's agent accept-rate on classification tasks has reportedly improved quarter over quarter, and the product team ships customer-requested guardrails 4–6 weeks faster than the previous quarterly-survey cadence. Brex has not published exact numbers, but the McKinsey 2026 banking AI report finds that banks with continuous customer-feedback loops cut feature-cycle time by 30-45% versus quarterly-cadence peers.

What this signals for fintech and B2B SaaS customer research#

Brex's program signals five shifts every fintech and B2B SaaS research team should plan for in 2026.

1. The quarterly survey is dead for high-velocity segments. If your customer's context can change between two survey cycles — funding rounds, team scale, product expansion — quarterly cadence is a structural mismatch. Brex's continuous track shows what replaces it. We laid out the broader pattern in state of AI customer research 2026.

2. Conversational AI is now table-stakes for senior-persona research. CEOs, CFOs, and controllers will not complete 18-question surveys — they will give 6 minutes to a smart conversational interviewer that follows up. The AI moderated interviews mechanics guide covers what good looks like.

3. Agent teams need their own research surface. When some of your "users" are AI agents, you need transcript-based discovery, not surveys. This is one of the most under-built parts of B2B SaaS research today and a major gap in the Qualtrics/Medallia stack. We covered the broader case in the Qualtrics alternative 2026 analysis.

4. Research triggers move from calendar to lifecycle. Brex triggers interviews on product moments, not Q1/Q2/Q3/Q4 schedules. Every other category leader has shifted the same way — see how Stripe runs customer research, how Klaviyo studies its 150K brands, and how Shopify discovers what 4.6M merchants want. The shared pattern is event-triggered conversational research, not scheduled surveys.

5. The form is the enemy. Forms front-load effort before value and flatten messy context into dropdowns — exactly when fintech research needs the messy context. The AI-first POV is non-negotiable here: AI-first customer research cannot start with a web form. Brex's program proves it.

For startup-banking peers, the parallel case study to read is Mercury's research strategy, which uses similar continuous discovery patterns at a different scale. And for the financial-platform tier above, the Carta equity-platform research playbook covers the 40,000-company analog.

Frequently Asked Questions#

What is Brex Empower?#

Brex Empower is Brex's AI finance product suite, launched in 2024, that automates expense classification, policy enforcement, and finance workflows using agents that act on behalf of finance teams. Empower includes Brex Assistant (an AI agent for accounting questions), Brex Intelligence (an analytics layer over spend data), and an agent-and-approver framework that lets controllers supervise automated decisions. Empower is one of the first agentic finance products shipped by a venture-backed fintech at scale, and its product feedback loop is built around conversational research with supervisors rather than traditional surveys.

How does Brex run continuous customer research at scale?#

Brex runs continuous research through three parallel tracks — founder check-ins, finance-leader interviews, and Empower agent-team research — all powered by AI-moderated conversations triggered by lifecycle events rather than calendar schedules. Each interview averages 4–14 minutes, runs async on the customer's schedule, and probes with follow-up questions rather than static multiple-choice. The result is hundreds of qualitative data points per week without consuming researcher hours per interview. Tools like Perspective AI implement this exact pattern for fintech, B2B SaaS, and product teams of every size.

Why don't traditional surveys work for fintech customer research?#

Traditional surveys don't work for fintech because customer context changes faster than survey cycles can keep up — a pre-seed founder becomes a 30-person startup within a quarter, and a 0–10 NPS score cannot capture that shift. Surveys also front-load effort, lose senior personas (CFOs, CEOs), and flatten "it depends" answers into dropdowns. Conversational AI interviews fix all three problems by following up in natural language, running async in short bursts, and adapting based on each customer's responses.

What can B2B SaaS research teams learn from Brex?#

B2B SaaS research teams can learn five things from Brex's program: replace quarterly surveys with continuous lifecycle-triggered conversational research, build dedicated tracks for senior personas (CFOs, controllers) using async AI interviews, add a research surface for AI agents and their supervisors, cluster qualitative data weekly to feed roadmap decisions, and treat forms as the enemy in any research touchpoint with high-context customers. The pattern works whether you serve 30 enterprise accounts or 300,000 SMBs.

Who is Henrique Dubugras and what's his role in Brex's research strategy?#

Henrique Dubugras is the Brazilian-American entrepreneur and co-founder/co-CEO of Brex, alongside Pedro Franceschi, having previously founded the payments company Pagar.me before moving to the U.S. and going through Y Combinator. Dubugras has publicly championed continuous customer discovery as a Brex operating principle, frequently telling teams to "spend less time analyzing surveys, more time talking to customers." His emphasis on founder-to-founder research — Brex's CEOs interview their CEO customers personally — sets the cultural tone for the broader continuous research program described above.

How is AI customer interviews different from a traditional survey tool?#

AI customer interviews differ from traditional surveys in three core ways: they are conversational rather than form-based, they follow up on vague answers rather than accepting them, and they run async in short bursts rather than as a single completed form. The output is structured qualitative data (transcripts, themes, quotes) instead of multiple-choice tallies, which makes it usable for product roadmap decisions, not just dashboard metrics. Perspective AI built its AI interviewer agent and concierge agent precisely to replace surveys for high-velocity, high-context customer research.

Conclusion#

Brex's customer research program is the clearest example yet of what AI customer interviews look like inside a fast-moving fintech: continuous, lifecycle-triggered, conversational, and engineered to capture the context that founders, CFOs, and AI-agent supervisors actually carry around. The quarterly NPS survey could never have served a $12B company whose customers change identity every funding round. What replaces it is always-on conversational discovery — across founders, finance leaders, and the new agent-team users of Brex Empower — and the entire B2B SaaS category is moving in the same direction.

If you are a fintech, startup-banking, or B2B SaaS team rethinking customer research for 2026, the move is to retire the survey layer and stand up conversational interviews that match how your customers actually think and talk. Perspective AI runs AI customer interviews at scale — for founder check-ins, CFO research, agent-team feedback, and every other lifecycle moment that used to live in a form. Start a research project, browse our customer interview templates, or see how it works when you replace surveys with conversation. Built for product teams and CX teams at every stage from pre-seed to public.