Marqeta's AI Strategy: How the Card-Issuing Platform Listens to Developers and Customers in 2026

TL;DR#

Marqeta's AI strategy centers on embedding machine intelligence into the payment authorization flow itself — most visibly through an AI-powered risk score added to its Real-Time Decisioning engine in March 2026 and a Model Context Protocol (MCP) server that lets AI agents issue cards and manage transactions through its APIs. The card-issuing pioneer processed $382.5 billion in total processing volume in 2025 (up 31% year over year) on $624.9 million in net revenue, and reached GAAP net income profitability in Q1 2026 with TPV of $112 billion. But Marqeta's customers are developers and product teams at fintechs like Block's Cash App, Ramp, DoorDash, and Klarna — people whose reasons for expanding, stalling, or churning are technical, subtle, and almost never captured by a support ticket or an NPS survey. That is the gap. AI in the transaction stream is solving fraud and latency; it is not telling Marqeta why a developer chose a competitor's BIN sponsor or why a program manager hesitated to migrate. Conversational AI customer research — interviews that follow the reasoning, not a form that flattens it — is how a developer-first platform closes that loop.

What is Marqeta's AI strategy?#

Marqeta's AI strategy is the company's plan to embed artificial intelligence directly into payment infrastructure — real-time fraud decisioning, agentic-commerce APIs, and developer tooling — rather than bolting AI onto a separate analytics layer. It is an infrastructure-level approach: Marqeta builds AI into the moment a card transaction is authorized, and into the APIs that AI agents themselves now call, positioning the platform as the payment rails for both human-led fintechs and machine-led agentic commerce.

For B2B platform businesses, an AI strategy has two halves that are easy to conflate. The first is AI in the product — the fraud models, the autonomous agents, the latency improvements that customers buy. The second is AI in understanding the customer — how the company learns why developers integrate, expand, or leave. Marqeta has moved aggressively on the first. The second still runs largely on tickets, account reviews, and surveys, which is exactly where the most defensible competitive insight gets lost.

Marqeta by the numbers: a developer-first card-issuing platform#

Marqeta is one of the largest modern card-issuing and embedded-finance platforms, built API-first for developers rather than for end consumers. The scale is substantial and the customer base is concentrated among technical teams:

• $382.5 billion in total processing volume (TPV) in fiscal 2025, up 31% year over year, according to Marqeta's 2025 annual report and Q4 earnings release.
• $624.9 million in net revenue for 2025, up 23% year over year.
• $112 billion in TPV in Q1 2026 (up 33%), with $166 million in net revenue and the company's first quarter of GAAP net income profitability, per Marqeta's Q1 2026 results.
• Net revenue retention historically between 110% and 120%, reflecting expansion within existing fintech accounts.
• Roughly 45–50% of 2025 net revenue concentrated in a single segment of large fintechs led by Block's Cash App — a concentration that makes understanding why those technical buyers stay or grow existentially important.

The customers behind those numbers are not consumers filling out a checkout form. They are engineers at Ramp issuing virtual and physical cards with custom spend limits, product teams at DoorDash and Uber wiring instant payouts, and program managers at companies like Klarna, Bitpanda, and Perpay deciding whether to migrate an entire card portfolio onto Marqeta's rails. Their needs are specified in API design choices, latency tolerances, compliance requirements, and BIN-sponsorship terms — nuance that a five-point satisfaction scale was never built to hold. This is the same dynamic we mapped for other API-first companies in our analysis of how Stripe runs AI customer research across 4 million businesses and Plaid's approach to listening across 8,000 fintech developers.

Where Marqeta uses AI today#

Marqeta currently applies AI in three production areas: real-time fraud and risk decisioning, agentic-payment APIs, and developer-facing platform tooling. Each is a genuine product investment — and each illustrates why "AI in the transaction" is a different problem from "AI in understanding the customer."

AI-driven risk decisioning in the authorization flow#

In March 2026, Marqeta added an AI-powered risk score to its Real-Time Decisioning (RTD) engine, scoring transaction risk at the exact moment of authorization. According to Marqeta's announcement, the score evaluates 300+ real-time transaction attributes against historical behavioral patterns, adapting to each customer's cardholder base and market shifts. The business case is sharp: it is designed not just to block fraud but to reduce false declines, which quietly cost merchants revenue. The timing matters — global payment fraud is projected to rise 153% between 2025 and 2030, per industry estimates cited alongside the launch. Marqeta's RiskControl suite now spans KYC, 3D Secure, RTD, and end-to-end disputes management.

Agentic payments and the MCP server#

Marqeta built a Model Context Protocol (MCP) server so AI agents can connect directly to its payment APIs — issuing a card, reviewing transactions, even filing a dispute autonomously. MCP is the open standard introduced by Anthropic for connecting AI applications to external systems; Marqeta's implementation gives agents a standardized, secure integration layer into the full card lifecycle. It is a clear bet that agentic commerce — software acting on a user's behalf — becomes a meaningful share of transaction volume, a view echoed in Everest Group's research on agentic payments reinventing payments for the AI era.

Developer experience as a product surface#

Marqeta's growth has always run on developer experience — well-designed APIs, a sandbox, and documentation that engineers can self-serve. CEO Mike Milotich, confirmed as permanent CEO in September 2025 after serving as CFO and interim chief executive, has framed the strategy as "innovation and profitable growth," with a specific push to make it easier for issuers to migrate onto the platform after the successful migration of Klarna's card program in Europe. New CTO Lukasz Strozek joined in May 2026 to lead that engineering agenda. The throughline: Marqeta wins or loses on whether developers find the platform fast, flexible, and trustworthy to build on.

The B2B listening gap: why tickets and surveys miss the "why"#

Marqeta's customer-understanding problem is structural: the signals it collects today — support tickets, periodic account reviews, and NPS-style surveys — capture what happened, not why a technical buyer made a decision. This is the bottleneck that the most sophisticated payment infrastructure in the world cannot solve from the transaction stream alone.

Consider the kinds of questions that actually determine Marqeta's net revenue retention:

• Why did a fintech's engineering lead choose a different BIN sponsor for its EU expansion — was it timeline, compliance comfort, pricing, or a single bad sandbox experience?
• What made a program manager hesitate for two quarters before migrating a live card portfolio, even after committing on paper?
• Where did a developer hit friction in the first 48 hours of integration that never became a ticket because they just... tried a workaround and silently downgraded their plans?

None of these surface cleanly in the existing toolkit. Support tickets are filed only when something breaks badly enough to warrant the effort; the industry-wide reality is that most dissatisfied customers never complain and simply reduce usage. Surveys flatten the answer into a dropdown before the respondent has finished a thought — which is the core problem we argue AI-first customer research cannot start with a web form. And account reviews are mediated by a salesperson's recollection, not the developer's own words. For a platform whose differentiation is technical subtlety, losing the technical reasoning is losing the whole game. The pattern repeats across infrastructure companies — we documented the identical dynamic for Twilio's 10 million developers and Databricks' forward-deployed-engineering model.

What conversational AI customer research unlocks for Marqeta-style platforms#

Conversational AI interviews close the B2B listening gap by letting developers and product teams explain their reasoning in their own words, at scale, with an AI interviewer that follows up on every vague or high-value answer. Instead of forcing a program manager to rate "migration ease" from 1 to 5, an AI interviewer agent asks what specifically slowed the migration, then probes the answer — surfacing that the real blocker was uncertainty about EU BIN-sponsorship timelines, not the API at all.

This is the difference between a metric and a mechanism. NPS tells Marqeta a number went down; a conversation tells it which part of the developer journey broke and why. That is the same reason we argue teams should move beyond NPS to capture the reasoning behind the score, and why a modern voice-of-customer program is built on conversation rather than collection.

For a developer-first platform, three use cases are immediately high-leverage:

1. Migration-decision research. Interview program managers who almost migrated and those who did, at the moment of decision, to learn what closed or killed the deal. This directly protects the 110–120% net revenue retention that drives the business.
2. Onboarding-friction discovery. Run conversational onboarding intake for new developer accounts so the first-48-hours friction gets captured as it happens — not reconstructed from a ticket that never gets filed.
3. Expansion and churn signals. Use an always-on concierge agent to interview accounts showing usage shifts, capturing the "why now" behind expansion or stall before it shows up in the revenue numbers.

Because Perspective AI can run hundreds of these interviews simultaneously, a product team gets qualitative depth at survey scale — and a CX team gets the early-warning system that tickets can't provide. For teams assembling the broader toolkit, our guide to the customer research stack modern product and CX teams actually use maps where conversational research fits, and our buyer's framework for AI customer-engagement software covers how to evaluate it.

The 2026 context: AI in the product is not AI in the relationship#

The lesson from Marqeta's 2026 moves is that infrastructure companies are racing to put AI inside the transaction while still listening to customers with tools built for a survey-era cadence. Marqeta's AI risk score, MCP server, and agentic-payment APIs are real, shipped, and impressive. They make the product smarter. None of them make the company smarter about why a developer chose it — and for a platform whose moat is technical trust and developer experience, that second kind of intelligence is the one competitors can't copy.

This is the broader thesis we develop in our overview of AI-powered customer experience from first touch to renewal: the companies that win the next phase are the ones that apply conversational AI to understanding the customer with the same seriousness they apply it to serving the customer. Marqeta has the data, the developer relationships, and the AI ambition. The unlock is pointing some of that ambition at the reasoning inside its own customer base.

Frequently Asked Questions#

What is Marqeta's core AI product?#

Marqeta's most prominent AI product is the AI-powered risk score it added to its Real-Time Decisioning engine in March 2026. The score evaluates more than 300 real-time transaction attributes at the moment of authorization to detect fraud and reduce false declines. Marqeta also operates an MCP server that lets AI agents call its payment APIs to issue cards and manage transactions for agentic commerce.

How big is Marqeta?#

Marqeta processed $382.5 billion in total processing volume in 2025, up 31% year over year, on $624.9 million in net revenue. In Q1 2026 it reached $112 billion in TPV and recorded its first quarter of GAAP net income profitability. Its customer base is concentrated among large fintechs, with roughly 45–50% of 2025 net revenue tied to a single segment led by Block's Cash App.

Who are Marqeta's customers?#

Marqeta's customers are developers and product teams at fintechs, neobanks, and commerce platforms rather than individual consumers. Named customers include Block's Cash App, Ramp, DoorDash, Uber, Klarna, Bitpanda, and Perpay. Because these buyers make technical, API-level decisions, understanding their reasoning requires deeper research methods than satisfaction surveys provide.

Why don't surveys work for understanding developer customers?#

Surveys fail for developer customers because they flatten technical reasoning into ratings and dropdowns, and they rarely capture the "why" behind a decision. A program manager who hesitates to migrate a card portfolio or a developer who hits silent onboarding friction won't explain that nuance on a 1-to-5 scale. Conversational AI interviews close this gap by following up on vague answers in the customer's own words.

What is agentic commerce, and why does Marqeta care about it?#

Agentic commerce is software acting autonomously on a user's behalf to make purchases and manage payments. Marqeta cares because it positions the company's APIs as the rails AI agents use to issue cards, review transactions, and file disputes. Its MCP server, built on Anthropic's open Model Context Protocol standard, gives agents a secure, standardized way to access the full card lifecycle.

How can a card-issuing platform run customer research at scale?#

A card-issuing platform can run customer research at scale by deploying AI interviewer agents that conduct hundreds of conversations simultaneously and follow up automatically on high-value answers. This delivers qualitative depth at the volume of a survey, capturing migration-decision reasoning, onboarding friction, and expansion signals that tickets and NPS miss. Perspective AI is purpose-built for exactly this conversational-research model.

Conclusion: turn AI toward the customer, not just the transaction#

Marqeta's AI strategy is a strong proof point that embedding intelligence into payment infrastructure works — AI-driven risk decisioning, an MCP server for agentic payments, and a relentless focus on developer experience have helped drive $382.5 billion in 2025 TPV and the company's first profitable quarter in early 2026. But the moat for a developer-first platform is technical trust, and trust is earned by understanding why customers build on you, expand, or walk away. That understanding does not live in a ticket queue or an NPS trend line; it lives in the reasoning, and reasoning only comes out in conversation.

Perspective AI lets product and CX teams at platform businesses run AI-powered customer interviews at scale — capturing the technical "why" that surveys flatten and tickets never record. Start a research study in minutes, see how it works across existing studies, or explore what an AI interviewer can do for your developer and customer base.