Feb 5, 2025 · 30m · a16z
How AI is Powering Payments, with Greg Ulrich of Mastercard
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of a16z In the Vault, Greg Ulrich, Chief AI and Data Officer at Mastercard, joins Marc Andrusko to discuss how the payment giant integrates traditional machine learning and Generative AI. Ulrich details Mastercard's strategic framework, organizational governance, data trust standards, and emerging trends like multimodality shaping financial services.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The host as informed peer 2.3 Guest teaching 3.4 Guest disagreement 0.1 The host pushing back 0.0
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 24:18 Gently rejecting the premise of unbridled adoption excitement
Greg softly challenges Mark's prompt about ecosystem enthusiasm by clarifying that adoption is far from unbridled due to accuracy and regulatory risks in financial services.
Hardest push from the host ▶ 18:37 Inquiring if AI ROI requires new evaluation machineryMark presses beyond general corporate governance to ask whether AI investments require a completely separate ROI evaluation framework compared to standard vendor management.
Biggest teaching moment ▶ 27:00 Explaining the significance of model self-awareness of limitsGreg educates the listener on why reasoning models that say 'I don't know' represent a monumental leap forward for enterprise trust compared to confidently wrong models.
The host holds their own ▶ 10:59 Articulating startup data friction versus incumbent trust advantageMark demonstrates strong enterprise software domain expertise by identifying the core trust hurdle early-stage AI startups face when asking clients for data.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| Greg Ulrich's Background and Career Journey | 1 | 2 | 0 | 0 | Mark warmly introduces Greg and jokes about a16z taking credit for his role. Greg outlines his career path from university to non-profits, predictive analytics at APT, and strategy at Mastercard. | |
| Traditional Machine Learning vs. Generative AI | 2 | 3 | 0 | 0 | Mark frames the shift from traditional machine learning to generative AI post-ChatGPT. Greg explains that traditional ML remains superior for structured data and forecasting, while Gen AI serves knowledge management and unstructured data. | |
| Strategic AI Framework and Key Mastercard Deployments | 3 | 4 | 0 | 0 | Mark demonstrates background research by referencing Mastercard's recent product announcements. Greg outlines his core framework—Safer, Smarter, More Personal, Stronger—and details applications like Decision Intelligence, Shopping Muse, and RAG onboarding assistants. | |
| Data Safeguarding and Building Trust in Enterprise AI | 3 | 3 | 0 | 0 | Mark articulates the primary barrier early-stage AI startups face regarding enterprise data sharing versus incumbent trust. Greg details how Mastercard approaches startup partnerships through programs like Start Path while maintaining strict data governance. | |
| Hub-and-Spoke Organizational Governance and ROI | 3 | 4 | 0 | 0 | Mark introduces organizational governance concepts like hub-and-spoke models and business unit P&L dynamics. Greg explains how enterprise AI coordination avoids duplication while allowing business units to innovate, and how ROI is tracked via developer efficiency and employee satisfaction. | |
| Staying Informed on External AI Developments | 2 | 3 | 0 | 0 | Mark inquires about ritual practices for tracking external technological developments amidst internal operational duties. Greg shares how he consumes external media and brought seven external AI leaders and academics to brief Mastercard's board. | |
| Ecosystem Adoption Sentiment and Risk Mitigation | 2 | 4 | 1 | 0 | Mark asks if ecosystem sentiment is characterizable as unbridled excitement. Greg offers a polite counter-perspective, noting significant enterprise caution due to risk, hallucination concerns, and regulated industry standards. | |
| Future AI Horizons: Multimodality and Reasoning | 2 | 4 | 0 | 0 | Mark asks what future technological developments excite Greg most. Greg highlights multimodality across financial workflows and the critical breakthrough of reasoning models recognizing the boundaries of their knowledge. |
Statements from this episode (11)
Ulrich: Multimodal AI is uniquely valuable for financial invoice reconciliation
“When we think about things in financial services, big invoice reconciliation, bill pay, you'll have these things come in as PDFs, you'll have them come in as images, you'll have them come in as text in a variety of ways. As we start looking at things holistica…”
Ulrich: Mastercard has used AI for fraud detection for decades
“We've been using AI for decades. It's inherent in everything we do in fraud detection, right?”
Ulrich: Traditional ML beats GenAI for structured data and fraud management
“If you have structured data, you're doing forecasting models, a lot of the fraud management, traditional artificial intelligence, machine learning, Is going to be more efficient, more effective, and certainly more cost effective as a way to do it.”
Ulrich: Mastercard scores real-time transaction fraud in milliseconds using GenAI
“The product we have is called Decision Intelligence, and that's where in the milliseconds we have between when a transaction passes through our network from when you tap to buy something in a store, it passes through MasterCard to go to the issuer. We're provi…”
Ulrich: Mastercard's Shopping Muse uses GenAI for online retail recommendations
“In personalization, we have something called Shopping Muse, which is basically enabling the in-store experience online. So you have a chatbot effectively where you can type in your own language and ask for recommendations. ... It's allowing you to interact lik…”
Ulrich: Mastercard uses RAG and internal databases for customer onboarding AI
“What we've done is we've created a digital assistant that uses RAG and points to all these technical databases, the Q and A's we've had over time, where we've tested and fine-tuned these models to enable customers to onboard our products easier by automating a…”
Ulrich: Centralizing all enterprise AI functions into one department is foolish
“Artificial intelligence, like electricity, like it, it's just a part of so much of the organization. It would be foolish, dangerous, and probably impossible to try to strip out all those elements and centralize them into one place.”
Ulrich: Mastercard brought seven external experts to brief its board on AI
“When we presented to our board of directors on AI in September, the way we structured is I had seven external speakers come in, right? To give different perspectives from an investor perspective. The CEO of one of the leading LLMs. We partnered with Databricks…”
Ulrich: Generative AI solutions are not yet ready for direct consumer use
“There's enthusiasm, but I, it's definitely not unbridled as people are very worried about the accuracy, the hallucination, the efficacy of these, and particularly as you start pointing these to customer facing solutions, there's a very high bar on what that's …”