Episode Overview

Podcast: Y Combinator Startup Podcast
Episode: This Startup Secretly Detects Fraud For Fortune 500s
Date: March 31, 2026
Main Theme:
This episode features an exclusive announcement: the fraud and compliance AI company Variance comes out of stealth, revealing its $21-million Series A. Co-founder Karine shares how Variance automates high-stakes fraud detection and compliance for Fortune 500s—often as their “secret weapon.” The discussion explores the technical challenges, real-world impacts, and the founders’ mission-driven journey from Apple engineers to running a lean, high-impact startup.

Key Discussion Points & Insights

1. Stealth Mode, Big Announcement, and Secret Customers

• Announcement: Variance exits stealth after three years, powering fraud detection and compliance for Fortune 500s and major platforms like GoFundMe.
• Series A: Raises $21M (00:30).
• Secrecy Rationale: Variance handles highly sensitive use cases; publicizing customer applications could aid fraudsters.
  • “We're building systems that are often used by the bad guys, but we're building them for the good guys.” — Karine (01:34)
• Customers hide their use of Variance as an advantage in the ongoing “cat and mouse game” with fraudsters (01:57).

2. What Variance Does: Use Cases and How the Tech Works

• Core Product: Purpose-built AI agents automating content and identity review, risk, and compliance at scale (00:49).
• GoFundMe Example:
  • Detects scam fundraisers during crises (e.g., the sudden death of a public figure leading to a spike in fraudulent campaigns) (03:52–04:59).
  • AI agents validate user identity, investigate fundraiser legitimacy, and ensure legal compliance before campaigns go live (05:05–05:48).
• Other Applications:
  • Marketplace identity checks (e.g., delivery driver onboarding via selfie/ID, KYB—know your business—verifications).
  • Complex graph analysis to surface hidden risk (e.g., shell companies linking to sanctioned actors) (06:20–07:12).

3. Data — The Core Technical Challenge

• Data Integration: Data often scattered or hidden in unstructured formats, legacy UIs, or disparate systems (09:35–10:58).
• Solution: Agents extract, unify, and reason over petabytes of cross-system, cross-format data—sometimes even scraping internal tools (11:18–11:45).
• AI Agent Evolution: Unlike old rule engines or classifiers, modern agents reason across open web and internal data, delivering “self-healing” fraud systems (12:07–13:58).

4. Real-World Impact: From Preventing Fraud to Saving Lives

• Political Misinformation & Threat Detection:
  • Detected coordinated, state-sponsored fraud rings during elections using relational, contextual analysis across data graphs (14:22–15:29).
  • “…able to detect really complex fraud rings… pushing one narrative…” — Karine (14:25)
  • Identified credible threats of physical harm, potentially preventing real-world violence: “Once it's detected and investigated by Variance, it's usually going to be in the hands of law enforcement.” (16:09).

5. Variance’s Team and Engineering Culture

• Size: 12 people, 5 engineers (16:30).
• Lean Output: With heavy use of coding agents, five engineers produce output comparable to a 25-person team (19:01–19:28).
• Ownership: Every team member is hands-on; engineers “own” end-to-end solutions. “Everyone is a manager of a small team of AI agents, which is really interesting.” (19:29)
• Innovative Use of AI Agents: Even non-technical staff (customer success) can build/ship basic features using AI tools, bypassing traditional engineering bottlenecks (19:29–20:09).

6. Founding Story: Mission-Driven from Day One

• Founders: Karine (data engineer) and Michael (ML engineer) met at Apple’s fraud engineering as a service team, with complementary skillsets (20:18–20:54).
• Origin Motivation: Deep conviction to solve a specific industry pain observed first-hand—“We always felt a really strong sense of duty to put our very specific and quite rare pair of skill sets to the good of the industry.” (29:36).
• YC Application: Sparked by desire to see this product realized; not a generic startup story—hyper-focused on this niche (20:54).

7. First Customer, Early Struggles, and Startup Hardships

• Landing First Customer (IAC):
  • Eight months of persistence; built system for compliance at Ask Media Group under IAC holding (22:24–23:39).
  • Automated previously human-moderated workflows that were “semi-impossible” for classifiers (23:23).
  • Pre-dated ChatGPT; founders adapted rapidly as OpenAI models improved, impacting their tech stack and costs in real-time (24:06–24:15).
• Near-Fatal Founder Accident:
  • Karine hit by a truck just after massive growth spike and conference appearance (25:58–26:19).
  • Major existential threat: CEO incapacitated; company structure risk due to founder-led sales dependence (27:00–27:56).
  • Recovery and resilience: Leaned on team, recognized need to scale and distribute responsibilities (27:56–28:48).

8. Never Pivoted: Commitment to the Mission

• Contrast with Typical YC Startups:
  • Stayed focused on initial hypothesis/problem for years, regardless of industry or technical evolutions (28:48–29:36).
• Sense of Duty: Built company around “the good guys” using advanced AI to stay ahead of adversaries (29:36–31:10).

Notable Quotes & Memorable Moments

• “We're building the systems that are often used by the bad guys, but we're building them for the good guys.”
  — Karine (01:34)

• “You’re their secret weapon. They don’t want anyone to know what their secret weapon is.”
  — Host (01:57)

• “Every person who signs up to do a GoFundMe fundraiser… their request is being validated by Variance’s software before it’s allowed to go live.”
  — Host (05:37)

• “At scale, some investigations can lead to finding things that are really scary… Once it's detected and investigated by Variance, it's usually going to be in the hands of law enforcement.”
  — Karine (16:09)

• “We’re five, but I think in terms of software output we’re probably closer to a 25 people team.”
  — Karine (19:05)

• “I always felt a really strong sense of duty to put our very specific and quite rare pair of skill sets to the good of the industry.”
  — Karine (29:36)

Timestamps for Key Segments

• 00:30 — Announcement: Coming out of stealth; $21M Series A
• 01:34 — Secrecy, sensitive data, “building for the good guys”
• 03:52 — GoFundMe fraud example
• 05:37 — Variance’s invisible but vital presence in user-facing products
• 06:20–07:12 — Marketplace identity and KYB verification
• 09:35–10:58 — Technical challenge: integrating unstructured data
• 11:18 — Scraping internal dashboards; agentic data integration
• 12:07–13:58 — AI agents as a leap over traditional fraud systems
• 14:22–15:29 — Detecting fraud rings, misinformation, and state actors
• 16:30–19:28 — Small team, high impact, AI coding agents
• 20:18–20:54 — Founders' experience; Apple’s fraud engineering
• 22:24–23:39 — Landing IAC as first (enterprise) customer
• 24:06–24:15 — Building pre- and post-ChatGPT
• 25:58–28:48 — Founder injury, recovery, and resilience
• 28:48–31:10 — Staying true to the founding mission

Conclusion

Variance represents a new breed of startup that quietly powers critical trust-and-safety decisions across the internet. The founders’ rare expertise and unwavering focus transformed a legacy challenge—automating complex risk and fraud investigation—into an enduring, high-impact business. Their work not only protects large companies but, at scale, is helping support safer communities. This episode is a rich behind-the-scenes look at the reality of building in the shadows and the resilience needed to succeed.