Deepfakes that fool facial recognition. Synthetic identities built from stolen data. Fraud rings that adapt faster than banks can update their defenses. Financial institutions keep pouring money into AI, yet losses keep climbing. The problem isn’t technology—it’s approach.
Mikhail Matveev spent 15 years in data and analytics before cracking a problem most banks consider unsolvable. As Chief Data Officer at B9, a Miami-based financial app serving 1.6 million customers, he wanted to crack down on fraud without alienating legitimate users. The conventional wisdom said you had to choose: either tight security that frustrates people, or smooth onboarding that lets criminals slip through.
His team rejected that trade-off. After implementing AI-powered liveness verification, they cut fraud and reduced chargeback losses by 35% over a three-month period—without spiking false positives or adding friction for honest customers.
Matveev built his expertise across multiple markets—Russia, APAC, and now the U.S.—working from analyst to the C-suite. Analytics Insight magazine named him one of “The 10 Most Impressive Chief Data and Analytics Officers” in 2024. He has judged fintech competitions, including the best data strategy contest at VIO 2019, and mentors women in tech through the Women Go Tech Acceleration Program.
“Data without business context is just noise,” Matveev says. “Working in different markets taught me where the real problems hide—and where AI can actually move the needle.”
You’ve built an AI system that detects fake identities. What’s the core idea behind it?
We didn’t start with artificial intelligence. We started with customer complaints. People kept calling to say “that wasn’t me”—transactions they never made, accounts they never opened. So we dug into the data.
The root cause was identity theft. Criminals using stolen documents to open accounts, or hijacking existing ones to drain them. We built our defenses in layers—multiple checkpoints rather than a single gate.
The key innovation was combining biometric verification with behavioral analysis. It’s not enough to confirm someone’s identity at sign-up. What happens when a fraudster steals a phone from an existing customer? That’s where continuous monitoring kicks in. We watch for anomalies throughout the entire customer journey, not just at the front door.
The fraud landscape is shifting under our feet—deepfakes, face-swapping, AI-generated identities. How do you stay ahead?
It’s a constant chase. A few years ago, fraudsters used printed photos. Now they use face-swap technology sophisticated enough to fool basic checks.
We treat it as a moving target. Every time we catch a new technique, we feed it back into the system. When bad actors find a workaround, we update our rules within days, not months. We’re not chasing some perfect model that will never need updating. We’re building something that learns faster than the bad guys do.
What results have you seen?
After rolling out AI-powered liveness verification, we cut chargeback losses by 35% in three months. The overall risk rate dropped by 20%, which improved portfolio quality and reduced our financial exposure. And we achieved this without making life harder for legitimate customers. That balance is what most companies get wrong.
You could stop 100% of fraud tomorrow by rejecting every application. But then you’d have no customers. The real skill is stopping criminals while letting legitimate users through without hassle.
How does your data strategy support fraud prevention beyond the initial verification?
Fraud detection doesn’t stop at onboarding. We analyze behavioral patterns continuously—transaction frequency, spending categories, timing anomalies. When someone connects their external bank account, we can see their transaction history. But raw data looks like gibberish—random strings of letters and numbers.
We use one system to decode the messy transaction strings into plain language, while a separate numerical engine hunts for stability and spending patterns. Together they build a risk profile that evolves with the customer. Suspicious activity triggers review in real time, not after the damage is done.
Before B9, you built data infrastructure across Asia. What did you learn there?
At Cashwagon in Singapore I led a data team covering four key areas: reporting, data warehousing, business integrations, and analytics. Some numbers from that period still stand out.
We implemented a data warehouse that cut retrieval times by 40%. Moving to Power BI with real-time dashboards improved reporting accuracy by 40%. A collections model that personalized outreach based on transaction history saved more than half a million dollars a year.
We also reduced data quality issues by 35%. In financial services, bad data leads to bad decisions—and bad decisions cost real money.
What do business leaders typically get wrong about data teams?
They see us as a cost center. “Why do I need data people when marketing already tracks downloads and risk tracks defaults?”
Here’s the problem: everyone optimizes their own little piece. Marketing celebrates installs. Risk celebrates rejections. But who’s asking whether the customers marketing brings in are the same ones risk keeps turning away? Nobody connects those dots unless you have people whose job is to see the full picture.
I’ve worked across every part of the financial business, from operations to executive strategy. That’s how you spot the blind spots. Most companies have them because their data lives in silos that never talk to each other.
What advice would you give a company just starting to build data capabilities?
Don’t try to build the perfect system first. Find one team with a concrete problem, solve it, measure the impact, use that win to build credibility. Trying to construct enterprise-wide infrastructure before proving value is a recipe for frustration.
And never let your numbers be wrong. The moment an executive catches an error in your report, you’ve lost their trust. Maybe for months. Your job is to be the source of truth—if people can’t rely on your data, you have nothing.
Where do you see fraud prevention heading in the next few years?
The arms race will accelerate. Generative AI makes it easier to create convincing fakes, but it also gives us better tools to detect them. The winners will be companies that build adaptive systems—ones that learn from every attack and update automatically.
Analytics will evolve too, though more slowly. Current tools can produce reports but lack business context. They might flag unusual activity without understanding legitimate reasons behind it. Unlocking that capability will require cleaner data, clearer definitions, better documentation.
Companies building those foundations now will have a huge head start in three to five years. Everyone else will be scrambling.
What does success look like in your role?
Speed. When the CFO asks why chargebacks spiked last week, the answer should arrive in hours, not days. When a product manager proposes a new verification step, historical data should predict its impact before anyone writes a line of code. When fraud patterns shift, defenses should adapt before losses pile up.
Infrastructure and dashboards are just the plumbing of the operation. At the end of the day, our only real job is to shrink the distance between a raw question and a business-ready insight.
