As financial fraud becomes faster, more sophisticated and increasingly AI-driven, banks are being forced to rethink the foundations of fraud prevention. In this TechBullion exclusive interview with Vittesh Sahni, we explore how artificial intelligence is transforming the fight against financial crime, and why the future of fraud defence will depend not on automation alone, but on the strategic partnership between AI and human expertise.
From deepfake-enabled scams and real-time payment fraud to the rise of agentic AI systems capable of conducting investigations autonomously, Sahni explains how financial institutions can modernise legacy fraud controls while maintaining accountability, regulatory compliance and customer trust. He also discusses the growing “AI versus AI” arms race between banks and cybercriminals, the importance of reducing false positives, and how human-augmented AI is helping institutions deliver faster, safer and more seamless digital banking experiences.
As banks compete not only on security but also on customer confidence, this conversation offers valuable insight into the technologies, governance models and strategic decisions shaping the next generation of intelligent fraud prevention.
Q1. Please tell us your name, a little more about yourself, and what problems you are solving at Coherent Solutions.
I’m Vittesh Sahni, Sr. Director of AI at Coherent Solutions, a global digital solutions engineering firm based in Minneapolis. My job is to help large enterprises figure out how to use AI to create digital value.
In banking, the problem I keep seeing is the same one in different forms. Fraud is getting faster and smarter, while the systems most banks use to stop it have not kept up. Fraud teams are buried in alerts, real customers get blocked by mistake, and the whole setup was not built for criminals who now move at the speed of software.
That is where my team comes in. We help banks build AI systems that work alongside human experts, not in place of them. Machines handle the speed and the volume. People stay in charge of the judgment calls. I recently authored Coherent Solutions’ whitepaper, Future of Finance: How AI is Advancing Fraud Detection in Banking and Financial Services, which lays out the full strategy and roadmap. Done right, fraud prevention stops being a cost the bank apologizes for and starts becoming a reason customers trust them.
Q2. Tell us more about static fraud controls, including static AI models, and why they are no longer fit for purpose in banking.
A “static” fraud system is one that is set up once and rarely changes. That used to be fine, because fraud patterns moved slowly. Today fraud patterns move much faster.
Three things have changed. Criminals now use cheap AI tools to fake IDs, clone voices, and create deepfake videos. Money moves faster than ever, so the window to catch fraud has shrunk to seconds. And the criminals themselves are using AI to attack at industrial scale.
Against all that, a fixed rule like “flag anything over $5,000 from this country” is identified and bypassed within hours. A model that was trained six months ago is already out of date. The warning signs of a system in trouble are too many false alarms that frustrate real customers, no way to explain why decisions were made, and no system for learning from yesterday’s mistakes.
Banks need systems that learn continuously, with human experts watching over the high-stakes calls.
Q3. What does “AI versus AI” mean in the context of modern financial fraud?
It means that both sides, the bank and the criminal, are now using the same kinds of AI tools. Whoever learns faster wins.
On the criminal side, AI is being used to mass-produce fake identities, fake documents, and convincing deepfake videos. It can plan and run attacks automatically, picking the moment fraud teams are least staffed.
On the bank’s side, AI can score a transaction in milliseconds, summarize a complex case in seconds, and handle straightforward investigations from start to finish, only flagging the genuinely tricky ones for a human.
The strategic question for banks is no longer “do we have AI?” It is “how fast does our AI learn, and how quickly can our team act on what it learns?” That is the real contest.
Q4. How could agentic AI change the way fraud teams investigate and respond to suspicious activity?
“Agentic AI” just means AI that can take a task and work through it on its own, the way an analyst would, instead of waiting for a person to tell it the next step.
Today, when a fraud alert pops up, an analyst opens five or six different systems and pieces together the story by hand. That takes hours per case. With tens of thousands of alerts a week, the math simply does not work. Most banks are not losing the fraud fight because they cannot detect things. They are losing because the queue of cases to investigate keeps growing.
Agentic AI changes that. It pulls evidence from across the bank’s systems, writes up the case, ranks the most urgent ones, and shows its work so a human can check it. Some recent deployments are reporting up to 80% fewer false alarms, with investigations that used to take weeks now finishing in minutes.
The point is not to remove people. It is to free experienced investigators from the busywork so they can focus on the cases that actually need their judgment.
Q5. How can banks use human-augmented AI to detect fraud faster without losing accountability?
“Human-augmented AI” means the AI does the heavy lifting and the human stays in charge. We sometimes call this “human-in-the-loop.”
The logic is simple. AI is better at things humans struggle with, like looking at millions of transactions at once, spotting tiny patterns, and never getting tired. Humans are better at things AI struggles with, like making judgment calls on unusual cases, taking responsibility for the outcome, and talking to a real customer with empathy.
In a well-designed system, AI handles the easy 80% by clearing low-risk alerts and drafting case summaries. People handle the trickier 20%, where their experience actually matters. The results are dramatic. For example, AI-based anti-money-laundering tools can find two to four times more suspicious activity while cutting overall alert volume by more than 60%. Coherent’s own work with a Midwest regional bank produced similar results, helping the bank stay compliant during customer onboarding without making the experience slower or more frustrating.
Accountability is made cleaner in this model. The bank still owns every decision, and exactly who did what, the AI or the human, is documented and reviewable.
Q6. Where should human oversight sit when AI recommends blocking payments or freezing accounts?
One simple rule: The bigger and harder-to-undo the decision, the more a human needs to be involved.
For small, easy-to-reverse actions like asking a customer to confirm with a one-time code, AI can act on its own and humans can spot-check the results. For things that affect a customer but are still reversible, like holding a single payment or flagging an account for review, AI should recommend and a human should confirm, but in seconds, not hours.
For the big stuff, freezing an account, filing a fraud report to regulators, or cutting off a customer relationship, a qualified human always makes the final call. The AI gathers the evidence and drafts the case. The person decides. Regulators in the US, the EU, and elsewhere are increasingly clear that purely automated rejections without human review are not acceptable.
The amount of autonomy you give the AI should grow as it earns trust, and it should also be possible to pull it back when fraud patterns shift.
Q7. What should banks consider when choosing between rules-based, AI-driven and hybrid fraud systems?
In 2026, almost no serious bank should be picking just one approach to AI. The right answer for most is a hybrid, meaning rules and AI working together.
Rules are simple and easy to explain, but they are rigid and criminals navigate them quickly. AI is adaptive and can spot patterns rules cannot see, but it needs good data and careful explanation. A hybrid approach uses each where it is strongest. AI catches the new and the unusual while rules act as a clear safety net for things like sanctions screening and regulatory thresholds, where transparency is non-negotiable.
The real decision for a bank isnot “rules or AI?” It is answer how much does the institution need to be able to explain to regulators? How fast does this decision need to be made? How clean is our data? Where is our team’s biggest pain today? Start where the answers are clearest, and grow from there.
Q8. How can financial institutions reduce false positives while maintaining strong fraud protection?
A “false positive” is when the bank’s system blocks something that turned out to be perfectly legitimate. It is the most underrated cost in fraud, because every false block can damage the customer experience and wastes staff time.
Three things move the number.
First, context. An old rule sees only the transaction in front of it. A modern system also sees the device, the customer’s recent behavior, the network of connections behind the account, and the merchant’s risk profile. With that fuller picture, AI tools are catching two to four times more real fraud while cutting alerts by more than 60%.
Second, learning from mistakes. Every false alarm is a lesson, but only if the system is set up to capture it. Many banks lose this signal because the analyst’s notes never make it back to the model. Fixing that loop is often worth more than buying a new tool.
Third, lighter-touch responses. Not every suspicious sign needs to be a hard block. A quick in-app confirmation or a one-time code often does the job with a fraction of the customer pain. The right scorecard for a bank is not just “how much fraud did we catch?” It is the balance of fraud caught, real customers helped, and staff time saved.
Q9. How can AI-led fraud prevention improve customer trust and digital banking experiences?
In 2026, fraud prevention is a customer experience function, not a back-office one. Every wrongful block is a trust event, and every quick, painless save is a moment of confidence.
AI changes the experience in three ways customers feel without ever needing to understand the technology.
First, invisible protection. Hundreds of checks happen quietly in the background, looking at the device, the typing rhythm, the connections behind an account, all without bothering the customer.
Second, the right-sized response. When something does look off, the bank can pick the lightest possible check, like a quick app confirmation, instead of declining the payment outright. The customer feels protected, not interrogated.
Third, faster recovery. When fraud does happen, AI helps close the case in hours rather than weeks. The money comes back faster, the communication is clearer, and the customer walks away thinking “my bank had my back.”
Done right, fraud prevention stops being something a bank apologizes for and becomes something a bank competes on. The institutions that win the next decade will not be the ones with the most automation. They will be the ones with the smartest partnership between human judgment and AI, and the customer experience to prove it.
Vittesh Sahni · Sr. Director of AI, Coherent Solutions · coherentsolutions.com
