Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Banks can use AI to sift through enormous volumes of transactions quickly, but fraud prevention still needs accountable people to handle ambiguity and consequential decisions. That is the central argument Vittesh Sahni, Senior Director of AI at Coherent Solutions, makes in a May 25, 2026 interview with TechBullion. His recommendations are a strategy, not proof that a particular system achieves the performance figures he cites.
What Sahni means by human-augmented AI
Sahni’s proposed division of work is straightforward: AI processes high-volume signals and routine cases; people examine unclear cases and remain responsible for important decisions. As he puts it, “Machines handle the speed and the volume. People stay in charge of the judgment calls.”
As an Amazon Associate I earn from qualifying purchases.
That does not mean a person must manually inspect every alert. Sahni offers an illustrative split in which AI handles 80% of easier alerts and people address the 20% that are trickier. He presents this as an example, not a measured or universal staffing formula. The practical aim is to focus human attention where context, uncertainty, or the consequences of an error make judgment especially important.
Where rules, AI and hybrid systems fit
Sahni argues that fixed rules and conventional models can struggle to keep pace with changing fraud patterns. That is his framing in the interview, not a claim that every rules-based system is ineffective. Rules remain useful where a requirement is stable, explicit and easy to explain; AI can help identify less obvious or evolving patterns in large datasets.
#1 Best Overall
| Approach | Where it can fit | Key questions for a bank |
|---|---|---|
| Rules-based | Transparent, fixed requirements and conditions | Can the rule be kept current without creating excessive alerts? Can staff explain why it fired? |
| AI-driven | High-volume analysis and patterns that may be harder to specify in advance | Are data quality, validation, explanation, monitoring and escalation adequate for the intended use? |
| Hybrid | Rules for clear requirements alongside AI for more adaptive pattern detection | Which component triggered the action, who reviews uncertainty, and who owns the final decision? |
The choice is not only about detection speed. Banks also need to consider integration with existing data, decision latency, false-positive workload, how readily staff can explain a result, the reversibility of the proposed action, and who is accountable for it. The Bank of England and Financial Conduct Authority’s 2019 survey reported machine-learning use in areas including fraud detection and anti-money-laundering, with alerting and human-in-the-loop mechanisms among commonly reported safeguards. The survey received 106 responses, and the regulators cautioned that its findings were not statistically representative of the whole UK financial system; it is historical UK context, not a current global adoption estimate. Bank of England and FCA, “Machine learning in UK financial services”.
How to place human review around fraud interventions
Not every fraud signal warrants the same response. Sahni’s recommendation is to scale oversight with the potential harm and difficulty of reversing an action: “The bigger and harder-to-undo the decision, the more a human needs to be involved.” A bank can operationalize that risk gradient by distinguishing between a verification request, a temporary hold or review, and a consequential step such as freezing an account or ending a customer relationship.
- Use proportionate checks for lower-impact uncertainty. A request for additional verification can address a suspicious signal without immediately blocking a legitimate customer.
- Route unresolved or higher-risk cases for review. A temporary hold or investigation can give staff time to assess transaction context and available evidence.
- Require stronger oversight for consequential actions. Define who may authorize actions that are difficult to reverse, what information that person needs, and how the decision is recorded.
- Feed investigation outcomes back into controls. Review whether alerts were genuine, mistaken or inconclusive, and use those outcomes to improve rules, models and escalation procedures.
This is a recommended operating approach, not a universal legal rule. Requirements depend on jurisdiction and the decision being made. RBI Deputy Governor Shirish Chandra Murmu said in a speech delivered in Mumbai on September 11, 2026: “Responsibility rests with the regulated institution, and boards and senior management must understand the models they deploy, their limitations and the consequences of their use.” His remarks concern Indian financial institutions and support the principle of institutional accountability; they do not establish a universal ban on automated decisions. BIS, “Trusted innovation – shaping the future of finance”.
Reducing false positives without weakening protection
False alarms are not just an inconvenience for investigation teams: they can add friction for customers and consume attention that could be directed to credible threats. Sahni recommends using richer transaction context, incorporating investigation outcomes, and choosing a graduated response—such as verification—rather than defaulting to a hard block whenever a signal is uncertain.
Rank #3
In the interview, Sahni says some recent deployments saw up to 80% fewer false alarms. He also says AI-based AML tools found two to four times more suspicious activity while reducing overall alert volume by more than 60%. These are claims attributed to Sahni; the interview does not identify the deployments or underlying studies, so the figures should not be read as independently verified benchmarks or guaranteed results. A bank evaluating a system should ask what was measured, against which baseline, over what period, and how missed fraud and customer impact were assessed alongside alert counts. TechBullion interview with Vittesh Sahni, May 25, 2026.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Governance must match the system’s reach
Automating more of the workflow does not transfer responsibility away from the institution using the system. Governance needs to cover the data a model receives, the purpose for which it is used, validation before deployment, monitoring after launch, escalation paths, and the authority to intervene when results are unreliable. Staff also need enough explanation to investigate an alert and understand the basis for an action.
The Bank of England and FCA survey is one historical example of regulators documenting safeguards such as validation, governance and human involvement in machine-learning use. Separately, Murmu’s 2026 remarks emphasize that boards and senior management should understand deployed models and their limitations. Together, these sources support careful oversight; neither establishes that one governance design fits every institution or jurisdiction.
Recommended Free Tools
What the interview establishes—and what it does not
The interview is a useful account of Sahni’s recommendations, not an independent assessment of their effectiveness. Coherent Solutions, his employer, also published a related finance research page naming Sahni and Chief Strategy Officer Shawn Torkelson as contributing experts. That page outlines work involving agentic systems, biometrics, graph analytics, data strategy, governance and lifecycle management. Coherent Solutions, “Discover the Future of Finance: How AI is advancing fraud detection in financial services”.
The interview also refers to an unnamed Midwest regional bank and an onboarding benefit, but does not identify the bank or provide measurements that would support quantifying the result. It therefore cannot serve as a verifiable case study. More broadly, the interview’s performance figures lack named underlying deployments or studies in the account itself.
For a technology-focused reader, the useful takeaway is not that automation should replace judgment or that people must review every alert. It is that banks need to design the handoff: which signals can be handled routinely, which cases require human context, and which decisions demand a clearly accountable owner.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errors
