Financial institutions that are deploying gen AI are seeing efficiencies, but quantifying ROI is difficult. 

Returns from traditional machine learning are more quantifiable than returns from generative AI, Valley Bank Chief Data and Analytics Officer Sanjay Sidhwani told FinAi News.  

“Traditional AI is a tool that gives binary yes or no decisions, which are easily quantifiable,” Sidhwani said. “With generative AI, that link is kind of missing as of now because outcomes are much more qualitative.”

Until an institution pauses hiring, speeds up a process or downsizes for some reason, measuring returns from gen AI is very difficult, Sidhwani said.  

“A lot of the times, to deploy gen AI we might have to change the entire processes compared to traditional AI, which is usually an add-on to the process,” he said. 

For example, Valley Bank has seen its anti-money laundering process become 60% more efficient through the use of AI, Sidhwani said. Returns are “quantifiable because it’s fraud costs that we didn’t incur or fewer investigations we had to do.” 

But when it comes to gen AI-driven chatbots, outcomes are more qualitative for customers, he said. 

The difference 

Traditional AI projects often target repetitive tasks with clear historical data, making before-and-after comparisons straightforward, Joan McGowan, head of U.S. financial services consulting at data and AI company SAS, told FinAi News. 

Fraud detection, credit risk scoring and forecasting models have decades of history and a clear link to reduced losses, lower capital requirements and faster processing of everyday work, she said. 

“Generative AI introduces new unstructured data types (such as text) and complex outputs, which do not fit traditional ROI models,” McGowan said. “Gen AI systems are transformative but harder to quantify, relying on qualitative gains like faster analysis, smoother customer engagement and enhanced productivity.” 

Most of the gains that the $62 billion Valley Bank and its peers are clocking are still from traditional machine learning, Sidhwani said.  

Common industry benchmark 

Gen AI is still fairly new, and the financial services industry is going to take time to properly quantify gains from its deployment, Nik Kale, principal engineer, CX engineering, cloud security and AI platforms at Cisco, told FinAi News. 

“The industry will eventually converge on benchmarks for gen AI ROI, but we’re not there yet,” Kale said. “We still lack a shared language for measuring things like ‘decision acceleration,’ ‘context synthesis’ or ‘cognitive savings.’” 

Today, enterprises often start with proxy metrics such as interaction volume, task completion time and human-in-the-loop reduction, but these are stepping stones, he said.  

“True gen AI ROI will require a common taxonomy around agent behavior, orchestration reliability and impact on end-to-end business processes,” he said. 

Traditional ML delivers ROI through measurable model performance; gen AI delivers ROI through organizational leverage, Kale said. 

“Because gen AI touches workflows, not just predictions, its value is both harder to calculate and potentially much larger than anything we’ve seen from classical ML,” he added. 

Register here for early-bird pricing for the inaugural FinAi Banking Summit 2026, taking place March 2-3 in Denver. View the full event agenda here.