Agentic AI Breaks The Bottleneck Between Data And Revenue-Generating Decisions

Companies have always been great at tracking numbers in spreadsheets, but they don't necessarily have a great track record in other areas. The other roughly 75% of their data—contracts, invoices, and customer chats—has historically sat untouched: a garbled mess that needs to be translated into insights. Early natural language tools could give you a customer's sentiment score, but they just fed those back to the marketing team in a slow-moving loop. Today, many organizations are moving beyond that model. Using generative and agentic systems, they can finally turn that data directly into knowledge, actionable decisions, and, thus, revenue without waiting to meet changing expectations around speed and personalization.

Enter Bijay Kumar. As the Director of AI with The LCF Group and former SVP of AI Products Programs at Mastercard, Kumar managed the AI infrastructure behind the company's fraud solutions. For Kumar, the leap to agentic AI solves the ultimate "but then what?" problem of early machine learning. As systems learn to process intricate data more autonomously, many leaders see the technology as a direct contributor to growth rather than just an efficiency play.

• Show me the money: As systems learn to process intricate data more autonomously, many leaders (including Kumar) see the technology as a direct contributor to growth rather than just an efficiency play. "What used to take 45 minutes on a complicated case can now be done in five minutes, with the insights already surfaced for the underwriter. Once we gravitate towards this ecosystem, there will be a quick shift in business ROI.”

Nowhere is that transition more obvious than in lending. Historically, processing a loan application frequently relied on legacy OCR pipelines that required rigid, custom coding. Developers typically had to manually map document zones and program the system to recognize specific headers, transaction details, and footers. In the U.S. alone, lenders juggle over 100 different bank statement formats. Legacy systems simply couldn't keep up, effectively capping automation at the top tier of banks.

• Ditching the heavy lifting: Many organizations are turning to newer orchestration workflow frameworks designed to bypass that constraint, leveraging intelligent pipelines that can navigate highly varied banking infrastructure with far less custom coding. Kumar says, "This can be done within a couple of minutes without requiring heavy software engineering. Agentic tech enables a metadata and schema extractor, composable tools integration, and workflow frameworks, making data assets and pipelines natively accessible to agentic systems."

When a system can only read half of your incoming documents, organizations leave massive amounts of money on the table. Kumar says that when automation lifts processing volume to around 80%, and standard approval rates sit at 30% to 40%, institutions can convert a portion of that previously untapped demand into top-line revenue growth. Since advances in AI can solve document formats, achieving that lift often means rethinking where humans sit in the workflow. When a routine $500,000 loan request requires a decision in under 30 minutes, underwriters frequently feel compelled to manually double-check results for unfamiliar document templates, turning what should have been a streamlined process into a 45-minute exercise.

• Trust issues: This challenge, for Kumar, speaks to how much (or little) organizations trust AI vs. how much they trust human employees. "When dealing with new document templates, even if there is good work on the software engineering stack, confidence is low because you cannot trust your software system. The processing and underwriting team used to double-check and validate the data, resulting in heavy manual processing to reconfirm whatever went through the pipeline."

• Five minutes flat: The solution, according to Kumar, is modern agentic pipelines that act as autonomous decision-support systems, empowering agents to focus on edge cases rather than routine parsing. "Irrespective of what kind of document it is, the processing completes in four to five minutes without any human in the loop. Yes, there needs to be people for oversight, but not at the scale we have right now."

Handing financial data over to autonomous agents naturally sets off privacy alarm bells. Kumar says that certain probabilistic models can be architected to maintain anonymity in regulated spaces by focusing on patterns in documents and interactions rather than on individual identities. That modeling approach, he noted, also helps address a long-standing workflow problem. Legacy systems forced human underwriters to act as high-speed data-entry clerks. The combination of manual review, time pressure, and varied formats introduced fatigue and unconscious bias into decisions, sometimes resulting in qualified customers being declined.

Instead, by handing initial data extraction over to autonomous agents, lenders remove that friction entirely. Humans can step in primarily for policy decisions and true exceptions.

• To err is human: Kumar says he has seen institutions use this pattern to capture up to 90–95% of actionable insights with greater consistency, delivering faster, more predictable lending decisions. "We have seen many cases where there was a good net income-to-debt ratio or net cash flow, but the loan was not getting approved. One contributing factor was the manual processing involved. Underwriters could easily overlook some of the insights. Quality was an issue because of the human in the loop."

For early adopters in the financial sector, the implications of these back-office upgrades are spilling over into front-office customer experiences. The same agentic models that parse unstructured lending documents are beginning to deliver hyper-personalized recommendations, helping customers navigate dense product catalogs with tailored suggestions. Kumar goes further: he expects agentic transactions to emerge as a third major payment rail alongside card and online, with autonomous purchasing agents operating directly within the customer journey. This gives him ownership of a specific, bold forecast rather than presenting it as a passive industry observation. It is a direction some industry watchers expect could influence the future of e-commerce.

Kumar goes further: he expects agentic transactions to emerge as a third major payment rail alongside card and online, with autonomous purchasing agents operating directly within the customer journey." This gives him ownership of a specific, bold forecast rather than presenting it as a passive industry observation.

For Kumar, these shifts answer the "but then what?" question once more—agentic payments, AI-guided product discovery, and the move away from traditional search toward tailored recommendations are all converging into a single, more intuitive customer journey. The companies that act on that convergence now, he says, won't just streamline operations; they'll fundamentally reshape how customers find, evaluate, and buy. "With agentic technology, we're moving from processing data to predicting and guiding better customer outcomes in real time."