Soon, your first job in finance may be arguing with a machine; B.Com isn’t preparing you for that

By K. Nagavani Kaggallu

Somewhere in a bank office in 2029, a 24-year-old credit officer, two years out of a B.Com from a college in a Tier-2 Indian city, is doing a job her seniors would barely recognise. By the time a loan application reaches her desk, the software has already done what used to be the work. It has pulled the bank statements of the applicant, read the GST filings, scored the risk, checked the result against the bank policy manual and drafted the rejection letter. Her job starts where the job of the machine ends. It is not to make the decision as most of us think. It is to decide whether the machine should be allowed to make decisions.

Nobody in her three-year degree taught her how to do that. And this is the gap Indian commerce education has not noticed yet, though it will decide the careers of everyone graduating between now and 2030.

Also read IIT Placements 2026: Many in BTech Computer Science aim for trading firms, not Google; internships show why

Finance Education: Broken sequence

For decades, financial education ran on the commandment to know your numbers. Read the balance sheet, understand the cash flow, work out the ratios, and then decide. The whole sequence assumed a human did the deciding. That assumption is now negotiable. Everyone knows AI calculates faster than you. The shift that matters is that banks and NBFCs have begun experimenting with handing over pieces of the decision itself.

RBI Governor Sanjay Malhotra recently compared the potential of AI and its effect on lending to what UPI did to payments. If you are a commerce student under 25, that comparison should land hard, because you have probably never stood in a bank queue to transfer money. UPI bypassed the old process of standing in queue. Malhotra is suggesting AI could do the same to credit. Means it will change who, or what, approves a loan or an overdraft. What does it mean to commerce and financial education of today?

Researchers who analysed 1.5 million real ChatGPT and Gemini conversations, from users in the United States and India, found that people already lean on AI heavily for financial information and judgement. But most of them do not allow it to act. They ask, ‘What should I do with this money?’ But they do not ask ‘Do it for me’, unless they use an agentic AI.

The shift runs from information to recommendation, then to delegation and execution, and perhaps eventually to AI agents handing work to other AI agents. Each step transfers authority. Accuracy is relatively easy to measure; the tough part is to acknowledge whether the human understands what authority has been handed over and what remains theirs to decide. That leads us to a different form of delegation which is missing in finance education today.

Also read ‘Hacker’s bent of mind’: IIT Kanpur, Madras B.Cybersecurity course to train ‘cyber warriors’, says in-charge

Delegation literacy with financial, digital literacy

What would be called delegation literacy can be placed beside financial literacy and digital literacy without reducing to either.

Take a concrete version. A student learns to assess credit risk using two AI models. Both are roughly equally accurate. A conventional course asks which predicts better and it ends there. The more useful course asks: which one would you allow to make the decision? Now she has to examine what data each model saw and what it missed, how each one fails, whether its errors cluster on particular kinds of borrowers, what the regulator requires, and what a wrong call costs the person on the other end of the file. Two models, same accuracy, and possibly opposite answers about authority.

The lesson is harder than merely saying ‘AI is unreliable’ or ‘AI is superior’. Different decisions deserve different amounts of machine authority, and somebody has to be trained to set that dial.

Humans will also make mistakes. Research on robo-advisers has begun documenting automation bias. It happens when investors accept an algorithmic recommendation without checking it. Most AI debates worry about machine bias against people.

The bigger problem is human deference toward machines. A serious course would engineer both into its cases, mixing models that are confidently wrong, models that are statistically right but ethically indefensible, and models whose output the student tends to accept. Disagreeing with a machine on evidence is a skill. So is knowing when not to. We are yet to incorporate this in our finance and commerce courses.

The AI framework of the Reserve Bank of India (RBI), released last year has more than two dozen recommendations on governance, accountability, audit and capacity rather than on model performance. Regulators are preparing for a world where the question, “Who allowed this system to decide?”, gets asked in earnest. Universities are still teaching as if the human being always decides.

If banks remain accountable for AI-driven decisions, future commerce graduates will need to understand more than finance. They will need to trace where an AI decision came from, who authorised it, when to intervene, and who remains responsible when it fails. These are becoming employable skills. Courses in finance will need to teach AI governance, model oversight and human intervention alongside conventional accounting and risk analysis.

Auditing may increasingly shift from checking transactions to reconstructing how an AI system reached a decision. That creates a new career space in AI audit and compliance, requiring commerce graduates to understand data trails, model decisions, system logs and governance alongside accounting. So, how will the finance classroom look?

Also read BTech Courses: AI, computer science fuel enrolment boom to 5-year high, but may soon kill jobs, say experts

The new finance classroom

Instead of one more lecture titled ‘AI in Banking’, it may look more like a lab. A machine approves some loans and rejects others. Students get the data trail, the output of the model and the applicant's circumstances, and their task is to rule on the machine's answer: accept it, challenge it, demand more evidence, or stop the pipeline entirely. Marks would not go to whoever matches the machine. The marks are for those who can defend why a given decision should stay with the machine, return to a human, or be shared between them.

How can we evaluate them? Hand students a polished, plausible AI recommendation on a financial case and ask for the strongest reason it might be wrong. Then change the conditions mid-semester without warning, say the vendor swaps the underlying model overnight, and see who notices. That is a different exam, and a harder one. It is also much closer to the actual job.

None of this requires commerce students to become programmers. The more capable the machines get, the more the market pays for people who understand their limits.

Regulators use the phrase ‘human in the loop’. It sounds reassuring but may simply mean that a human clicking approve on whatever the screen suggests is in the loop and asleep at the same time.

Our credit officer, at her desk in 2029, reads the rejection from the machine one more time. The applicant is a first-generation shop owner with a thin credit history, exactly the kind of file where a model sees risk and a human sometimes sees a customer. She can let the letter go out, or she can pull the case back. Whichever she chooses, everything depends on whether anyone, in any classroom, ever taught her how to choose. That is where commerce education has to move to.

Dr. K. Nagavani Kaggallu is associate professor, Post Graduate Department of Commerce, Seshadripuram College, Dr. Manmohan Singh Bengaluru City University. She is also founder and chairman of Alpha Research Centre, Bengaluru.