Everyday Reflections

AI is reshaping sectors by automating tasks, given its forte in spotting patterns and anomalies. 

Take the banking, financial services and insurance industry, for instance. AI is being integrated for repetitive tasks involving technical analysis of documents, fraud detection, and customer support via chatbots. 

💸As per reports, Bajaj Finance is betting on AI systems to handle 85% customer-service resolutions, loan application checks, and support tasks, and thus boost operating efficiency over 18 months.

💸Kotak Mahindra Bank has slashed SME loan processing time from 7.5 to 2.5 weeks (by 60%) using AI/ML digital underwriting tools and end-to-end transaction processing. 

This streamlining of work processes can free up staff to deal with decision-making and strategy. And there’s scope for further advancement as AI learns from past patterns. 

Yet, in terms of full-scale automation, the human element is crucial. 

At Bajaj Finserv, 10,000 loans are disbursed per month using AI, but humans oversee the entire process due to fears of hallucination, Sanjiv Bajaj, CMD of Bajaj Finserv, is quoted as saying in a PTI report.

And according to an EY report, the accuracy of AI predictions and potential for bias (due to training data) are concerns for the financial services industry, which require human oversight.

Meanwhile, in the tech sector, companies like Amazon and Google have been gung-ho about LLMs’ abilities to generate code. 

🖥️Microsoft CEO Satya Nadella said that around 30% of the company’s code is now written by AI. 

🖥️Meta CEO Mark Zuckerberg predicted that for one of its projects, half the development is going to be done by AI.

Here the issue lies in the quality. An NPR report suggests that software engineers on the ground are struggling to fix messy AI-generated “workslop” and end up reviewing each line of code. That’s bound to hamper productivity. 

A study by nonprofit research institute METR shows that experienced software engineers using LLMs took 19% longer to finish tasks compared to peers who didn’t — contrary to the engineers’ own expectations.

Automation’s strong track record in fields like aviation has created the belief that more automation automatically means fewer errors. But a review of 74 studies (Goddard et al., 2012) shows the opposite risk: when systems are highly reliable—but not perfect—people tend to over-trust them, leading to a 12% rise in errors from accepting wrong AI suggestions and slower recognition of unusual issues. In short, reliability can make humans complacent.

Rather than aiming for full automation, organisations should use hybrid systems that return control to humans when AI confidence drops, and invest in automation literacy so employees know when to trust and verify AI outputs. The aim is balanced, not removed, human judgment.

The idea of end-to-end automation that keeps humans out of the mix is still a while away. But in the meanwhile, work will increasingly get redesigned to factor in this new reality of human + machine collaboration. 

#ArtificialIntelligence #Software #Engineer #BFSI #Automation #AI