Smart calls, human credit: the guarded rise of AI in India's microfinance sector

SummaryMicrofinance companies are scaling up AI to boost productivity and strengthen fraud checks, while keeping core lending decisions largely manual amid data quality challenges and evolving privacy rules.India's microfinance institutions are expanding the use of artificial intelligence across customer calls, cybersecurity and internal operations as lenders look to improve efficiency, reduce costs and strengthen risk management, according to executives.The adoption marks a shift from experimental deployments to operational use cases, although firms remain cautious about applying AI to lending decisions, where much of the assessment still relies on field verification and human judgment.While lenders are increasingly buying AI capabilities, some are avoiding software-as-a-service (SaaS) models for customer-facing applications, preferring to host AI models within their own infrastructure to retain control over sensitive borrower data."We are not taking SaaS-based AI solutions. The agents are hosted on our own infrastructure. We are using Azure, but the hosting is on our infrastructure itself, so no data is flowing outside," said Avinash Yadav, chief information officer at Spandana Sphoorty Financial Ltd. The listed microlender has a market capitalisation of ₹2,117.37 crore. On Thursday, the company's shares closed up 0.8% at ₹265.55 on BSE.Belstar Microfinance—an arm of Muthoot Finance Ltd with assets under management of ₹8,222 crore—follows a similar approach. Dhanasekaran S., chief technology officer at Belstar Microfinance, said the company keeps its AI running within systems it controls rather than sending customer data to external AI providers, giving it greater control over sensitive information.Another microlender, Fusion Finance, meanwhile, stores customer information on Amazon Web Services (AWS), with additional safeguards around sensitive information. Fusion has assets of ₹7,407 crore.Key TakeawaysMicrofinance lenders deploy AI for calls, fraud detection, and cybersecurity operations. Companies prefer hosting AI on their own servers, avoiding third-party SaaS data risks. AI now handles most collection calls, improving loan officer efficiency by 15-20%. Credit assessment stays manual; household income data remains too unstructured for AI. Industry body sees early AI underwriting pilots, but stresses human judgment is essential."All customer data is on our AWS cloud in a secure environment. Aadhaar information is encrypted. We have also developed an in-house AI model to automatically mask historical Aadhaar records so auditors only see masked data," said Susheel Kumar Menon, chief information officer at Fusion Finance.Automated collection callsOne of the biggest applications of AI has emerged in customer calling, particularly collections, where lenders previously relied on large teams of callers and field staff.Fusion Finance explained that AI now handles nearly all collection calls, replacing manual calling with multilingual AI systems.At Spandana Sphoorty, AI acts as the first point of contact before field officers visit borrowers."Instead of sending field staff directly, AI can first call customers, check whether they have funds, ask when they will be available, and even send a payment link if they are ready to pay. Only after trying all these channels does the physical visit happen," said Yadav.He estimates that if even 30-40% of overdue accounts can be resolved remotely, field staff spend less time travelling and more time serving customers.The gains are already visible. "Earlier, if my loan officer was handling 50 customers, now that loan officer is able to handle 60 customers. There is a straightaway 15-20% improvement in efficiency," he said.Lenders are also using AI to detect fraud before and after loans are disbursed. At Spandana Sphoorty, AI-powered welcome calls verify loan details directly with borrowers."The AI confirms the loan amount, tenure, interest rate and whether any additional amount was asked