{"id":1235,"date":"2026-02-10T08:42:00","date_gmt":"2026-02-10T08:42:00","guid":{"rendered":"https:\/\/www.spokeo.com\/pathfinder\/?p=1235"},"modified":"2026-01-19T20:43:11","modified_gmt":"2026-01-19T20:43:11","slug":"uses-of-ai-in-debt-collections","status":"publish","type":"post","link":"https:\/\/www.spokeo.com\/pathfinder\/uses-of-ai-in-debt-collections\/","title":{"rendered":"AI in Debt Collections: New Study Gives Insight Into What Works, What Doesn\u2019t"},"content":{"rendered":"\n

The past few years have seen a steady rise in the business use of artificial intelligence (AI), a trend that came to mainstream society with the public release of ChatGPT late in 2022<\/a>. Since then, the allure of \u201cgenerative\u201d AI tools, such as ChatGPT and Claude, has only grown, and it\u2019s easy to understand why: they\u2019re interactive, general-purpose, and capable of doing many of the things that humans can do.\u00a0<\/p>\n\n\n\n

Theoretically, then, automated debt collection driven by AI (or human-led collections enhanced by AI) could be a potent, or even revolutionary, next step for an industry that relies on a lot of tedious, manual work. So, let\u2019s take a look at AI in debt collections and how it fares in specific use cases. <\/p>\n\n\n\n

AI in Debt Collections<\/h2>\n\n\n\n

The term \u201cAI\u201d was coined in the 1950s<\/a> and has been used to describe several different technologies over the course of the following decades. For discussion purposes, this post will focus on the two specific types of AI you\u2019re most likely to encounter in collections.\u00a0<\/p>\n\n\n\n

One is called machine learning (ML)<\/a>, and it\u2019s built into a number of tools you already use. It\u2019s trained to recognize patterns in data, to draw inferences from those patterns, and then to act on what it infers. Many of the services used within the collections industry use this type of AI for analytics, including Spokeo for Business<\/a>, which uses a proprietary machine learning algorithm to help identify the best contact information for debtors.\u00a0<\/p>\n\n\n\n

The other is generative AI tools, including ChatGPT, Claude, and xAI\u2019s Grok, as well as Microsoft\u2019s Copilot suite, which integrates these models into productivity tools. These are what are called large language models (LLMs), a specialized form of machine learning trained on vast quantities of text-based data, without being limited to a specific task or intended purpose. They are, so to speak, a technological \u201cSwiss Army knife\u201d that can be turned to a wide range of purposes. These models can be used for tools like improved chatbots that can interact with debtors very much as human agents would, with spoken or written responses. They can even be used for fully automated debt collection. <\/p>\n\n\n\n

The Promise of Automated Debt Collection<\/h2>\n\n\n\n

AI\u2019s potential use-cases within collections hold significant promise, from improving analytics to enhancing the productivity of human agents. Interest in the technology is on the rise: in TransUnion\u2019s 2024 Debt Collection Industry<\/a> report, only 27% of respondents said that their companies were not considering AI implementation in some form. That\u2019s down from 40% in the 2023 report, a very significant one-year change of focus (we named the increasing use and adoption of AI as one of the top five industry trends of 2025<\/a>).\u00a0<\/p>\n\n\n\n

The boldest potential use case is direct AI debt collection. A functioning, fully automated debt collection model would neatly address a number of the debt industry\u2019s staffing issues, from recruitment to retention and burnout<\/a>. For agencies struggling to manage staffing costs and tight revenue margins, the appeal of such a model is clear.\u00a0<\/p>\n\n\n\n

A limited number of enterprises, in the US and globally, are currently experimenting with fully automated options. One such company, in China, has been using automatic AI callers (as well as humans) since 2018 to reach out to delinquent accounts. Its experience was the subject of a recent research paper, which analyzed the differences in results between human and AI callers. In theory, low-cost AI agents should have proven to be a cost-effective alternative to humans. In practice, the outcomes were more nuanced.\u00a0<\/p>\n\n\n\n

\"AI<\/figure>\n\n\n\n

Human-led vs AI-led Collections<\/h3>\n\n\n\n

The company in question is a Chinese lender that operates a retail storefront as a means to drive loans (the opposite of many scenarios in the US, in which retailers grant credit in order to drive sales). Its policy calls for any delinquent accounts with an outstanding principal of less than 300 yuan (roughly $42USD) to go permanently to an AI caller for handling. Debts above that amount are handled by an automated AI through the fifth day of delinquency, then shifted to a human (if necessary) for further follow-up on the sixth day. For its own analysis, the company also allocates a random 10% of debts over 300 yuan each month into a head-to-head trial, in which half get AI callers for those first five days, and the other half are contacted only by humans from start to finish. <\/p>\n\n\n\n

The research paper\u2019s authors were given access to the company\u2019s data from April 2021 to December 2023, covering 22 million delinquencies that met their criteria (you can read the full paper on <\/a>AccountsRecovery.net<\/a>). The outcomes were instructive:\u00a0<\/p>\n\n\n\n