# HMRC deploys AI to categorise businesses by ‘propensity to pay’ tax debts > Source: > Published: 2026-09-15 13:40:37+00:00 ##### Department has created an algorithm that is used to support the work of collectors by placing firms into segments based on what it determines is the likely timeliness of payment HM Revenue and Customs has shed light on details of an artificial intelligence-powered algorithm used to categorise millions of businesses based on their “propensity to pay” tax debts. Once firms have been segmented, the tool is then used to guide how the department’s collection teams communicate with firms, *PublicTechnology* understands. A newly published transparency record reveals that the propensity to pay (PtP) tool “uses machine learning to analyse past debt-repayment behaviour to predict businesses’ likely engagement with the debt-recovery process”. Insights derived via the algorithm are then used “to segment debtors and inform the debt journey, including the frequency and content of letters sent by HMRCs Debt Management function, [which] helps to optimise the recovery of overdue tax”, the record says. An example of how this works in practice is that, where PtP detects signs of financial difficulty on the part of a business, HMRC’s comms would be more likely to focus on the department’s Time to Pay offering, which provides the option of repaying arrears via an instalment plan, rather than in one lump sum. It is understood that all four million UK-registered businesses in scope of one or more elements of HMRC’s tax regime are segmented via the technology. New assessments are made by the model each month – which can result in firms being moved from one segment to another. Automated checks are designed to detect any anomalies in monthly changes, while the operation and efficacy of the overall model is subject to a broader yearly assessment. HMRC debt-collection staff, meanwhile, are given “training and guidance to understand what the PtP score that they can see on their system means, and how they might want to take this insight into consideration to support their conversations with debtors”. The tool does not itself determine the payment arrangement to be put in place. This is only decided following conversations between HMRC and the company in question, *PublicTechnology* understands. The transparency document says: “Members of the public cannot appeal decisions made by the tool, as the model output is not an outcome and no decision is made as a direct result of the PtP model. The debt collection journey would happen regardless of the tool and the PtP model is only one factor in determining the tailored part of the debt collection journey. At any point the customer engages we will agree an appropriate outcome, such as payment or instalments.” The department indicated that these kinds of segmentation tools are widely used in debt recovery and financial services. “We’re always looking at ways to better understand our customers and the insight from this model is just one of many tools we use to help us decide how to support businesses with tax debts,” an HMRC spokesperson added. “We want to work with businesses to find the best possible solution, which can include paying what they owe in instalments.” ### Related content - [HMRC algorithm for automated council tax valuations offers ‘greater consistency and quality’](https://www.publictechnology.net/2025/08/06/economics-and-finance/hmrc-algorithm-for-automated-council-tax-valuations-offers-greater-consistency-and-quality/) - [HMRC shared 5,000 citizen records with local councils in pilot ‘to analyse vulnerable and overlapping debtors’](https://www.publictechnology.net/2021/05/06/economics-and-finance/hmrc-shared-5000-citizen-records-local-councils-pilot-analyse-vulnerable-and/) - [HMRC Transformation Roadmap: New citizen services, more revenue and ‘cultural change’ across department](https://www.publictechnology.net/2025/07/23/economics-and-finance/hmrc-transformation-roadmap-new-citizen-services-more-revenue-and-cultural-change-across-department/) The operational record states that “debts from VAT, PAYE and Corporation Tax are first linked at a debt customer level”, before “a gradient-boosted decision-tree model analyses historical data, including data on past debt repayment behaviours and predicts the likelihood of a customer repaying their debt in the near future”. A decision tree operates like a flowchart, moving through a journey of various scenarios and options before arriving at an ultimate projected outcome. Following completion of this process, firms are given a PtP score between 0 and 1 that is then “fed into downstream IT systems” and combined with other sources of information to support ongoing engagement between HMRC and the business. The development of the algorithm built on work that took place during the coronavirus crisis in which HMRC created “a simple model which segmented debtors based on how impacted they may have been” by the pandemic. Before Covid, “the majority of customers received the same debt collection journey instead of a more tailored one”, the record says. In its efforts to establish a more sophisticated model to be used in the longer term, HMRC determined that “simpler methods, like logistic regression, were not… suitable”. Instead, “a variety of supervised machine learning techniques were considered, specifically logistic regressions, decision tree based models and neural networks, [before] decision tree models were chosen over neural networks due to being explainable and less computationally expensive”. The [record](https://www.gov.uk/algorithmic-transparency-records/hmrc-business-propensity-to-pay) adds: “The performance of different types of decision tree algorithms (single decision tree, random forest and gradient boosting) were compared, and gradient boosting was selected due to having the highest performance (accounting for the risk of overfitting).” The tool is run using Viya AI and analytics platform offered by tech giant SAS.