Explainable AI for Fair Debt Recovery in UK Energy

AI you can stand behind: fair, compliant debt decisions for a UK energy supplier

AI next best action for debt recovery, built with customer fairness and compliance at its core

Energy debt is rising, and more of it sits with customers who are struggling. We are on a live journey with a UK energy supplier, leveraging advanced AI capabilities to choose the next best action for every customer in arrears. The approach replaces one-size-fits-all collections with individual, explainable decisions, built for an Ofgem-regulated market where vulnerability, affordability and fair treatment has to be prioritised.

The challenge

The debt challenge is here to stay. 6.6 million UK households, 23%, are financially vulnerable, and 2.2 million are extremely vulnerability. A further 56,000 are forecast to join them in H2 2026, and energy debt is set to rise by £2bn by Q1 2028. Fewer households are in arrears overall, but a smaller group of extremely vulnerable households now holds larger balances that are harder to recover.

For suppliers, this creates a complex and challenging balance:

• Cost: every pound left unrecovered raises the cost to serve.
• Customer harm: every badly judged action risks harming a customer who is already struggling.
• One-size-fits-all collections: customers are put into groups, so everyone in a group gets the same letters at the same pace, whatever their circumstances.
• Regulatory scrutiny: every decision must be fair, explainable and defensible to Ofgem.

Off-the-shelf AI isn't the answer. General-purpose AI is not built for sensitive and thoughtful decisions such as these: they don't know a supplier's customers, they can give different answers to the same question, and they can't always show their reasoning to a regulator. Who builds the AI matters as much as the AI itself.

Our approach

We brought our deep debt collection expertise in energy retail to design the debt strategy and customer journeys and built the AI alongside them. Operational changes and change management moved in step with the technology, so people were taken on the journey and the AI is trusted and adopted, not left on the shelf as a proof of concept.

Our AI drops the groups and treats every customer as an individual. It builds a full picture of each customer from first-party data (account, payments and debt history), third-party data (credit risk, affordability and life events) and interaction data, and it adjusts their treatment as their situation changes. For every customer, the engine weighs each possible next step against its likely outcome and what it costs to deliver. It then picks the step that creates the most value for the business and for the customer, across early collections, mid collections, late stage collections and escalations, each with an accelerated or slower pathway.

Developing the models and building in trust

We developed purpose-built, empirical machine learning models trained on real debt outcomes from the supplier's own history. The model  learns from hundreds of small decision rules, each built to correct what the last one got wrong; together they spot patterns no single rule, or person, could.

• Explainable drivers: every prediction can be traced to the factors behind it, such as debt age and value, payment pattern, past journey outcomes, credit-based risk and engagement.
• From prediction to decision: predicted best treatment for debt recovery is weighed with vulnerability and business guardrails applied first, to produce one next best action per customer.
• Always-on monitoring: model drift, input data drift, calibration, fairness across customer groups and data quality are monitored continuously. When a check crosses its threshold, a retraining loop is triggered, and recommendations improve through feedback from customer outcomes.

Compliance by design

• Guardrails first: vulnerability and fairness rules always come before value.
• Human in the loop: debt collection agencies review every account they receive before acting.
• Assured every day: every vulnerable customer's treatment is checked daily against the rules, giving an extra layer of assurance and a clear audit trail.
• Fully auditable: the model version, inputs and reason are stored for every decision.

The impact

• Individual, fairer treatment: every customer receives a next best action that reflects their circumstances rather than a fixed path.
• Designed to recover more at lower cost: the AI capability targets more debt recovered, lower cost to collect, greater digital engagement and faster resolution.
• Confidence for the business and the regulator: calibrated, explainable and auditable decisions that operations teams, auditors and Ofgem can all understand.
• Adopted, not shelved: AI built alongside the debt strategy, journeys and operations, so it works in the real world at scale.

Beyond debt: the energy retailer of the future

We believe AI, embedded end to end across customer experience, efficiency and growth, will change what an energy retailer does, how it is organised and the skills it needs.

• Experience and efficiency, from processing work to preventing it: AI spots and fixes issues before customers notice, and conversational AI resolves routine contact in the channel they choose.
• Growth, from segments to individuals: every customer gets the right offer, price and next step for them, at the right moment.
• Operating model, from operations teams to an AI-enabled Super Team: product owners, function experts and AI specialists work together to get every decision right.
• Skills, from volume to judgement: a smaller team of highly skilled energy advisors, supported by AI, handles the complex and sensitive conversations.