On 23 July 2026, I had the pleasure of hosting and facilitating an EXUS Virtual Roundtable with Collections and Recoveries leaders from banking organisations across Jordan, Indonesia, Mauritius, Papua New Guinea, Central and Eastern Europe, and North Africa.

The session was supported by Marios Siappas, EXUS Chief Knowledge Officer, Dimitris Papadopolous, EXUS Chief AI Officer, and Miranda Ko, our Product Marketing Specialist. Together, we explored a question that is becoming increasingly urgent for collections teams everywhere: How can AI move from promise to practical impact?

What made the discussion especially valuable was the diversity of perspectives. This was not a theoretical conversation about AI in ideal conditions. It was a grounded discussion shaped by different markets, regulatory environments, levels of technology maturity, customer behaviours, and operational realities.

That is where one of EXUS’ core strengths came through clearly: Global Insights, Local Impacts.

AI in collections cannot be treated as a single universal playbook. The global themes are consistent, but the way they need to be applied locally can differ significantly.

Some organisations already have machine-learning models in production, predicting forward flow, self-cure, bucket deterioration and customer propensity. Others are still focused on building the foundations: core collections technology, digital communication channels, reliable data, and the operating model needed to support future AI adoption.

Both positions are valid. AI maturity is not only about having a model live in production. It is about whether the organisation has the data, processes, governance and people needed to turn intelligence into better decisions.

During the roundtable, we discussed four practical areas where AI is beginning to reshape collections:

1. Predictive intelligence
Using account, payment and behavioural data to estimate outcomes such as self-cure, collectability, forward flow and deterioration.

2. Strategy augmentation
Turning predictive signals into next-best actions, including treatment selection, prioritisation, contact timing and channel choice.

3. Digital customer engagement
Using generative AI, chatbots and virtual agents to support straightforward early-stage interactions through channels such as WhatsApp, SMS, Viber and web messaging.

4. Human productivity and assurance
Supporting agents and supervisors with live scripting, call summaries, performance analysis, quality assurance and compliance review.

A major theme was contactability. Many collections teams still rely on repeated contact attempts across different channels and times of day, without always knowing which attempt, channel or timing actually led to the outcome. Participants discussed how behavioural signals, such as mobile-app activity, login times, previous responses and channel preference, could help teams make smarter decisions about when and how to contact customers.

But the discussion also recognised the limits of behavioural data. A mobile-app signal may be useful in early delinquency, but less meaningful if a customer stops using the app as their situation worsens. Different customers also respond differently to different channels. Some may not be able to answer calls during working hours, but may respond well to SMS or another asynchronous channel later in the day.

This is where local context matters. The best contact strategy is not simply the most automated one. It is the one that reflects customer behaviour, regulatory boundaries, consent, cybersecurity requirements and operational capacity in each market.

We also discussed the role of existing infrastructure. AI adoption does not always require organisations to replace what they already have. For example, EFS can integrate with established dialler and telephony platforms, using low-latency speech-to-text to convert voice interactions into transcripts. Those transcripts can then support real-time scripting, call summarisation, performance analysis, quality review and compliance monitoring.

In that sense, AI can operate as an intelligence layer around existing contact infrastructure, rather than forcing a disruptive replacement of it.

Measurement was another important topic. Participants identified several ways to assess AI performance, including control measures, financial outcomes, operational efficiency and customer impact. Cure rates, collectability, kept arrangements, cost-to-collect, handling time, customer satisfaction, complaints and contact frequency all have a role to play.

However, benchmarking needs care. Comparing a virtual agent with a human agent only makes sense if the comparison accounts for case complexity, customer segment, delinquency stage, treatment eligibility and channel availability.

Governance was perhaps the most important thread running through the discussion. In some markets, cloud deployment is restricted and AI must operate on-premise. That creates infrastructure, cost and resourcing implications, especially where GPUs and specialist technical skills are required. Cybersecurity requirements can also affect which communication channels are available for customer engagement.

Explainability is therefore essential. Organisations need to understand what action the AI recommended, which data and rules informed that recommendation, whether a human intervened, and what outcome followed. Governance cannot be something added at the end. It needs to be part of the design and operating model from the beginning.

Finally, the roundtable reinforced something that is sometimes overlooked in conversations about AI: people remain central.

AI can reduce routine administrative work, support repetitive interactions and help teams make better decisions faster. But vulnerability, empathy, unusual circumstances and complex negotiation still require human judgement. The goal is not to remove people from collections. It is to enable experienced teams to spend more time where they add the most value.

For me, the roundtable was a strong reminder that AI in collections is not just a technology conversation. It is a strategy, governance, data, people and customer conversation.

The opportunity is global. The impact must be local.

Turn AI ambition into practical collections impact

Every collections organisation is at a different stage of its AI journey. Whether you are strengthening your data and technology foundations, exploring predictive models, or looking to scale AI-enabled customer engagement, the right approach must reflect your market, operating model and regulatory environment.

Talk to an EXUS expert to explore how AI can deliver measurable, responsible and locally relevant impact across your collections and recoveries operations.