James Towner, Chief Growth Officer at ArvatoConnect and Martin Stevenson, VP of Sales and Partnerships from XpertRule, explain why our partnership marks a shift towards more transparent and reliable AI in high-stakes complaints environments.
Managing customer complaints shouldn’t just be seen as an operational inconvenience. It’s a defining test of organisational credibility and has high financial stakes.
In the first half of 2025 alone, 1.85 million complaints were recorded by financial services firms, alongside a sharp rise in cases escalated to regulators such as the Financial Conduct Authority and the Financial Ombudsman Service. Not to mention the over £500 million these firms have to pay out in redress every year.
Consumers are more informed. More empowered. More willing to escalate. If they believe they have been treated unfairly, they will pursue it.
There’s also the irony that the very AI tools helping organisations improve, are the same ones empowering customers to draft more complex complaints, meaning their volumes and sophistication are only likely to increase.
For organisations, this creates a fundamental pressure point: scale complaints handling quickly without compromising fairness, compliance or trust. Fail to do so, and the consequences are regulatory, financial and reputational.
Behind the scenes, complaints teams are under sustained pressure. Cases are more complex, calls are longer, documentation requirements are heavier and regulatory expectations are sharper.
Consumer Duty has raised the bar permanently. Firms must now demonstrate consistently good customer outcomes – and evidence them. Yet our report Navigating Consumer Duty shows that just 3% of organisations believe they have fully met compliance standards.
That isn’t a marginal gap. It is a systemic one.
History shows how quickly complaint volumes can escalate. From PPI to car finance commission rulings, complaint events can surge almost overnight, creating multi-billion-pound exposure.
Simply hiring more advisors isn’t a long-term strategy, but a temporary reaction. It takes months to train handlers to competency while operational costs continue to rise and complexity continues to grow.
And then there’s also vulnerability to consider.
Almost half of UK adults identify as having at least one characteristic of vulnerability. Yet 44% say they have never had a positive experience relating to that vulnerability, according to our report AI, Digital Transformation and Vulnerable Customers.
This is the reality organisations are facing with rising expectations, intensifying scrutiny and very little margin for error.
AI is necessary but not all AI is fit for purpose
In this environment, AI is no longer optional but deploying the wrong kind of AI in regulated complaints handling can introduce a new risk.
Large language models have transformed what’s technically possible. They can summarise, draft, analyse and uncover insights at remarkable speed – but they are probabilistic systems. The same input can generate different outputs
In low-risk environments, that variability may be manageable, but in regulated complaints handling, it’s not.
When decisions affect outcomes, regulatory exposure or vulnerable customers, consistency is non-negotiable. Regulators expect decisions to be defensible, customers expect them to be fair and organisations need clear, traceable audit trails that demonstrate exactly how an outcome has been reached.
Complaints handling demands four uncompromising standards: accuracy, consistency, explainability and auditability. Meeting those standards requires more than experimentation with generative agentic. It requires structure, governance and deliberate design.
A smarter approach to regulated complaints
This is why the ArvatoConnect and XpertRule partnership – with XpertRule being an AI specialist with more than 40 years’ experience developing deterministic AI systems for the financial services sector – has been established.
Deterministic AI agents are built for high-impact decision-making, where outcomes must be reliable and repeatable. The same validated answer is delivered every time. No drift. No ambiguity. No unexplained variation.
By combining this type of AI, which provides structured and transparent decision logic with carefully governed generative capabilities, we have developed a complaints solution specifically designed for high-scrutiny environments.
This means agents are presented with relevant information and receive real-time, compliant guidance to help them navigate complex cases without sacrificing speed but also importantly, judgement. Administrative burden reduces, decision pathways remain fully traceable and auditability is embedded from the outset.
It’s not about replacing people. In fact, our research shows that while 56% of vulnerable customers believe AI could meet their needs as well as a human, 78% still want some level of human interaction.
Our solution is designed with a true human-in-the loop approach, with agents keeping the full decision authority at critical moments and the final outcome, and any redress our goodwill measures resting with them as the expert. What changes is the level and quality of support they receive.
In essence, it helps make every agent your best agent, able to respond to complaints at speed.
And when a human chooses to override the system, it’s not hidden, instead it’s understood and captured. This means it’s transparent, testable and predicable – and importantly that regulators, auditors, team leaders and clients can inspect how every step and decision has been reached.
This is particularly vital when stakes are high. In complaints handling, “good enough” is not good enough.
The risk of standing still
The consequences of failing to modernise complaints handling are not theoretical, they are already visible.
Costs will continue to rise, inconsistency will increase, regulatory scrutiny will intensify and customer trust will erode. And as we’ve already seen, the financial exposure is significant with over £500 million paid out in redress each year and multi-billion-pound complaint events capable of materialising almost overnight.
While some organisations are intentionally making it harder and harder to formally complain – and sometimes even insisting on snail mail as a deliberate barrier – there’s a better way. By using AI effectively, it removes the need for these roadblocks altogether. Instead of hiding from complaints, organisations should face them head-on, resolve them decisively and even welcome them as opportunities to improve.
In a high-stakes environment like this, smarter AI, purpose-built for regulated, high-impact decision-making, is not a future aspiration, but an operational requirement and strategic imperative.
