It is easy to forget, in the current wave of AI hype, that most of the real work still happens offline. In call centres. On WhatsApp. Through back-and-forth emails written by tired humans handling repetitive issues that were supposed to have been solved already.
For Adrian Defta, founder of ProveIt, that disconnect is personal. His company helps UK-based Romanians navigate the bureaucracy of complaints and claims, especially in the wake of the car finance mis-selling scandal. But ProveIt did not start as a tech-first company. It started on TikTok, with Adrian filming short clips to explain financial basics to an underserved audience. From there, ProveIt became a service, one that coaches users through complaint processes they have never attempted before.
But success brings scale problems. And scale problems in financial services, especially those involving vulnerable users, are not easily solved by throwing humans at them.
That is where agents come in.
Most people working in financial services have seen what automation can do: improve speed, reduce costs, and handle routine tasks. But automation alone hits a ceiling when the tasks are complex, contextual, and regulated.
Agents, AI-powered systems with goals, memory, and reasoning capabilities, offer a step change.
At ProveIt, the agentic approach is beginning to reshape how the company operates. Rather than replace humans, agents are helping scale the quality and consistency of service without compromising trust. Templates are chosen, emails are drafted, threads are monitored, and sentiment is interpreted. All with human validation where needed, but without the manual drudgery that bottlenecks most small teams.
Crucially, these agents are not static. They are interactive, contextual, and sometimes multilingual, which is ideal for a customer base that mixes English and Romanian in the same email. The agent knows who the customer is, where they are in the process, and which of the 60-plus templates should be applied. And it gets better with every case it sees.
ProveIt’s work in the Dootrix Agentic Incubator is exploring how different agentic patterns can be applied to real-world workflows.
Two have emerged as essential:
The Personal Advisor: This agent guides users through complex complaint processes step by step. It provides instructions, clarifies next actions, and offers assurance, just like a human coach would. In many ways, it replicates what ProveIt staff do every day, but with the ability to scale infinitely.
The Sentinel: This agent watches for anomalies. For example, if a customer receives two identical emails and becomes confused, the Sentinel can detect the duplication, infer that two separate complaints may be in progress, and guide the user accordingly. It does not just see patterns; it interprets them.
Together, these patterns allow ProveIt to deliver a premium, high-trust experience without building an unmanageable back office. They also make compliance easier. Because each agent has a bounded role and a defined process, outputs are explainable, repeatable, and auditable, which is essential in a regulated space.
ProveIt’s story may seem niche, but its implications are not. Every financial services provider has a long tail of processes that involve:
Customer-specific reasoning
Document or template-based logic
Regulator-visible audit trails
Human-in-the-loop decision making
These are exactly the kinds of workflows agentic AI is designed for. Unlike traditional automation, agents can handle variation. They can read between the lines. They can personalise. Most importantly, they can help customers feel heard at a fraction of the cost.
That last point matters. In a world where consumer trust is fragile and compliance risk is high, delivering a supportive, human-like experience at scale is no longer a luxury. It is a differentiator.
ProveIt’s next leap is a proactive one. The team is experimenting with digital personal assistants that not only respond to user requests but also raise awareness of issues a user might not even realise they have. For example, "You might be owed money from this finance agreement. Would you like me to check?"
That is the future: agents that act like advisors, not just assistants. Advisors that understand financial context, regulation, language barriers, and behaviour. Advisors that work around the clock and never get bored of copy-pasting template 34.
It is early days. But the building blocks are already in place: templated reasoning, document categorisation, language understanding, and context-aware memory. All the things traditional CRMs and automation scripts were never designed to handle.
Adrian puts it simply: "You would not run a business today without email. Soon, you will not run one without AI."
But not all AI is equal. The next wave in financial services will not be about chatbots that hallucinate or dashboards that automate. It will be about agents, structured, contextual, feedback-driven systems that work alongside humans to handle the long tail of difficult, regulated, human-facing tasks.
At ProveIt, that journey started not with technology, but with belief. Belief that underserved users deserved more. Belief that trust could be rebuilt. Belief that helping someone write their first complaint letter could change the way they see the system.
Now, it is being scaled by agents. And that is worth paying attention to.