VIEWPOINT: Providing conversational AI in hours, not days

Ask an end customer what they want from their finance provider, and the answers are rarely exotic. They want to move a payment date because their payday has changed, find out what settling early would cost, or update their details after moving house. But more importantly, they want to achieve this in the time it takes to send a message, without holding for an agent or searching through an app for the right screen.

The gap between that expectation and the reality of originations or servicing operations is expensive. Requests like these still flow through contact centres, where a trained operator listens, interprets the customer’s needs, and enters the change into a solution such as Alfa Systems on their behalf. Digital self-service has been a great leap forward, but every journey a customer can complete themselves is one that someone has had to anticipate, then design, build and maintain: one screen for changing a payment date, another for requesting a settlement quote, and another for updating an address. The list of requests a customer might reasonably make will always be longer than the list of screens anyone has had time to build.

We've all experienced poorly performing customer service chatbots, but at Alfa we've also seen some genuine successes in use with our customers. We wanted to put our APIs to the test and ‘eat our own dog food’ by trying our hand at one such scenario: where a customer changes the payment due date on their car loan, using their own words rather than following a predefined digital journey.

Proving the idea in less than a day

Everyone at Alfa enjoys a generous time allowance to spend innovating early-stage ideas, but it's not always easy to find a time when everyone's available to work in pairs or groups. On top of that allowance, we organize four full-day events spread across the year where people can collaborate on ideas, many of which have made their way into our core product, Alfa Systems, as well as our broader operational processes. Everyone naturally aims to achieve something within a single day. Ideas have mushroomed in complexity over the years as people's ambitions grow, and this has often required us to come up with creative ways to bootstrap parts of the technology stack that aren't important to the task at hand. Doing so maximizes the time we spend doing the thing that matters - experimenting, not setting it up.

We chose an off-the-shelf agent platform called Chatbase, which is used to build AI agents that can be trained on business data, deployed into user-facing channels, and connected to custom actions. Those custom actions are then configured to call external tools and services (like Alfa Systems), with the aim of configuring the agent to walk the user through the change payment date business process via Alfa Systems' APIs. Connecting it to Alfa Systems was simply a matter of configuring new API calls and providing the AI agent with sufficient context to determine when to use them and how to handle their responses. In just a couple of hours, we had a customer conversation making real calls into an existing Alfa Systems environment and completing the payment date change. This was an experiment rather than a production deployment, but that was exactly what we needed: enough of the end-to-end interaction to understand whether the idea had shape, where it broke down and what would be required to take it further.

The low setup cost made the experiment accessible to a wider range of people at Alfa, whether they had technical expertise or were completely new to AI concepts.

Most of the day was spent describing behaviour in natural language and iterating on the interaction without writing any code. Configuring the APIs remained the most technical step, but this activity is well documented and supported by Thea Chat, Alfa's domain-tuned AI assistant. With direct access to integration guides, API references and system documentation, AskThea makes the process approachable even for users unfamiliar with Alfa Systems. The modest initial outlay also means we can broaden our experimentation to assess a wider range of processes; it lowers the cost of curiosity.

Where the capability actually lives

The AI tooling turned out to be the least remarkable part of the day. Commoditized agent platforms are making it increasingly straightforward to interpret a customer’s request, collect missing information, and call an external service. The platforms will continue to develop, and providers may choose different technologies as the market matures, but the harder-to-replicate capability sits underneath the conversation.

Everything the agent did was possible because Alfa Systems already exposes granular APIs across the servicing lifecycle, and already applies the business rules governing what a user can and cannot do. The agent knew how to change a payment due date safely, long before an AI agent was involved. Decades of investment in that depth have allowed a relatively thin conversational layer to produce a working customer outcome in a single day, using an existing, unmodified product environment. The AI provided a new way for the customer to express their intent; Alfa Systems provided the controlled servicing capabilities needed to act on it.

The distinction also prevents providers from becoming dependent on a single conversational technology. Alfa Systems’ APIs are agnostic about what sits in front of them: the same capability could support a contact centre application, a customer portal, a mobile experience or whichever agent platform a provider chooses.

For providers considering where AI belongs in their servicing operations, this changes the starting question. Choosing an agent technology matters, but so does the platform underneath it - can it expose the right servicing capabilities, apply the necessary controls, and support new channels without rebuilding the underlying process each time?

For Alfa Systems customers, that foundation is already in place.