Becoming AI-first, with the human still in the loop

At the FCI 58th Annual Meeting in Lisbon, Trade Treasury Payments (TTP) spoke with Karol Leszczyński, Product Development Manager at Comarch, the Poland-based software house of around 5,000 people that builds technology for financial institutions, insurers, healthcare and beyond.

While much of the industry conversation centres on advising clients to adopt AI, Leszczyński was keen to look inward and talk about what a technology provider should be doing within its own walls.

“In a company like ours,” Leszczyński said, “implementing AI is a challenging task, because there are different professions across the company. A programmer will use a different tool than an analyst, so we need to be sure each person uses the right one. The most important thing at the beginning is education — understanding what AI can give us, the risks and the opportunities. That’s why we created an AI academy, with four levels of certification that everyone, including the CEO, has to pass.”

The response internally has been positive, perhaps because, rather than displacing people as many AI commentators have feared, the tools have shifted where their time goes.

“People are able to work more efficiently,” Leszczyński said. “Developers who used to write simple code can now hand it to AI and focus on the more sophisticated work, and it’s the same story with our data analysts. At the beginning, there was some fear that they would lose their jobs, but nothing like that happened. They are simply working in a more productive way.”

That same principle (using AI to augment rather than replace) carries over into the product. On the Comarch Factoring platform, the team began with payment reconciliation, which is a module that clients themselves have identified as having the most to gain.

“Around 90% of invoices are already matched automatically in our system,” Leszczyński said, “but there was still a gap of 10%. Working on reconciliation with AI, we are able to reach almost 100%. But we still don’t let AI do all the work automatically. AI gives the information to the operator, and at the end of the day, it is still a human decision.”

Looking further ahead, Leszczyński sees the dashboard itself giving way to conversation, driven by a generation that expects answers in real time.

“A few years from now, the next generation of software will be pure conversation with the system,” Leszczyński said. “You will see a white screen, ask a question about a debtor or a report, and receive exactly the information you need at that moment. People will expect a real-time decision, real-time rating, real-time money in the account. They will not wait one, two or three weeks.”

In this type of environment, the longer-term challenge may not be how to develop the technology that is able to do this, but instead be about how to keep pace with it.

“I truly believe that 20 years from now, no one will talk about AI,” Leszczyński said. “We will be used to having it in our software, our cars, our homes, everywhere. The challenge is to be first and not to fall asleep behind the wheel — to make sure we are implementing the right solutions and heading in the right direction. And we can only do that through conversation with our clients, asking what they need.”