Chatbots that never scale, credit engines stuck in testing, fraud tools that look good in a demo but don’t survive integration – banks have plenty of AI experiments, but far fewer success stories.
Research suggests that only 11 per cent of banks have generative AI fully live in production, even though 43 per cent are still in the process of rolling it out. Most executives agree adoption is now a matter of competitiveness: more than 80 per cent say banks that fail to implement AI will fall behind their peers. Yet for many institutions, the step from trial to large-scale deployment remains elusive.
It’s this reality that Jason Cao, CEO of Huawei Digital Finance, set out to address when he launched the company’s new FinAgent Booster (FAB) at the recent Huawei Connect 2025 conference in Shanghai. He suggests that banks might want AI, and the models exist, but they still lack the engineering, workflows and supporting infrastructure to actually make it work.
“AI in finance is already evolving from an assistant role to core business scenarios such as customer engagement, risk management, even end-to-end processes,” he said. “But from the institution’s point of view, there is still a lack of many, many things. It sounds fancy, but not so easy.”
Making AI stick beyond the pilot stage
Cao describes FAB as Huawei’s way of bottling up the engineering lessons the company has learned from years of working with financial institutions.
Instead of every bank building agents from scratch, FAB provides ready-made workflows and connectors designed to shorten the gap between a promising demo and a production service. FAB already includes more than 50 scenario workflows and demos drawn from real financial use cases, giving banks a head start instead of a blank page.
The aim is not to replace banks’ own systems but to give them a set of templates and tools that slot into existing processes – whether that’s plugging into legacy platforms through MCPs (micro-component plug-ins) or supporting new AI-native applications.
That balance could prove most valuable for mid-sized banks. The largest players often have the money and specialist teams to grind through long deployments, but smaller institutions are under pressure to modernise with fewer resources.
As Cao puts it, speed matters: experimenting quickly, making mistakes early, and moving forward without having to reinvent every workflow. FAB, he argues, is designed to lower the barrier so that even banks without deep in-house AI teams can get agents into everyday use.
“It’s like engineering,” Cao said. “If one person has already tested many ways and found the best workflow for, say, a loan process, the next bank doesn’t need to repeat all that work. They can just take the workflow we’ve prepared and move forward more easily.”
From legacy systems to global rollouts
One of the hardest realities for banks is that no two environments look the same. Each institution carries its own mix of workflows, compliance rules and legacy systems. That makes any ‘plug-and-play’ promise sound ambitious. Cao acknowledges the challenge but says FAB is built to handle those differences.
“We have a lot of legacy applications, but banks are also building new AI-native applications,” he explained. “For the AI-native ones it’s faster, but you still have to connect with legacy. With FAB, we provide functions to make that easier. Using MCPs [micro-component plug-ins], your agent doesn’t only work with AI-native systems, it can also connect with legacy systems. In this way the connection is much easier, based on our engineering experience.”
Huawei says it has accumulated more than 150 MCPs so far, covering functions across banking, insurance and securities: the kinds of common processes that otherwise slow down adoption. That global practicality matters, because AI regulation and system maturity differ widely from market to market. FAB, in Cao’s telling, is meant to help institutions move faster without having to rebuild everything for each new environment.
Why speed is the real advantage
Cao argues that the real differentiator isn’t just having the right models or tools but moving fast enough to learn what works. In his view, banks that hesitate risk wasting time chasing a perfect plan instead of getting practical experience.
“The work with AI doesn’t have a proven way,” he said. “There’s not a clearly defined path, so people are exploring. In this case, speed is very, very important – even if you make mistakes, it’s better to make them earlier.”
On the engineering side, FAB is tuned for speed as well: Huawei reports that its customer-facing agents can hit over 90 per cent accuracy in intent recognition while delivering responses in milliseconds.
Huawei’s approach, he added, is to co-create with financial institutions and share lessons between them. A solution tested in one market can then be refined and reapplied elsewhere, saving time for the next bank down the line. That cycle of trial and reuse is what FAB is meant to accelerate.
AI is a long game, not a quick win
While banks often ask about immediate ROI, Cao believes AI needs to be treated as a longer-term investment. Expecting too much, too soon, he warns, can backfire if leaders set unrealistic goals.
“We definitely should not underestimate the value AI can bring in the long run, but we also cannot overestimate what it can do in the short run,” he said. “Sometimes people think next year AI will bring big value, but such expectations can hurt an organisation. ROI is not so clear today. It’s like a child – you can’t ask a five-year-old what return they bring you, but you know they are growing.”
For Cao, the point is not to hold back, but to treat AI adoption as a step-by-step process. Banks that start early, move quickly and build on shared lessons are the ones most likely to see real business value in the years ahead.
The post Jason Cao on How Huawei’s FinAgent Booster Is Turning AI Pilots Into Everyday Bank Platforms appeared first on The Fintech Times.
