The Rise of Agentic AI in Fintech
Agentic AI changes fintech operations by taking over the sequence of lookups between systems, not the decisions at the end of them. The interesting shift is not that a model can reason; it is that a model can carry a task across four dashboards that a human used to open one at a time.
The unit of work was always a sequence
Support work in a consumer finance app looks like one decision — refund, explain, escalate — and is really a dozen lookups stacked underneath it. Which product is this, when did the transaction settle, has this customer asked before, what does the risk flag say, is the partner's status the same as ours. Each lookup is trivial. The cost is that they live in different systems and a human has to be the bus between them.
That is what agentic automation actually removes. Not judgement — the judgement is seconds — but the ferrying. The first time I built a workflow that did this end to end, the measured result was a 40% cut in manual intervention, and what that number is really counting is somebody stopping opening tabs.
Context-aware is the load-bearing half
An agent that can call tools is not interesting on its own; most of the failure modes come from exactly that. The useful property is context-awareness: knowing which lookups matter for this customer at this moment, and stopping when the answer has arrived. Most of the work in building one is telling it when not to continue, which is the opposite of what the demos emphasise.
The second thing demos under-emphasise is escalation. An agent that cannot say "I am not allowed to decide this" will decide it anyway, because the model has no innate sense of permission. Permission has to be built as an edge in the system, and it has to be reachable by the customer as well as by the operator. In my day job the escalation paths are the design; the agentic workflows run inside them.
What I expect to look ordinary in two years
The boring prediction is that agents become plumbing. The visible part of a support team will not be an AI answering customers; it will be fewer handoffs, and a queue where the human sees everything the machine already found the moment they open the ticket. Whoever gets the context assembly right wins, and it is not the team with the cleverest model — it is the team that wrote down what the human was looking up and why.
Filbert Henrico is a Technical Customer Success Lead at FINETIKS in Tangerang since June 2025, with seven-plus years of customer success and technical sales across Southeast Asian fintech, including Xendit, Peakflo, and Brick. His focus includes API integrations, agentic AI, workflow automation, conversational AI, UAT, and launch readiness. Bachelor of Marketing, Victoria University, Bandar Sunway, Malaysia, 2016–2019.
Stuck on something similar?
These pages describe how I approach the work, not a fixed methodology. If you are scoping something similar and want a second opinion on the trade-offs, send me the problem in the messy state it is actually in.
Keep reading
All essaysWhich AI use case should a team build first?
Build the one where a human already does the task every day, the correct answer is easy to recognise, and a wrong answer is cheap to undo. Those three conditions are what let you finish the project: the workflow gives you data, the recognisable answer gives you a test, and the cheap undo gives you permission to ship.
How do you automate customer support without breaking escalation?
Map every escalation rule a human currently applies, and automate only the work that sits above the rules. Escalation is not a feature you add to an automated flow — it is the boundary of the flow, so the boundary has to exist and be written down before anything is automated. Teams that automate first and define escalation later end up with a system that handles everything confidently, including the things it must not touch.
What does production readiness mean for a voice AI agent?
A voice agent is production-ready when it can end a call safely without ever needing to be clever. That means four things are tested against real audio, not transcripts: it recognises the caller's actual speech, it handles being interrupted, it has somewhere to hand off to a human, and every action it takes against a real system has a limit it cannot exceed.