Who can See Your Customers’ Data? Follow One Record Through an Offshore Fintech Operation
Banking security conversations gravitate toward the intruder: the stolen credential, the phishing campaign, the ransomware note. That emphasis is understandable, and it leaves a gap. A great deal of financial data exposure involves nobody breaking in. It involves people who were given access legitimately, using systems that were approved, under permissions nobody has revisited since the week they were granted.
For any fintech running part of its operation through an external partner, this is the exposure that repays attention. It is also what a standard procurement questionnaire is least equipped to uncover.
The most direct way to see it is not to read a policy. It is to pick one customer record and trace every place it stops.
Stop one: the agent’s screen
A customer contacts support about a declined payment. An agent opens the record.
Ask where that agent is physically sitting, and on what device. A supervised production floor and a laptop at a kitchen table present very different risk, and plenty of operations run both without the distinction appearing anywhere in the contract. Firms evaluating fintech back-office operations in the Philippines or any other offshore location should establish this at the level of data category rather than as a blanket answer — which record types may be opened outside a controlled site, and what technically stops the rest.
Any partner who treats this as a one-word answer has not examined it. The credible answer is a segmentation rule with enforcement behind it.
Stop two: the notes
The agent types a summary. Here is where a surprising amount of sensitive material gets created rather than merely viewed.
Free-text fields routinely accumulate things no schema anticipated: a partial card number pasted for reference, a customer’s explanation of a medical situation behind a missed payment, a screenshot attached to a ticket. None of it was designed into the data model. All of it is now in your estate, often in a system with looser controls than the core banking platform.
Worth asking: are free-text fields scanned for sensitive patterns, and what happens when one is found?
“Nobody has ever found this problem by reading a policy document,” said John Maczynski, chief executive of PITON-Global, the Philippines-focused BPO sourcing advisory. “You find it by taking one record and asking, at every stop, who touched this and why were they allowed to.”
Stop three: the quality sample
A proportion of interactions gets reviewed. That means a second person, usually with broader access than the agent, pulls up the record, the recording and the transcript.
Quality platforms are frequently third-party tools, which makes them sub-processors in their own right. They are also frequently absent from the sub-processor list, because they were adopted by an operations team as a productivity tool rather than procured as a data system.
Stop four: the model
This is the newest stop on the journey and the one most agreements do not describe at all.
If an automated system drafts a reply, classifies the dispute, scores fraud risk or summarises the case for a supervisor, that system has read the record. Where does it run? Under whose contract? Was it in the assessment anybody signed, or was it added eight months later because it made the team faster?
Automation layers tend to arrive incrementally and for good reasons, each addition sensible on its own. Collectively they reshape the processing estate without anyone updating the document that describes it.
Read: Some of the Most Effective Business Lending Companies for Service and Product Based Small Businesses
Stop five: the archive
Eventually the interaction is stored — the ticket, the recording, the transcript, the quality score.
The question here is not whether retention policy exists. It is whether anybody can find a specific record again. Ask what the process is for locating a single interaction from several months back, who carries it out, and how long it took the last time the partner tested it internally. During selection that is a question about process, not a request to see another client’s material. After go-live, try it yourself on a record of your own.
If a record cannot surface inside the window a regulator allows, that is a failure in its own right, however well guarded the data was at every earlier stop.
The control that tells you most
Having traced the path, one question reveals more about the partner than the rest combined: how quickly does access end when somebody leaves?
This matters because of what it measures. Policies are written once. Access revocation happens continuously, every time an employee resigns, transfers or is dismissed — which in a service operation is often. A partner who can state the median elapsed time from exit to deactivation, and produce the record behind it, is running the control. One who describes the procedure without the number has a document.
Ask for both the typical case and the worst one over a recent period. The gap between them is usually where the story is.
“Revocation is the one control that gets exercised every single week whether anybody is watching or not,” Maczynski said. “A firm that cannot hand you the number is telling you it does not measure it.”
Why the picture is changing
Two shifts are altering this landscape, and they compound each other.
Smaller teams, broader permissions
The first is team size. With software now closing the routine traffic, support functions across the sector are contracting and skewing more experienced. That is good for quality and awkward for segregation of duties: with fewer people, each necessarily holds broader permissions, and the separation that worked comfortably at eighty people is harder to sustain at twenty.
An estate that keeps quietly growing
The second is the growth of the processing estate described at stop four. More components touch each record than did two years ago, and the documentation has not kept pace.
Taken together, they mean a security assessment ages faster than it used to. The useful discipline is not a more detailed initial review but a trigger for repeating it — a new automation component, a move to hybrid working, a material change in team size — any of which should prompt the trace to be run again.
What to do with the findings
Walking the path above produces a list that is usually shorter and more specific than a questionnaire response, and far easier to act on.
Expect to find at least one system touching customer data that is not named in any agreement. Expect at least one permission set that is broader than the role needs, granted during implementation and never trimmed. Expect the free-text fields to contain material nobody intended to store there. These are normal findings, not signs of a bad partner, and all three are fixable once visible.
What distinguishes a strong partner is the reaction. Some will walk the path with you and volunteer the awkward stops. Others will route you back to the certification pack.
Both answers are informative. Only one of them tells you where your customers’ data actually goes.
