Hidden Bank Fees Exposed: How AI Will End Fee Smuggling

Agentic AI tools now act as persistent financial advocates, reading every line of every statement on a customer's behalf and flagging charges that would previously have gone unnoticed for years. When software can instantly quantify what a "ledger administration fee" costs annually and benchmark it against alternatives, the information asymmetry banks relied upon simply ceases to exist.

There is nowhere left to hide.

For decades, banks built an elegant asymmetry into the relationship with their customers. The bank had a team of specialists, the time to design products, and the lawyers to draft disclosure documents that were technically correct and practically unreadable. The customer had a life to live and, at the end of a long day, roughly zero appetite to reconstruct what “ledger administration fee” actually meant in rand terms across a calendar year. That asymmetry is not weakening or gradually eroding under competitive pressure, it has been structurally removed, and the instrument that removed it is sitting on your phone.

1. The Architecture of Opacity

Fee smuggling is not accidental. It is a design philosophy refined across generations of retail and business banking, and the mechanism is straightforward: distribute the cost of your margin across enough line items, wrap each one in jargon sufficiently opaque to discourage inquiry, and rely on the one absolute truth about your customer base, which is that they are busy people who will absorb a small irritation rather than spend four hours on a Saturday reverse engineering your fee schedule.

The toolkit has always had three instruments. The first is the named but inexplicable fee, “ledger fee,” “statement retrieval levy,” “account maintenance charge”, each of which appears on a statement and carries just enough bureaucratic gravitas to discourage follow up, because nobody calls their bank to interrogate a R12.50 ledger fee and that, precisely, is the point. The second is the bundle fee, where you pay for thirty things packaged as one product, use six of them, and absorb the marginal cost of the twenty four you never touch without ever seeing it broken out, because unbundling would expose the arbitrage and nobody is going to offer to unbundle it for you.

The third instrument is the spread, and it is the most sophisticated of the three because it leaves no trace. When you pay a USD subscription on your business card and your bank applies an exchange rate that sits 3.7% away from the interbank midrate, no fee appears on your statement. There is no line item, no disclosure, simply a number that is larger than it should be and presented as though it were just the rate. It is indistinguishable from the real rate unless you already know what the real rate is and have both the knowledge and the time to verify it, and the model has always depended on most customers having neither. Multiply that invisible spread across thirty SaaS subscriptions and you have manufactured a meaningful revenue stream from nothing, disclosed nothing, and the customer never had the information required to object to anything in particular.

2. The Loyalty Laundering Problem

Banks have historically offered a partial answer to customers who noticed the cost accumulating, and that answer is rewards. The problem with that answer is that rewards programmes do not return your money, they return a fraction of your money, at a time and in a form of the bank’s choosing, typically in a currency you did not ask for and may not fully value. Offering a rewards programme to a customer whose margin you have manufactured through rate opacity is not a relationship gesture. It is more like a burglar who breaks into your house on a Tuesday and comes back on Wednesday to return the microwave, framing the gesture as evidence of a fundamentally caring relationship.

Genuine loyalty recognition looks structurally different. It means the bank that has held your deposits for eight years charges you less than it charges a new customer, because you are demonstrably less risky and more valuable. It means the fee structure becomes more transparent as the relationship deepens, not less. It means you are recognised as a participant in a long term arrangement and not processed as a revenue unit who has simply not yet found the switching process worth the effort. The distinction matters because the conditions that made customers tolerate the difference have changed.

3. What Actually Happened This Week

Colin Iles is a B2B marketer in South Africa who asked an AI model to do something that would previously have taken a specialist accountant the better part of a working day: pull his bank statements, calculate his actual monthly banking cost in full including FX spread effects, and compare that figure against the published fee schedules of every major South African bank. The result was not ambiguous, his current business account was running at roughly four times the cost of the cheapest credible alternative, and the dominant driver was not the headline fee he already knew about but the FX spread being applied invisibly across his SaaS subscriptions. He published the finding and within hours it had tens of thousands of impressions, which tells you something about how many people have been carrying a version of the same suspicion without the means to confirm it. You can read his post here: Colin Iles Banking Fees Comparison

The AI did not do anything exotic. It read documents, did arithmetic, and compared numbers against published rate schedules, all tasks that humans have always been technically capable of performing. The constraint was never intelligence, it was time and friction, and both of those are now gone. The asymmetry that the opacity model depended on has been structurally removed.

4. Agentic Proctology

The phrase is unglamorous by design, because what we are describing is a rigorous examination of the places where value disappears and nobody previously had the attention span to look. Agentic AI systems are particularly well suited to this work because they do not experience the psychological friction that causes a human to abandon a fee comparison exercise after the third PDF of published rate schedules. They will read every document, apply every comparison, surface every discrepancy, and produce a structured output that a human can act on in ten minutes rather than across a weekend.

The methodology Colin demonstrated is not complex. You load your statements, ask the model to calculate your actual cost, ask it to compare that cost against published alternatives, ask it specifically about the FX treatment applied to international transactions, and ask it to produce a report. That entire process is now available to any business owner with a laptop and a subscription to one of the major AI models, the barrier is low enough that it will become routine, and at the point where it becomes routine the information environment that fee smuggling depends on no longer exists.

5. The Business Model Question

If your banking product relies on fee structures that customers would not consent to if they understood them, you have a problem that the next product cycle will not solve, because this is not a technology problem that a better app or a refreshed onboarding flow addresses. It is a business model problem, and the business model was built on an information environment that is permanently changing.

The opacity model had a long run because information was expensive, comparisons were genuinely difficult, and switching carried real friction. All three of those conditions are degrading simultaneously, and when the comparison becomes clear enough the switching friction becomes worthwhile even for customers who previously would never have bothered. The banks positioned well for this are the ones that never needed the fog, Capitec’s performance in Colin’s analysis was not accidental, it is the output of a pricing philosophy that does not depend on the customer failing to notice something, and that is a structurally defensible position in a way that the alternative simply is not. Transparent pricing also turns out to enable something valuable, which is the ability to compete on genuine value rather than on the customer’s limited bandwidth for investigation, and customers who chose you because your fees are honest are more durable customers than customers who stayed because switching was annoying.

6. No More Hiding Place

The AI is not going to get worse at this kind of analysis. The models will improve, the tooling will become more accessible, and the workflows will eventually be packaged into products that any small business owner can run without composing a single prompt. The one click bank fee audit will be built, it will find an audience immediately, and when it does every percentage point embedded in an undisclosed spread becomes a potential viral moment, every bundled fee structure becomes a potential comparison report, and every rewards programme that returns less than it extracted becomes a potential published finding.

The era of fee smuggling is not ending gradually under gentle competitive pressure. It is ending abruptly, under the scrutiny of tools that do not get tired, do not miss line items, and do not accept that standard industry practice is a sufficient answer to the question of whether a customer received what they paid for. If your business model depends on the customer not looking too closely, the customer is about to look very closely indeed.

This post was prompted by Colin Iles’ excellent analysis of South African business banking costs, published on LinkedIn this week. The methodology he used is available to anyone. The findings are worth reading: https://www.linkedin.com/posts/coliniles_i-asked-claude-which-bank-would-work-out-share-7460934475145994240-_fLu?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAgXxf8Bv8Uc33KhkxGlawRhGdJ63UyIZb8

7. Right, But How Do I Actually Do This?

You have read this far. You are now either mildly annoyed at your bank or deeply annoyed at your bank. Here is exactly what you do next. This is written for someone who has never used an AI model before and has no intention of learning to code anything.

Step 1: Get your bank statements as PDFs.

Log into your business banking portal and download your last three months of statements as PDF files. Almost every South African bank lets you do this from the accounts or statements section of their internet banking. Save them somewhere you can find them, your desktop is fine.

Step 2: Download your SaaS subscription details if you have them.

If you pay for Google Workspace, Microsoft 365, Xero, AWS, Slack, or any other USD or EUR denominated service on your business card, download or screenshot those invoices too. The AI will use these to cross reference what you were billed in rand versus what you should have been billed at the interbank rate.

Step 3: Get Claude or ChatGPT.

You need one of two tools. Either works.

For Claude: go to claude.ai on your phone or laptop, create a free account with your email address, and you are in. The free tier handles this task comfortably.

For ChatGPT: go to chat.openai.com, create a free account the same way.

If you want the most thorough analysis, the paid tiers of either (roughly R200 to R350 per month) give you larger document handling and more detailed output, but the free versions will do the job for three months of statements.

Step 4: Upload your statements.

In Claude or ChatGPT, look for the paperclip or attachment icon in the chat input box. Click it, select your PDF statements, and upload them all at once. Both tools accept multiple PDFs in a single conversation.

Step 5: Paste this prompt exactly.

Once your files are uploaded, paste the following into the chat and press send:

“I have uploaded my business bank statements for the last three months. I want you to do a complete cost analysis. Please: (1) list every fee charged, what it is called, how often it appeared, and the total cost over the period; (2) identify any foreign currency transactions and calculate the effective exchange rate I was charged versus the interbank midrate on those dates, noting the spread in percentage terms; (3) give me a total monthly banking cost including estimated FX spread effects; (4) compare that total against the published fee schedules for Capitec Business, FNB Business, Nedbank Business, Standard Bank Business, Absa Business, Investec Business, TymeBank Business, and Discovery Bank Business for a comparable account type; (5) identify which bank would have been cheapest for my actual transaction pattern and by how much. Present the output as a clear table where possible.”

Step 6: Read the output and sit with it for a moment.

The model will work through your statements and produce a structured breakdown. It will not always be perfect, particularly if your statements have unusual formatting, so read the output and ask follow up questions if something looks wrong. You can ask things like “can you check that FX spread calculation again” or “I do not recognise this fee, can you explain what it might be” and it will respond in plain English.

If the results show a meaningful gap between what you are paying and what you could be paying, you now have the specific numbers, the specific line items, and the specific comparison you need to either call your bank and ask for an explanation, or to start a switching conversation. The whole process takes about fifteen minutes.