A £6 overdraft fee is the smallest complaint a bank receives. A customer who was in credit sees it, says it isn’t fair, and wants it back. Handled one at a time, each of these complaints is upheld, refunded and closed, and nobody ever asks why they keep arriving. This post follows four of them on the demo firm’s data and shows how the case research links them to one fee run, and how the root-cause analyst turns four small complaints into one question for the people who own the fee engine.

In short

  • Four customers, one run. Muhammad Ellery, Ama Zhao, Leo Somerton and Eilidh Penhale each complained about an unarranged overdraft fee of £6.00 charged on 3 October 2026. Case research shows all four fees came from the same fee run, BATCH-2026-10-03-OD.
  • The run reached 104 customers. The firm’s records show that run charged 104 customers £624.00 in total. Four complained. The other hundred are the reason root-cause analysis exists.
  • The cluster is flagged, not fixed. The root-cause analyst marks “Fee charged in error · Everyday Current Account” as possibly systemic. A person takes ownership; whether to remediate customers who never complained is the firm’s decision.
  • Each complaint is still handled on its own clock. Two are closed and upheld with £6.06 refunded. Two are open, and their acknowledgements are overdue: the clock doesn’t wait for the root cause.

Why one complaint is not enough

Regulators expect complaints to be used for more than the complaint. In the UK, the FCA’s complaints sourcebook asks firms to put in place management controls that identify and remedy recurring or systemic problems, and to consider whether a problem found in one complaint may affect customers who did not complain. The practical difficulty is that the evidence for a systemic problem is spread across handlers, channels and weeks. Each handler sees one £6 fee. Nobody sees the run.

Industry
data

The FCA’s Dispute Resolution: Complaints sourcebook (DISP 1.3) requires firms to have effective and transparent procedures for handling complaints and, in DISP 1.3.3R, to put in place appropriate management controls and take reasonable steps to ensure that in handling complaints they identify and remedy any recurring or systemic problems. The guidance beside it describes analysing the causes of individual complaints to identify root causes common to types of complaint, and considering whether those root causes may also affect other processes or products, including those not directly complained about.

Source: FCA Handbook, DISP 1.3.3R and 1.3.6G. Other regulators have similar expectations; check the rules that apply to your firm.

The four complaints

The demo uses Fernhollow Bank & Insurance, a fictional firm with invented customers, accounts and fees. These are the four complaints, in the customers’ own words as the classifier quoted them:

CaseCustomer and channelReceivedWhat they said
SVC-2026-000010Muhammad Ellery, web form9 Oct 2026“I noticed a charge of £6.00 described as an unarranged overdraft fee. I have never had an overdraft and my account has always been in credit.”
SVC-2026-000011Ama Zhao, chat16 Oct 2026“You’ve charged me an overdraft fee when I had money in the account. That’s not fair and it’s not the first time I’ve seen people complaining about this online.”
SVC-2026-000008Leo Somerton, chat20 Oct 2026“Hello, an overdraft fee was taken on 3 October.” … “My balance was positive, around £3201.” … “I’m quite annoyed to be honest.”
SVC-2026-000009Eilidh Penhale, call note30 Oct 2026Cust said: “You’ve charged me an overdraft fee when I had money in the account. That’s not fair … Sort it out.”

Four channels, four weeks, four different handlers in a real firm. In two of the chats, only the customer’s lines count: the classifier ignores what the adviser typed, so an adviser’s “sorry you’re unhappy” never turns a query into a complaint. In the call note, the customer’s quoted words carry the dissatisfaction. Each contact was classified as a complaint about the current account, and the complaint clock started from receipt.

What case research finds

When a case is researched, the case researcher reads the customer’s accounts and six months of fees from the firm’s records. It looks for two kinds of fee: the same fee charged twice on the same day, and fees the message names. Each of these customers names the unarranged overdraft fee, and each has exactly one, charged on 3 October 2026, for £6.00.

Every fee in the firm’s records carries the fee run that charged it. So the researcher can do something a handler working one case cannot: look at every other open or closed case whose fees in question came from the same run, and ask the firm’s records how far that run reached.

Muhammad Ellery's case, Customer and fees tab: one unarranged overdraft fee of £6.00 charged on 3 October 2026 from fee run BATCH-2026-10-03-OD, a warning that the run charged 104 customers (£624.00) and that 3 other cases complain about the same run, his accounts, and related cases on the same fee run
The fee, and the run behind it. One £6.00 fee on Muhammad Ellery’s Everyday Current Account. The run charged 104 customers £624.00, and three other cases complain about it.

On Muhammad Ellery’s case the research says: one fee in question (£6.00); no prior complaints; three other cases on the same fee batch; 157 similar complaints from other customers in the inbox. The related panel lists the three cases on the same run by reference, and separately the cases that share a root cause in the last 90 days.

Fee run BATCH-2026-10-03-OD

Muhammad Ellery, 6.00

Ama Zhao, 6.00

Leo Somerton, 6.00

Eilidh Penhale, 6.00

100 more customers, no complaint

SVC-2026-000010

SVC-2026-000011

SVC-2026-000008

SVC-2026-000009

One run, 104 customers charged, four complaints. The fee run is in the firm’s records; the cases link to it through research.

From four cases to a cluster

The root-cause analyst works across the whole book, not one case. It groups every complaint the firm knows of by root cause and product: confirmed cases with tags, complaints in the inbox that the classifier tagged but nobody has opened a case for, and the firm’s prior complaints handled before InsightService. A group is flagged as possibly systemic when several complainants share a fee run, or when a cause reaches at least twice the systemic threshold of customers in the last 30 days and at least double the 30 days before. The threshold is SERVICE_SYSTEMIC_CUSTOMERS, five by default: illustrative policy, set to your own.

The clusterWhat the analyst recorded
Root cause and productFee charged in error · Everyday Current Account
Complaints111 complaints from 105 customers; 72 in the last 90 days
The trend39 customers in the last 30 days, up from 17 in the 30 before
The shared run4 complainants share fee batch BATCH-2026-10-03-OD, which charged 104 customers (£624.00)
Where decided71% upheld or partially upheld
SignalPossibly systemic

Both signals fire on this cluster: a shared fee run and a rising cause. Of the 119 clusters in the demo, two are flagged. The other is process failure on home insurance, which is rising (12 customers in the last 30 days, up from 3) without a fee run behind it.

The Root causes page: 1,368 complaints analysed (20 on cases, 1,166 in the inbox, 182 prior), 2 possibly systemic, complaints by root cause and source, the six biggest causes by month, and the cluster table with Fee charged in error on the Everyday Current Account flagged possibly systemic because of fee run BATCH-2026-10-03-OD
Root causes. The fee-run cluster at the top of the table, flagged possibly systemic, with “Take ownership” beside it. The thresholds are labelled illustrative.

yes

no

yes

no

Complaints: cases, inbox, prior

Group by root cause and product

Shared fee run with 2+ complainants?

Possibly systemic

Rising: last 30 days at least 2x threshold and 2x the month before?

Cluster tracked, not flagged

A person takes ownership

How a cluster is flagged. Both thresholds come from configuration.

What happens to each complaint

The root cause does not replace the individual complaints. Each one is still a complaint on its own clock, owed an acknowledgement, a decision and a final response. Two of the four have been taken all the way through in the demo; two are still open.

Leo Somerton, SVC-2026-000008: closed, upheldWhat the agents prepared, and what a person did
1
Triage and researchagents

The triage agent quoted the dissatisfied sentence. The case researcher found one fee in question, £6.00 on 3 October.

2
Root cause taggedservice agent

Accepted the researcher’s proposal: fee charged in error.

3
Remediation worked outremediation calculator

£6.00 refund plus £0.06 interest at the illustrative 8% simple rate: £6.06. Below the £250 threshold, so no second approver is needed.

4
Approvedservice agent

Approved with the note “Fee from the faulty overdraft run”.

5
Acknowledgement sent, lateservice agent

The reply drafter’s acknowledgement, accepted and sent on 21 November, against an illustrative due date of 25 October. The clock records it as missed, and QA sampling would fail that check. In the demo nothing is delivered: SERVICE_COMMS_LIVE is off.

6
Outcome recordedservice agent

Upheld: “The overdraft fee was charged in error by a faulty fee run.”

7
Final response sent and case closedservice agent

The final response letter, with the refund and the template referral-rights wording, sent well inside the illustrative eight-week deadline of 15 December. No control blocked the close.

Eilidh Penhale’s case followed the same path, including a late acknowledgement; the QA sample later failed her case on exactly that check. Muhammad Ellery’s and Ama Zhao’s are still under investigation, and both show the same prepared decisions: tag the root cause as fee charged in error, put a £6.06 remediation on the case, use the drafted acknowledgement, and uphold. On both, the acknowledgement is overdue against the illustrative five-day policy: 38 days late for Muhammad Ellery, 31 for Ama Zhao. That is visible on the case header and on the case queue, and the case controls will not let either close without an outcome, a root cause and a sent final response.

The case queue in InsightService: open complaints and queries with the customer, what it is about, stage, the complaint clock status such as Final response overdue or Acknowledgement overdue, the final response due date and the owner
The queue. Each complaint about the run is still a case with its own clock; overdue acknowledgements show in red.

What the people who own the problem do

“Possibly systemic” is a prompt for a person, not a conclusion. The root-cause analyst prepares a single decision on the Root causes page: take ownership of the flagged clusters. A person who accepts it becomes the owner on the record, with a note. From there, the work is outside the complaint queue:

  • Confirm the fault. The run is evidence that something went wrong with fees charged on 3 October. Whether every one of the 104 fees was wrong is for the team that runs the fee engine to establish.
  • Decide on proactive remediation. Refunding customers who didn’t complain is a business decision with its own governance. InsightService shows the reach of the run; it does not refund anyone, and nothing is paid from it.
  • Fix the cause. The fee-run defect itself, so the next run doesn’t repeat it.
  • Keep handling the complaints. Complaints that arrive about the same run keep being handled individually, with the researcher linking each one to the run.

For the head of complaints

Four complaints about £6 are a reporting line. One fee run that reached 104 customers is a conversation with the business owner and, depending on the outcome, possibly your regulator. The cluster gives you the evidence in one place.

For the service operations lead

Your handlers don’t need to spot the pattern. The research links each new complaint to the run, and the prepared remediation and drafts are the same for every case, so four handlers reach four consistent answers.

For the board

The board should hear about the run, not the four complaints: how many customers it reached, what it cost them, who owns the fix, and whether customers who didn’t complain are being put right.

What this shows

Complaints are the cheapest early warning a firm has, because customers do the detecting. The cost of using them is joining up the evidence: the fee and its run from the firm’s records, the complaints across channels, the trend over months. In InsightService that join is deterministic. The same data gives the same clusters every time, every count is a real complaint, and the flag is raised for a person to own rather than acted on by a machine.

Keep reading