Complaint management for lenders, explained.

A lender's complaint program in plain words: every grievance written down, answered on time and read for patterns

The life of one complaint, in five steps

1. A borrower tells the lender something went wrongOn a call, by email or letter, at a branch, in a public post, through a state regulator or the CFPB's company portal.
2. It is written down, sorted and given its deadlinesProduct, issue, risk flags, and every legal clock it starts. This is where a notice of error or a fair lending signal is caught, or missed.
3. It is investigated and answeredFrom the account records, in writing where the rule requires it, before the earliest due date.
4. It is read together with the othersGrouped by what happened, compared with what the lender changed, traced to a cause and sized to everyone affected.
5. The cause is fixed and reportedThe owner of the process corrects it for every customer; the trend and the fix go to management and the board.

Steps 2 to 4 are the subject of this guide. The people, lenders and figures in the example are fictional.

No legal definition, three clocks from one letter

Channel, taxonomy, theme and root cause in plain words

Thirteen terms are defined in the one-page summary.

What counts as a consumer complaint at a bank, and the seven doors complaints come through

In shortA complaint is any expression of dissatisfaction about a product or service, resolved on the spot or not. Examiners expect complaints from every channel to be "recorded and categorized" the same way. A program that logs only written, unresolved complaints sees a small and late slice of what its customers are saying.

The CFPB's examination procedures ask whether "consumer complaints and inquiries, regardless of the channel through which they are submitted, are appropriately recorded and categorized", and, where a lender runs several complaint units, whether "it employs a common set of practices".4 The same procedures ask examiners to gather complaints about the lender from the CFPB, the prudential regulator (the agency that supervises the bank's safety and soundness), "state regulators, state attorneys general offices or licensing and registration agencies, and private or other industry sources".4 An examiner can therefore see complaints the lender itself never logged.

A practical definition starts broad: any expression of dissatisfaction, oral or written, about the lender's products, services, staff or vendors, whether or not it was resolved at first contact. Many lenders add inquiries that suggest a problem, and complaints about third parties that act for them: collection firms, payment processors, dealers, subservicers.

A Federal Reserve examiner advised banks that contract with third parties to require "similar tracking mechanisms, with routine reporting to the bank".9 The FDIC found a third-party provider involved in 6,356 of the complaint cases it handled in 2025, up nearly 48% from 4,282 in 2024.10

The seven doors, and who usually owns each

Management watches the portal most closely, yet most of a lender's complaints arrive through other doors. Even inside the CFPB's own process, nearly all complaints come in through its website, with smaller numbers by phone, by referral from another agency and by mail.

A lender's own mix looks different, because its customers mostly call. No public survey measures how lenders' complaints split by channel. The useful measure is your own: count last quarter's complaints by door, including the ones solved on the call, and compare that with what your complaint log holds.

Section 1034, the 2017 exam procedures and Regulations X, Z and E: the law a complaint program answers to

In shortNo single rule creates a complaint program. Section 1034 of the Dodd-Frank Act makes the CFPB collect and answer complaints and requires large banks to respond to regulators. The CFPB's 2017 examination procedures and the interagency rating system say what a good program looks like. Product rules (Regulations X, Z and E) put hard deadlines on some complaints. In 2025 and 2026 the CFPB shrank and withdrew guidance, states added powers, and the statutes did not change.

What the 2025–2026 changes mean for a complaint program

Less federal enforcement has not removed a single deadline. The CFPB's funding cap fell to about $467 million for fiscal 2026, it said it would cut its supervisory "events" by half and focus on banks rather than nonbanks, and it opened one new public enforcement action in the first 18 months of the new administration.20

Its November 2025 pledge says exam findings "will focus on pattern and practice violations of law where there is tangible and identifiable consumer harm".21 A pattern of harm is exactly what complaint data shows first, so the complaint log has become the place an examiner starts.

The May 2025 withdrawal removed guidance and left the law in place. It withdrew the CFPB's 2023 policy statement on abusive practices and its old policy statements on publishing complaint data.15 The public complaint database is still published; this guide queried it on October 31, 2026.11 Section 1034, the UDAAP statute and every Regulation X, Z and E deadline are unchanged.12,14

States have moved into the space. New York's FAIR Business Practices Act, signed December 19, 2025 and in effect 60 days later, on February 17, 2026, adds unfair and abusive acts to the state's consumer protection law; only the Attorney General enforces those two.19

CFPB 15 and 60 days, notices of error, billing errors and state dates: which clocks a complaint starts

In shortSet the clocks from what the message says, whatever door it came through. A portal complaint runs on calendar days. A written claim of a mortgage servicing error is a Regulation X notice of error, counted in business days without weekends and federal holidays. A written dispute of a card or credit-line charge can be a Regulation Z billing error, counted in billing cycles. Each clock is met on its own.

A notice of error needs no special words. Regulation X applies to "any written notice from the borrower that asserts an error and that includes the name of the borrower, information that enables the servicer to identify the borrower's mortgage loan account, and the error the borrower believes has occurred".1 An angry email that says the escrow analysis is wrong, signed and with the loan number, qualifies.

A phone call does not start the Regulation X clock, but it is still a complaint to log; under Regulation E an oral notice does start the investigation clock.17 The detailed mechanics of each of these desks are in the companion guides on RESPA error resolution and card disputes.

Counting a business-day deadline across two holidays

Take a written claim of an escrow error that reaches a servicer by email on Monday, October 12, 2026. That day is Columbus Day, a legal public holiday.24 Counting starts the next day: Tuesday, October 13 is day 1, and day 5 is Monday, October 19, the acknowledgment deadline.

The thirtieth business day, skipping Veterans Day on Wednesday, November 11, is Tuesday, November 24. Counted in calendar days, thirty days would end on November 11. A tracker that uses one calendar for every clock is wrong on at least one of them.

No date replaces another. The earliest due date sets the order of work, and each obligation is met on its own terms. Count a complaint once in volume and once per clock in timeliness, or a missed acknowledgment disappears inside a portal answer that went out on time.

A complaint taxonomy by product, issue and risk, so that legal problems are escalated the day they arrive

In shortA complaint taxonomy is a fixed list every complaint is coded against: the product, what happened, the legal risks it raises, and later the cause and the finding. Examiners expect complaints that raise unfair treatment, discrimination, unauthorized enrollment, imminent foreclosure or other legal issues to be "categorized and escalated". Coding inconsistently is the quiet failure: the same problem lands in different categories and no count ever rises.

The CFPB lists the complaints examiners expect a lender to escalate: those that "raise legal issues involving potential consumer harm from unfair treatment or discrimination, unauthorized product enrollment, account openings or upgrades (including the addition of ancillary products), improper sales practices, imminent foreclosures, or other regulatory compliance issues".4

A Federal Reserve article suggests tracking by "business line, legal entity, product type or service, complaint reason, law or regulation, and the disposition", so trends and high-risk issues can be analyzed.25 The FDIC codes each complaint it handles "with at least one product, issue, regulation, and finding".10 The finding, coded after the investigation, is what turns a count into evidence.

Five fields every complaint should carry

Which words in a complaint point to which legal risk

Two rules keep coding honest. Flag generously and clear carefully: a flag costs a reviewer minutes, a missed pattern costs a finding. And let only the person who owns a risk clear its flag. A fair lending flag, for example, is cleared by the fair lending officer after a review, never by the analyst who answered the borrower.

Consistency is the harder problem. When the contact center codes a call about a higher payment as a "payment inquiry", the servicing desk codes the same problem as "escrow dispute" and regulatory relations codes the portal version as "trouble during payment process", three small counts hide one large one. A taxonomy only works if every channel uses the same one, which is what the CFPB means by "a common set of practices".4

How a systemic issue hides across channels, and the three habits that surface it in days

In shortSystemic problems arrive through every door at once, in different words, so each queue sees a handful of something ordinary. Three habits surface them: group complaints by the event they describe, compare each group with its own normal level, and keep a calendar of what the lender changed beside the complaint data. Most systemic errors start with a change the lender made.

The CFPB told the industry in 2013 that it had seen complaint response "handled ... with the business line that received the complaint addressing the issue without any type of centralized method for tracking issues across the entity", which increases the risk that a problem found in one line "may not be identified or addressed in a similar manner within another".26 Thirteen years later that remains the most common shape of a late pattern.

In the example at the top of this guide, one transfer error produced forty complaints in four days. Split by door, no queue held more than 23 of them, and each team saw its own slice as normal for a week after escrow statements go out.

Three habits that surface a pattern

Template campaigns need the same discipline in reverse. The CFPB says its 2025 volumes came partly from third parties such as "credit repair organizations" and from "an increase in the number of complaints submitted in a single session".6 Grouping by the facts asserted keeps a campaign that sends the same words about unrelated accounts from looking like a crisis, while a real error described in a campaign's words still joins the borrowers who described it in their own.

Measure the time to see a pattern honestly. Start the clock at the first complaint later assigned to the group, from any channel, including calls, and stop it when the group is flagged and handed to a named owner. Count calendar days, because a statement run or a fee cycle does not stop at weekends.

Complaint root cause analysis: from the symptom borrowers saw to every account the error touched

In shortThe symptom is what customers felt, such as a higher payment. The root cause is the step in the lender's process that let it happen. Examiners ask for "comprehensive root cause analysis" and "retrospective remediation to all consumers harmed". The complainants point to the problem; the affected population, found in the lender's own records, sets the scope of the fix.

The CFPB procedures give two examples of what root cause analysis should find: "whether an information system coding error led to inaccuracies across a portfolio, or whether repeated data entry errors, in the absence of appropriate monitoring, resulted in significant inaccuracies". They then ask whether root cause findings are used in corrective action, "such as modifying policies, procedures, training, monitoring, and/or other appropriate business adjustments".4

Regulation X adds a duty inside each investigation of a notice of error: if the servicer finds errors other than, or in addition to, the one the borrower raised, it must correct those too.1

Four steps from a group of complaints to a fix

1. Pull the records behind the complaints, beyond what the borrowers wrote: account history, statements, the data that produced them, and any file received from another company.
2. Find what the complaints share that other accounts do not: a date, a branch, a system, a county, a transfer, a product version.
3. Compare the records that disagree, and test the obvious alternative. A rise in the county tax rate would also raise a payment; the arithmetic shows whether it explains the size of the rise.
4. Search the whole book for the same pattern and size it against the next operational deadline: the next statement run, disbursement or fee cycle.

In the example, the two things the 40 complaints shared, the September transfer and the two counties that discount early tax payments, pointed to the transfer file. The old servicer had paid the second-half tax early to take the discount; the transfer file still listed it as due on November 1; the new servicer's escrow analysis projected it a second time.

A rate rise was ruled out: the county's 2026 rate rose 1.8%, about $107 a year on a tax bill like Gloria's, against a shortage of $1,486.20. The same pattern sat on 1,140 loans. Fixing forty accounts would have left 1,100 borrowers paying a shortage they did not owe, and the November 1 tax run would have paid $3.39 million of tax again.

The rules around a transfer explain why this kind of error matters to an examiner. For 60 days after a transfer, a payment the old servicer receives by its due date "may not be treated as late for any purpose", and the escrow account rules govern how a shortage may be collected.27 A failure to transfer information accurately is itself one of the covered errors under the notice-of-error rule.1

Reading complaints for unfair practices and fair lending risk after the 2025 changes

In shortComplaints are the earliest evidence of an unfair, deceptive or abusive practice and of different treatment on a prohibited basis. The FDIC and the OCC now examine fair lending through disparate treatment, which is what a borrower's account of her own treatment describes. Few of these complaints use the legal words, so flag on the description and decide on review.

Unfair, deceptive or abusive

The statute defines an unfair practice as one that causes or is likely to cause substantial injury that consumers cannot reasonably avoid and that is not outweighed by benefits to consumers or competition. An abusive practice materially interferes with a consumer's ability to understand a term, or takes unreasonable advantage of a lack of understanding, an inability to protect one's interests, or reasonable reliance on the provider.14

The CFPB withdrew its 2023 policy statement on abusiveness in May 2025; the statute itself is unchanged.15 Federal Reserve examiners cited section 5 of the FTC Act, the bank version of unfair or deceptive practices, 16 times in 2025, the fourth most-cited provision.28

Complaints point to UDAAP risk when many customers describe the same surprise: a fee they did not expect, a promise a script made that the product did not keep, a charge for something they did not buy. The FDIC's complaint unit lists "account opened without knowledge" among its top issues for 2025, at 3% of complaints and inquiries about the banks it supervises.10

Fair lending: why complaints matter more since 2025

In August 2025 the FDIC updated its examination manual to assess discrimination under ECOA and the Fair Housing Act "only through evidence of disparate treatment".7 The OCC told its examiners in July 2025 that they should no longer examine for disparate impact, where a neutral policy falls harder on a protected group.8

The CFPB's Regulation B rule, effective July 21, 2026, states that ECOA does not authorize disparate-impact liability, and it narrows discouragement to statements a creditor knows or should know would lead a reasonable person to expect a denial or worse terms because of a prohibited basis; a challenge to the rule is pending.18 The statutes themselves did not change: discrimination on a prohibited basis in any aspect of a credit transaction is still unlawful, and so is a statement that discourages an applicant in that way.18

What changed is the kind of evidence that carries a case. A statistical gap with no story is a weaker lead; a borrower saying she was asked for something other applicants were not is exactly the story. And discouragement leaves no other trace: no application, no adverse action notice, nothing in mortgage reporting data. The complaint may be the only record that exists.

Fair lending complaints are rare, which is why counts never find them. The FDIC's complaint unit recorded 39 fair lending complaints in 2025, down from 62 in 2024, among 28,489 written complaints closed.10 In the example's bank, six complaints over six weeks from two branches said applicants were told to bring a co-signer after asking for service in Spanish. The comparison that tests them is a file review: how often were similar applicants at the same branches asked for a co-signer?

A gap like that is a reason to look further. The fair lending officer reads the files for credit reasons that justify each request, interviews the branch managers, and then decides. Whatever the answer, the complaints, the comparison and the reasoning sit together, which is what an examiner asks for.

Where complaint programs break: a notice read as ordinary mail, 157,000 late acknowledgments and queues split by channel

In shortThe expensive failures are classification failures: a letter that was a notice of error treated as ordinary mail, a dispute that was a billing error left without an acknowledgment. The cheaper but common failures are structural: queues by channel, narrow definitions, root cause that stops at the symptom, and boards that see only counts.

Two orders that turned on recognizing the complaint

Error resolution among the rules Fed examiners cite most

Error resolution is a standing exam finding. In 2025 Federal Reserve examiners cited 630 consumer compliance violations in 97 examinations of state member banks. Regulation E's investigation and written-answer requirements, §1005.11(c) and (d), accounted for 48 of them; section 5 of the FTC Act for 16.28

Seven complaint handling failures, why they happen and what catches them

The private exposure is real too. A borrower can sue a servicer under RESPA section 6 for actual damages and, for a pattern or practice of noncompliance, additional damages of up to $2,000; in a class action the additional damages can reach the lesser of $1,000,000 or 1% of the servicer's net worth.32

How many lending complaints reach the CFPB, how companies answer them, and what changed in 2026

In shortThe CFPB received more than 6.6 million complaints in 2025, and 88% were about credit reporting. Lending products are a small slice: about 114,100 about credit cards, 30,400 about mortgages and 28,500 about vehicle loans. In the first nine months of 2026, published mortgage complaints were up by half on 2025, and the share answered late rose.

That share is why national totals say little about a lender. The CFPB itself says the growth is "driven almost entirely by a dramatic growth in credit or consumer reporting complaints", and links the increase to third parties, to more complaints "submitted in a single session" and to new AI tools that file on a person's behalf.6 Lenders still feel it: they furnish (report to the credit bureaus) the account data those complaints dispute, and many receive the same templates by mail and email.

How companies close lending complaints

Mortgage complaints split mostly between payment problems and hardship. Of 24,697 mortgage complaints published for 2025, 12,680 were about "trouble during payment process" and 5,980 about "struggling to pay mortgage".11 The CFPB's 2025 report says escrow problems such as late fees from taxes or insurance paid late "appeared to be especially frequent after servicing transfers", that companies often attributed late payments to "third-party vendors, loan boarding, or prior servicers", and that some "admitted to escrow analysis calculation errors".6

Mortgage complaints up by half in 2026, and three times the share answered late

The share marked as not answered on time rose with the volume. For mortgage complaints received in the first nine months, it went from 1.4% in 2025 (245 of 17,740) to 4.3% in 2026 (1,150 of 26,549); for credit cards from 0.3% to 1.4%.11 Student loans are a separate story: in 2025 companies gave no timely response to 31% of student loan complaints.6

Regulators see complaints the CFPB never sends. The FDIC's consumer response unit closed 32,128 written complaints and call records in 2025, 21% more than in 2024; it acknowledged every written complaint within 14 days, investigated 15,405, and found 280 errors by banks and 108 violations of federal consumer rules.10 Credit cards were its top product, at 5,783 complaints, and credit reporting its top issue, at 35%.10

What no one publishes: the cost of handling a complaint at a lender, how many complaints an analyst can handle, the share of lender complaints that are notices of error, or how long lenders take to spot a systemic issue. Part C shows how to measure each of your own in a week.

Complaint software and AI read, code, clock and group; people decide what counts, what caused it and what the lender promises

In shortSoftware is good at the parts that are reading, matching and counting: capturing every channel, coding to a taxonomy, setting clocks from rules and calendars, grouping complaints, drafting answers from records. People should keep the decisions that commit the lender: whether a message is a complaint at all, whether a fair lending flag is cleared, which cause is accepted, what remediation is promised, and what is sent. The April 2026 model guidance leaves generative AI to each bank's own governance.

SR 26-2 and the CFPB's chatbot warning, applied to complaint tools

On April 17, 2026 the Federal Reserve, OCC and FDIC replaced the 2011 model risk guidance (SR 11-7) with SR 26-2, OCC Bulletin 2026-13 and FDIC FIL-15-2026. Its footnote 3 calls generative AI models "novel and rapidly evolving" and places them outside the guidance's scope. It adds that a bank's "risk management and governance practices should guide the determination of appropriate governance and controls" for tools the guidance does not cover, and that its principles still apply to traditional statistical models and non-generative machine learning.22

A language model that reads and codes complaints therefore falls to the bank's own AI policy and third-party risk program; the regulators give no checklist for it.

The CFPB's 2023 spotlight on chatbots, still posted, makes the point that matters most for complaints: automated tools "must comply with all applicable federal consumer financial laws", and a script that recognizes a dispute only from "specific words or syntax" may "fail to recognize that a consumer is invoking their federal rights".31 A notice of error does not need special words, so neither should the tool that recognizes one.

Can a lender tell an AI-written complaint from a person's?

Not reliably, and it should not try to act on a guess. NIST's 2024 review of synthetic content reports that several studies found text detectors "tend to perform little better than chance when given short texts", reaching about 90% on its accuracy measure only with samples of around 750 words, and it cites research titled "GPT detectors are biased against non-native English writers".33

Most complaints are short. The useful signals are facts the lender can check: the same narrative filed for unrelated accounts, a batch from one session, an account that does not exist, an error already answered with nothing new. No rule lets a lender skip a portal answer or a notice-of-error clock because software may have written the complaint.3,1

Five controls to ask of any tool

Complaint reporting for the board: what was found, who owns it, who was put right and whether the fix held

In shortComplaint counts rise when a lender does something wrong and when it makes complaining easier, so a count alone cannot tell good news from bad. Examiners ask whether complaint evaluations reach "the board and senior management" and whether management "takes appropriate prospective and retrospective corrective action". Report in four families: timeliness per clock, time to see and own a pattern, reach of the fix, and the integrity of the data.

The Federal Reserve's guidance on complaint data says complaint-derived risks "should be factored into the overall compliance assessment provided to the board", and describes an institution whose rating was downgraded after a UDAP issue visible in complaints was left out of its board report.25 The interagency rating describes a strong program as one where management "monitors consumer complaints to identify risks of potential consumer harm, program deficiencies, and customer service issues and takes appropriate action".5

One line per accepted issue is enough for a board: the cause in a few words, the owner, the accounts affected, the share remediated and the date the fix is due. A cause that is only proposed is reported as proposed, with its owner, so the board sees what is decided and what is not.

Who handles a complaint at a lender today, how often it changes hands, and the month lost before anyone sees the pattern

In shortComplaints arrive through several teams, each with a queue. Analysts read and key each one, pick a category and a deadline, and start a letter. Patterns are usually found at a monthly or quarterly trend review. Most of the elapsed time sits between complaints arriving in different queues and someone noticing they share a cause.

Analyst, complaints manager, fair lending officer, owner, CCO

Four hand-offs before anyone looks for a pattern

About seven weeks by hand, most of it waiting to see the shared cause

A month's wait matters because operational calendars do not wait. Statements keep going out, disbursements run, fees post. In the example, a month's delay would have carried the investigation past the October 27 file lock and into a $3.39 million duplicate tax payment.

A shared log, a case management module or complaint software

The escrow transfer error on screen: from an overnight spike to an accepted issue and Gloria Ferraro's answer

In shortThe example from the top of the page, inside the product. Overnight, six agents read 63 new complaints from every channel, joined 21 of them to a theme that doubled from 19 to 40, traced it to one line in the transfer file and found 1,140 loans with the same false shortage. Joelle Marchetti, the complaints management lead, sends it to the head of mortgage servicing as one issue; once he accepts it, she approves Gloria Ferraro's answer eleven days before it is due.

1

Ninety days of complaints, with what the bank changed above them

Tuesday, October 13, 2026. Joelle opens Themes: 1,286 complaints from July 15 to October 12, from seven channels, grouped into 34 themes. Each theme is a strip of daily bars colored by channel. Above the strips runs a band of the bank's own events: a mobile app release on July 21, HELOC rate notices on August 14, the Brightwater transfer on September 1, the escrow statements on October 8.

The overnight scan ran from 6:00 to 6:19 am. One row is flagged: "Escrow shortage after transfer", 19 to 40 overnight, likely systemic across three channels.

2

The theme, its growth and a proposed cause with both records side by side

The escrow theme holds 23 calls, 11 emails and 6 CFPB complaints; the first arrived October 9. There were no escrow complaints for 37 days after the transfer, until the statements went out. The proposed cause: the transfer file listed Gloria Ferraro's second-half 2026 tax as "Due Nov 1, 2026, $2,972.40", while the Marlow County record shows it paid on August 27 at $2,912.95 with the 2% early discount.

A tax-rate rise is ruled out. The screen says Marlow's 2026 rate rose 1.8% and puts that at about $38 a year; on a tax bill like Gloria's, two installments of about $2,972, a 1.8% rise is about $107 a year, still a small fraction of a $1,486.20 shortage. The 92% beside the cause is the solution's own confidence on sample data; the cause stays "proposed" until its owner accepts it.

3

Every affected borrower, well beyond the forty who complained

The root-cause investigator searched all 6,214 transferred loans for a paid installment listed as due. It found 1,140: 812 in Marlow County and 328 in Ashby County, $1.59 million of shortage billed, $1,391 a loan on average. Fourteen borrowers had already paid theirs in one sum, $19,870 to refund. The next tax run, on November 1, would pay $3.39 million of tax a second time; its file locks on October 27.

4

Gloria Ferraro's complaint, in her words, with every fact cited

She filed through the CFPB portal on Friday, October 9 at 7:42 pm. Her words are on the left with numbered notes: the shortage of $1,486.20, the payment change from $1,842.17 to $1,966.02, the county receipt, what she wants. The middle column shows what the agents found, each fact with its source. The clock keeper counts from the day she filed, the earliest date the portal clock could start: first response due October 24, final response December 8.

The drafted response admits the error and makes three commitments. They are marked in yellow: they depend on an issue the servicing head has not accepted yet.

5

One issue to the head of servicing, then the answer goes out

Joelle sends the theme to Tom Whitfield, head of mortgage servicing, as ISS-2026-031: 40 complaints linked, the evidence from the transfer file, county tax records and escrow analyses, and five remediation steps with owners and dates. The first is his: stop the November 1 disbursement for the 1,140 loans by October 27. He accepts.

Had she tried to approve Gloria's response first, the solution would have stopped her: the response promises what the owner had not yet agreed. Now the commitments are green. She approves, and the response goes out through the CFPB portal, eleven days before the due date.

The solution's design targets are 14 minutes from receipt to a classified, routed complaint with its clock, and a systemic error like this one flagged by day 4. They are targets, to be measured on your own complaints.

Nine more screens, from a notice of error counted in business days to the October report for the board committee

In shortOne theme is the story. These screens show the rest of the program: the themes one event started, a written claim of error with its business-day clock, a new email routed on arrival, the issues with their owners, a small theme with a fair lending flag, every due date on a calendar, the board report and the lender's own rules.

Select any screen to open it larger. The bank, borrowers, people, counties and figures are fictional.

Six agents read, code, group and draft; the complaints lead, the fair lending officer and the issue owner decide

In shortEach agent has one job and limits it does not cross. None of them sends a response, clears a fair lending flag, changes an account, holds a payment or contacts a borrower. The complaints management lead approves and sends every response; only the fair lending officer clears a fair lending flag; the owner of each process accepts its issue and signs off the fix.

The six complaint agents, and the line each one never crosses

Who approves responses, who clears fair lending flags and who accepts an issue

Every read, classification, flag, draft, edit, approval and send is kept on the activity record with who, what and when, agents and people alike. The agents read loan, escrow, payment and transfer records; they never change them. Complaint data stays in the bank's own workspace.

Kestrel Point Bank's complaint team in the screens

What the complaint software reads from seven channels and the servicing system, and the coded cases, issues and board report it produces

In shortThe solution reads the channels the bank already has and the account records behind them, read-only, and leaves behind one list of coded complaints with their clocks, themes with likely causes, issues with owners and plans, cited drafts waiting for approval, and the monthly trend report. The bank's systems of record stay where they are.

Eight systems a complaint program connects, from the CFPB portal to issue tracking

For each, check whether it has an API, a file feed, or only screens, and who owns it.

Your complaint log, taxonomy, templates and board report stay yours

Cited findings, no unmade promises, people decide

What a pattern found at the quarterly review costs you in refunds, repeat contacts and examiner attention

In shortThe cost of a complaint program is mostly hidden in two places: the hours analysts spend reading and keying, and the harm that keeps spreading between the first complaint and the day someone sees the pattern. Nobody publishes the cost of a complaint at a lender. The enforcement record shows the ceiling when a clock is never started, and your own last systemic issue shows the rest.

Unlogged complaints and late patterns: finding your figures

Silent borrowers, the public record and a state inquiry

Analyst hours released, and harm stopped before the next statement run, when every complaint is read the day it arrives

In shortTwo levers carry most of the value: the analyst minutes released when complaints arrive already read, coded, clocked and drafted, and the harm avoided when a systemic error is seen in days. The calculator asks for your own numbers; every default marked illustrative is an assumption to replace.

Analyst minutes, earlier patterns and deadlines counted right

Your complaint volume, analyst hours and harm avoided

At the defaults, 5,000 complaints a year and 30 minutes saved release 2,500 analyst hours, about $150,000 at $60 an hour; four issues of 300 accounts at $150 each, half avoided, add $90,000. A lender with a few hundred complaints a year and no cross-channel issues probably gains little, and should know that before buying anything.

Complaint program readiness: a written definition, one taxonomy, clocks as rules, record access and named deciders

In shortPick the answer closest to your program today. Eight questions score 0 to 2; the last is for reflection. The meter adds up your score as you go, and each question says why it matters and what a strong answer sounds like.

What your complaint program score means

Keyword lists, one deadline for every case and word clouds: warning signs in a complaint software vendor's answers

In shortAsk every vendor the same questions, us included. The answers show whether the tool recognizes legal events from the words borrowers actually use, counts each clock correctly, groups complaints by what happened, shows its evidence, and leaves the decisions with your people.

We found no published independent test of any complaint software's accuracy, as of October 2026. A blind re-read of your own past complaints is the fairest comparison, and the same test measures your current process.

A complaint software pilot decided on four numbers: recognition by flag, clocks correct, days to a pattern, analyst minutes

In shortMeasure first, then test on your own past complaints, where you already know the answers, then run one live month with your people approving every reply. Decide on four numbers. None of these steps commits you to a vendor.

1. Take your baseline: minutes per complaint by step, timeliness per clock, and the dates of your last systemic issue.
2. Write down the rules: the definition, taxonomy, clock table, approvers and fair lending routing.
3. Gather last quarter's complaints from every channel, with the codes, clocks and themes your team assigned and any issue you found.
4. Re-read them blind and compare: codes, notices of error recognized, fair lending flags, themes and the day each would have been flagged.
5. Run one live month in parallel, then decide with four numbers: recognition by flag, clocks correct, days to see a pattern, analyst minutes.

How to count each pilot number, and why it matters

From the CFPB's 15-day deadline to AI-written complaints: thirteen complaint handling questions answered

In shortThe questions complaint teams, compliance officers and buyers search for most, each answered on its own.