First, tag a month of complaints by root cause and separate failures from broken promises
Start with the complaints themselves, not the count. Take a month of them from every channel, including reviews, social replies and anything staff wrote on paperwork, and tag each one twice, the way the tutorial on tagging support contacts by reason separates the two: the reasonThe main reason a customer contacted support, tagged by the agent from a fixed, defined list so volume and other service metrics can be read by cause. the customer gave, in their own terms, and the root cause behind it. "Late delivery" is a reason; the cause might be a courier, a warehouse cut-off, stock shown as available when it was not, or a delivery promise at checkout that was never achievable. The five whys method carries each of the largest reasons back to a cause: the sofa was late because the fabric was made to order; the customer did not know because the product page promised a fast delivery for every sofa; the page said that because nobody told the web team which fabrics were made to order; nobody told them because no step in the process asks. The chain stops at a process the team can change, not at a person.
Then sort the causes into two groups, because they have different owners. A failure is something the business did wrong: the parcel was lost, the appointment was missed, the bill was wrong. A broken promise is something the business did as designed while the customer expected something else, because an advert, a product page, a confirmation email or a sales call led them to. The service recovery and complaints guide records that in many businesses a large share of complaints start from a promise, and those are marketing's to fix, while a failure belongs to whoever runs the step that failed, often outside marketing. A month tagged this way usually shows one cause far ahead of the rest, and the table below is read from that cause.
Three further readings sharpen the diagnosis. Count complaints against a base such as orders or active customers, so a busy month is not mistaken for a worse one. Count repeat contactsA customer contacting the business again about the same problem, the clearest sign that the first answer did not end the customer's work., the customers who complained twice about the same thing, because a rising share of them points at the handling rather than the failure. And read how long complaints waited for a first reply and for a decision, because a complaint that waits grows. Where the rise is a burst of reviews on a local listing within days, the diagnosis for a sudden wave of negative Google reviews first sorts real customers' complaints from fake reviews and from a story that is spreading; where the rise arrives as returns rather than complaints, the diagnosis for a rising return rate starts from the returns. This page starts from the complaint log.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Complaints cluster at one step of the journey, and customers quote the advert, the product page or the confirmation email | The wording on every channel for that step against what the step actually delivers, judged from complaints, returns and reviews | A marketing promise the experience does not keep: a delivery time, a feature implied, a price that excluded something | Fix the step or change the claim; correct or withdraw a broken objective claim everywhere it runs |
| Complainants say nobody acknowledged them, and failures that could have been contained arrive as formal complaints | Time from first contact to acknowledgement, and whether each step of the recovery sequence appears in the record | No recovery sequence, so a failure that could have been recovered becomes a complaint | Run the sequence on every failure: acknowledge quickly, apologise, understand, fix, make amends, follow up, learn |
| Complaints escalate while the customer waits for someone to approve a refund or replacement | How many cases waited for approval, how long they waited and how many were approved anyway | Front-line staff with no written remedies or limits, so every case waits for permission | A remedy table with a standard remedy per failure, a goodwill allowance and approval levels with a fast route |
| Nobody can say how many complaints arrived last month, and the same customer complains twice | Whether every channel feeds one log, and the share of complaints logged only after the customer chased | No procedure, so complaints are handled differently each time and some are never logged | A written procedure: a definition, one log for every channel, response promises, escalation stages and closure with a cause |
| Reviews quote replies that sound scripted or contradict each other, and repeat contacts follow a first reply | A sample of complaint replies read against the tone guidelines and a short scorecard | Replies that vary by agent, with no tone standard and no quality review | Tone guidelines for complaints and a weekly quality review that reads every complaint reply |
| The same failure returns month after month at the same rate, filed under different reasons | Whether closed complaints carry a root cause separate from the reason, and whether the monthly review reaches the owners | Complaints closed without a root cause, so the same failure recurs at the same rate | Record the reason at close, find the root cause for the largest reasons with the five whys, and review monthly with owners |
Fix, verify and prevent each cause
A promise the experience does not keep
The tutorial on checking marketing promises against the experience starts from the point this cause depends on: customers judge the experience against what they were promised, and marketing writes most of the promises, some stated outright and many only implied by a photograph, a tone of voice or a word like "easy". A delivery time that holds only before a cut-off, a feature implied by an image, a price that did not include something the customer assumed it did: each produces complaints from customers who were served exactly as designed.
Fix: run the promise check for the step the complaints cluster at. Collect the promise as it appears on every channel, including the inferred ones, find the moment the customer tests it, judge it from the complaints, returns and reviews as kept, partly kept or broken, and then fix the step or change the claim. The tutorial's rule for objective claims is strict: a stated time, price, result or feature that is broken is corrected or withdrawn in every place it runs, through the business's approvals process, not reworded in a softer tone. The recovery guide adds the reference point: where a customer complains after relying on a marketing claim, whether that claim was acceptable is judged against the Advertising Standards Authority's codes, and the recovery often has to correct the claim as well as solve the customer's problem. That is general direction, not legal advice.
Verify: complaints tagged to that step fall in the following months, and the sales script, the confirmation email and the agents' saved replies no longer make the promise the advert stopped making.
Prevent: keep a promise register with an owner and a date for each decision, and add a short promise check to every campaign plan before launch, because campaigns change promises faster than operations change delivery.
No recovery sequence
The recovery guide's sequence is acknowledge, apologise, understand, fix, make amends, follow up and learn, and its first observation is that customers judge what follows a failure far more harshly than the failure itself. Most people accept that things go wrong; few forgive being ignored, disbelieved or passed between departments. A business without the sequence turns ordinary failures into complaints and complaints into public ones, because speed runs through every step and a failure that waits grows in the customer's mind.
Fix: write the sequence into the handling of every failure, not only the ones already labelled complaints: a same-day acknowledgement that names the specific problem, an apology that takes responsibility without excuses, a check of the order history before the customer is asked to repeat it, the fix itself, amends proportionate to the real cost to the customer, and a follow-up that confirms the fix worked. Keep amends proportionate, because over-compensating a minor failure teaches customers that complaining pays and under-compensating a serious one reads as indifference. For complaints that start in public, the community management guide applies the same sequence in the open: acknowledge in public quickly, resolve in private, and note the resolution publicly where the person agrees; never argue, and never offer anything to have a review removed, which the recovery guide records the Competition and Markets Authority treats as unacceptable, alongside customers' rights to goods and services that match what was promised.
Verify: the time to first acknowledgement falls for every channel, and the share of failures that become formal complaints or public reviews falls with it.
Prevent: make the sequence the structure of the complaint record, so a case cannot be closed with a step missing, and read a sample of recovered failures each week to see that the follow-up happened.
Front-line staff with no written remedies or limits
The recovery guide names the three things staff need to act: defined remedies for common failures, discretion within a limit, and a fast escalation route for anything beyond it. Staff who must ask before every refund or replacement keep the customer waiting and pass them along, which tells them nobody is in charge, and that wait is where a complaint escalates. The tutorial on setting remedies and goodwill limits builds the two documents that end it.
Fix: list the failures from the complaint log in order of frequency, separate for each the part the customer may be entitled to from the goodwill the business adds on top, and write a standard remedy specific enough that two agents would offer the same thing. Set a goodwill allowance agents can use without asking, approval levels above it with a named approver and a promised response time, and a rule that safety issues, legal threats and customers in vulnerable circumstances go to a senior person regardless of amount. Have the rights column checked by someone qualified for the sector and country, because the tutorial states no periods or conditions and neither does this page.
Verify: the share of complaints resolved without escalation rises, read beside how satisfied complainants are afterwards so that speed is not bought by giving too much or too little, and the time a case waits for an approval falls to hours rather than days.
Prevent: record every remedy given, standard, discretionary or approved, and review the table each quarter from the cases that did not fit, raising the allowance where most escalations for one failure were approved anyway.
No procedure, so complaints are handled differently each time and some are never logged
The tutorial on writing a complaints procedure describes what happens without one: a complaint taken by phone survives only in the memory of the person who answered, a review goes unseen by support, a delivery form carries a note nobody passes on, and nobody can say how many complaints the business had last month or whether any customer waited too long. A rising count in that condition is partly a count of complaints that were handled late or twice.
Fix: write the procedure. A one-sentence definition of a complaint, with the instruction to treat a contact as one when unsure; every channel opened into a single log, with a named person moving reviews, social messages and field paperwork into it; the fields the log records, from the customer's own words and the outcome they want to the reason, the owner, the dates the promises were met and the root cause; and acknowledgement and response promises set from what the business can keep, which are its own promises rather than legal deadlines, tracked as a KPIA metric tied directly to an objective. someone owns; where another team must act before the reply can go, agree that hand-off time with them as a service level agreementA published promise between teams, or with a supplier, about what each type of request needs and how quickly it will be handled, agreed rather than imposed.. Name who investigates and who decides for each type of complaint, write the escalation stages as a route customers can use rather than a wall, and close each complaint only when the customer has had the full response and the actions are done. Regulated sectors have their own complaint rules and ombudsman schemes, which the tutorial does not cover and neither does this page.
Verify: the log's count for the next month includes complaints from every channel, the share of complaints that were chased before being logged falls, and the acknowledgement and response promises are met for most cases with the exceptions explained.
Prevent: publish the customer version of the procedure where customers and staff can find it, and keep the staff version current by revising it whenever the monthly review shows a step that does not work.
Replies that vary by agent, with no tone standard and no quality review
Response times show how fast complaints were answered, not whether the answer was right or sounded like the brand. The tutorial on reviewing support conversations for quality lists what a reviewer sees that no dashboard records: an answer that was confident and wrong, a reply that was polite and answered a different question, a tone right for a simple query and wrong for an angry customer. When replies to the same failure differ by agent, customers compare notes in reviews, and the second contact is a complaint about the first reply.
Fix: write the tone decisions that should not vary between agents into the customer service tone guidelines, including how the tone shifts for bad news, refusals, delays, apologies and angry customers and the phrases the business never uses, then set up the weekly review: a short scorecard drawn from the business's own standards, a random sample across agents and channels, and every complaint, escalation and low rating on top of it, read as whole conversations rather than last replies. Calibrate reviewers on the same conversations so an agent's score does not depend on who reviewed them, and give feedback with the example in front of the agent.
Verify: the scorecard's tone and correctness lines are met in more of the reviewed complaint replies than before, and repeat contacts after a first reply fall.
Prevent: when the same failure appears across several agents, treat it as a saved reply, a help article or a policy problem rather than an agent problem, send it to the owner of the cause, and revise the scorecard each quarter so it measures the outcome rather than the phrases agents have learned to use.
Complaints closed without a root cause
The recovery guide draws the line between a complaint that is a cost and a complaint that is an investment: the second is tagged by its root cause and feeds a monthly review, the first is closed and recurs. The tagging tutorial explains why the two tags have to stay apart: agents can record the reason reliably at the end of every contact, while the root cause usually needs investigation across many contacts and is worked out for the largest reasons in the review rather than guessed by each agent. A log that records only the symptom tells nobody what to change.
Fix: tag at close with one main reason from a short list in the customer's terms, keep the original reason on repeat contacts so they can be counted, set a rule for "other" with a note the list owner reads monthly, and for each of the largest reasons run the five whys to a cause a team can change, with the chain written down. Count reasons against a base, put an owner beside each of the largest, and take them to the monthly review with a few anonymised quotes in the customer's words, which persuade more than a chart.
Verify: at the next review, the complaints tagged to each cause that was given an owner and an action have fallen, and the agreement audit between agents shows the reasons are being applied the same way.
Prevent: change the reason list only on a fixed cycle with a change log, so the trend survives, and treat a cause that returns after its fix as evidence that the chain stopped one why too early.
Prevention: a promise register, a remedy table and a monthly root-cause review with marketing in the room
Each cause above is a missing document or a missing meeting rather than a bad agent, and the prevention is to keep three things current. The first is the promise register from the promises tutorial: every promise the business makes, stated or inferred, with where it runs, where the customer tests it, the evidence, the judgement and an owner and a date for each decision. It is read on a fixed cycle and whenever a campaign adds a new promise, and whoever repeats a promise, from the sales script to the confirmation email to the saved replies, is told when it changes. A register kept this way turns the largest source of complaints in many businesses into a list someone owns before the complaints arrive.
The second is the remedy table with its authority matrix, sitting beside the complaints procedure. Agents offer the standard remedy without asking, use the goodwill allowance when a case needs more, and reach a named approver with a promised response time for anything larger. Every remedy is recorded against its complaint, and the table is reviewed each quarter from the cases that did not fit. The recovery guide's point about proportion applies to the whole table: consistent remedies protect customers from indifference and the business from the most persistent complainants getting the most.
The third is the monthly root-cause review, which the recovery guide, the procedure tutorial and the tagging tutorial all converge on. The log is counted by reason and by cause against a base, the acknowledgement and response promises are checked, escalations and repeat contacts are read, and a few complaints are read in full in the customers' words. The largest causes go to the people who can remove them, with one agreed change, an owner and a date, and the next review checks whether the complaints tagged to that cause fell. Marketing is in the room because promise problems are marketing's to fix, and because the team that made the promise should hear what customers said when it broke. The weekly quality review feeds the same meeting with the patterns it found across agents, and the service and support hub sets out how support insight of this kind reaches the teams that can act on it.
Watch the public side as part of the same routine. When complaints about one issue start to spread across public channels, the recovery guide treats it as an early sign that the problem is taking over more of what people say about the brand, and suggests counting complaint-driven mentions within Share of VoiceYour share of market visibility. so a service problem is seen when it starts to shape perception rather than after. Keep the public replies to the standard the community management guide sets, acknowledged fast and resolved privately, and keep every public complaint in the same log as the rest. The troubleshooting collection for customer experience lists the neighbouring symptoms that often arrive with this one: support volume that keeps rising usually shares a cause with the complaints, and a churn rate that keeps rising is often where unresolved complaints show up next. Run the three documents and the one meeting for a quarter and the complaint count becomes a figure the business can explain, which is the first sign it is about to fall.
A sofa retailer traces its complaints to one delivery line (illustrative)
Rundleshaw Sofas
Rundleshaw Sofas, an invented online sofa retailer, logs complaints rising from 140 to 236 a month over 6 months, while its review rating on the main platform slips from 4.4 to 3.9. The support lead takes one month, 312 complaints once reviews and social messages are added to the email and phone cases, and tags each by reason and by root cause.
171 concern delivery dates. Every product page says "delivered within 14 days", but 60% of orders are made-to-order fabrics that take 6 to 8 weeks, and the only place the longer lead time appears is a line in the terms. Customers complain when the confirmation email gives no date, and again when the date arrives. Agents can offer nothing for a late sofa without the customer service manager, who takes a median of 2 working days to answer, so 58 of the month's complaints are second complaints from customers chasing a first one that was never logged or answered differently by a second agent. A quality read of 40 complaint replies finds 3 different explanations of the lead time, 2 of them wrong.
The retailer shows each sofa's real lead time on its product page and in the confirmation email, and withdraws the "within 14 days" claim everywhere it runs, including the agents' saved replies. It writes a remedy table for late delivery: a new date with the option to cancel for a full refund, the delivery charge refunded, a goodwill allowance of £50 an agent can use without asking, and a team lead who answers larger requests the same working day. It adopts one complaints log for every channel with a next-working-day acknowledgement promise, reviews 25 complaint replies a week against a 6-line scorecard, and brings the head of marketing into the monthly cause review. Three months later the log holds 118 complaints for the month, of which 31 concern delivery dates and 9 are second complaints, and the rating recovers to 4.3.
Common mistakes
- Counting complaints without a base. A busy month produces more complaints and more orders; the rate and the repeat share say whether the problem is real.
- Tagging the symptom as the cause. "Late delivery" tells nobody what to change; the five whys reaches something a team can fix.
- Softening a broken objective claim. A stated time, price or feature that is wrong is corrected or withdrawn, not reworded gently.
- Remedy without recovery. A refund without an acknowledgement, an apology and a follow-up leaves the customer feeling processed rather than heard.
- Approval routes slower than the customer's patience. A complaint that waits days for a manager escalates while it waits.
- Logging only some channels. Reviews, social comments and field paperwork stay outside, and the same customer complains twice.
- Blaming agents for replies the process produced. The same wrong answer across several agents is a saved reply, an article or a policy problem.
- Leaving marketing out of the review. Many complaints start with a promise, and the team that made it should hear what customers said when it broke.
