Almost every documented customer complaint process describes the same five or six stages, and almost every one of them works on paper. Complaints still go missing, still get answered twice, and still arrive a second time as a complaint about the silence. The stages are not where things break. The gaps between them are, and those gaps are made of three things: a complaint that exists in one person's inbox and nowhere else, a case with no owner, and a state nobody updates.
What follows is the path stated as states rather than as good intentions, with the conditions for moving between them, because a process where the transitions are undefined is a flowchart rather than a process.
The path, stated as states
A complaint is in exactly one of these at any moment, and each transition has a condition and a person.
| State | It enters when | It leaves when | Owner |
|---|---|---|---|
| Logged | The complaint exists in the register with a reference | An owner is assigned | Whoever takes intake |
| In progress | A named handler accepts it | A decision is made or the customer is awaited | The handler |
| Waiting on the customer | Information has been requested | The answer arrives, or the waiting period expires | The handler, still |
| Decided | The remedy and the reason are recorded | The customer has been told | The handler |
| Closed | The customer has been told and has not disputed within the reopen window | It reopens, or the record is archived | The queue owner |
Four details in that table do most of the work.
Waiting on the customer is a state, not an excuse. It still has an owner and it still has a clock. A case sitting in it for five weeks because nobody checked is the most common way a complaint quietly dies, and the fix is a rule: after a defined wait, the handler closes it with a note saying what was asked and what was not received, and the customer is told that.
Decided and closed are separate. A remedy agreed internally is not a resolved complaint. Merging the two states is how cases get counted as closed while the customer is still waiting to hear.
There is no state called escalated. Escalation is a change of owner, which is covered by the existing states. A separate escalated bucket becomes a place where difficult cases are parked without anybody holding them.
Five states is the ceiling. Processes with nine or eleven states are maintained accurately for about a month.
One register, whatever the channel
Complaints do not arrive through the complaints channel. They arrive by phone, in a reply to an unrelated email, in a public review, through a form, and via whoever the customer happens to know. A process that only covers the form covers a fraction of the volume, and the fraction it misses is biased toward the loudest cases.
The requirement is unglamorous: every complaint gets into one register, with one reference, regardless of how it arrived. That means someone taking a phone call logs it rather than resolving it verbally and moving on, and someone who receives a complaint in an email thread copies it in rather than replying and forgetting.
Getting this adopted depends almost entirely on how long logging takes. Thirty seconds and it happens. A form with twelve required fields, including ones only relevant to certain complaint types, and staff will handle the case informally and not log it, which is rational behaviour and fatal to the register.
So the intake route used by staff should be shorter than the one used by customers: who, what, when it happened, how it arrived, and a paste of whatever the customer wrote. Everything else gets filled in during handling. Where customer facing complaints already arrive through a form, the simplest arrangement is to point the staff route at the same queue, so phone and email complaints sit in the same list as submitted ones. The same pattern covers other kinds of incoming request, and a complaint should not be the one thing that lives outside it.
One consequence worth accepting deliberately: the number of recorded complaints will rise sharply in the first quarter after this is enforced, and none of that rise is new problems. Anybody reporting on the numbers needs to know that in advance, or the improvement will look like a deterioration.
The two clocks that matter
Three intervals get discussed and only two of them are useful.
Acknowledgement. Automatic, immediate, and not a measurement of anything. It exists to stop the customer chasing and to give them a reference. An acknowledgement counted as a response is the most common way a complaints report flatters the reality.
Time to the first substantive reply. Written by a person, addressing the actual complaint, even if the answer is that it is being investigated and by whom. This is the number that determines whether the customer complains a second time about the silence, and it is the number most responsive to process changes. Report it as a median, and report the worst case separately.
Time to resolution. Worth tracking and a poor management number on its own, because it is dominated by a small number of genuinely complicated cases. A median resolution time of four days with a handful at ninety days describes a functioning process. The average of the same set describes a crisis that is not occurring.
Set a target on the first substantive reply and publish it internally. One working day is achievable for most businesses. Two is comfortable. Whatever the number, the target should be met more often than not in the first month it exists, otherwise it becomes a number nobody looks at.
The clock also needs a defined start. Time from arrival, not from when someone opened it, otherwise a case sitting unread for three days shows as a fast response. This is the one measurement detail that a system has to handle, because a timestamp applied by hand is a timestamp applied kindly.
Ownership, and what escalation actually means
One name per case, always, including while it waits on the customer. A queue where cases belong to a team is a queue where the difficult ones belong to nobody, and the difficult ones are the ones that turn into public complaints.
Escalation is reassignment, not notification. Copying a manager into a thread changes nothing about who is responsible. Moving the case to that manager, with the state and the history intact, does. Three triggers are worth defining in advance, so escalation happens on a rule rather than on a mood:
- The remedy the customer is asking for exceeds the handler's authority.
- The case touches safety, a regulated matter, a legal threat, or the press.
- The first reply target has been missed, or the case has been open beyond a defined period.
That third trigger is the one that gets left out and the one that prevents the most damage. A case that has been open for three weeks without a reply needs to move to someone with the standing to unblock it, and the move should be automatic rather than dependent on the handler admitting the problem.
Where cases arrive as form responses, this is mechanical rather than a matter of discipline: an owner and a status attached to each response is the whole apparatus, and reassignment is a change of owner rather than a forwarded email with a new subject line.
Deciding the remedy, and who may give it
Most complaint processes are slow for a reason that has nothing to do with investigation. The handler knows the answer on day one and cannot authorise it, so the case waits for a decision that is never anybody's priority.
The fix is to decide the common remedies once, in advance, and delegate them. Something like: any refund up to a stated amount, any replacement of a stated class of item, and any apology with a goodwill credit up to a stated limit may be given by whoever is handling the case, without asking. Above that, escalation applies.
Two things follow from this. The vast majority of complaints get resolved on the first or second contact, which is where customer satisfaction after a complaint is actually won. And the escalation queue shrinks to the cases that genuinely need judgment, which means those get attention.
Whatever the remedy, record the reason alongside it, in a sentence. Not the category of the complaint, but why this outcome was chosen. Six months later that sentence is the only defence against inconsistent treatment of two similar cases, which is the complaint that follows a complaint.
Closing, and the one field that pays for the whole process
Closing has two parts and one of them is routinely skipped.
The part that gets done is telling the customer. State the outcome, the reason, and what has changed as a result. Confirm it in writing even when it was agreed by phone, because a complaint resolved in an undocumented conversation is a complaint that can reopen with a different account of what was agreed.
The part that gets skipped is recording the cause. Not the category, which describes what the customer complained about, but what actually went wrong: the step that failed, the instruction that was unclear, the supplier that shipped late. One line, written by the handler at closing, taking about fifteen seconds.
That single field is the difference between a complaints process that handles complaints and one that reduces them. Without it, the twelfth case caused by the same broken step is the twelfth unrelated complaint, and the pattern is invisible to everybody except the handler who noticed and has no way to say so.
Also define the reopen rule. A defined window in which a closed case can be reopened, keeping the original reference, prevents the alternative, which is a second complaint logged as a new case and a register that double counts.
Review the process, not just the complaints
Handling cases well and improving anything are separate activities, and the second one needs a slot in the calendar or it does not happen.
The recognised general guidance here is ISO 10002:2018, "Quality management. Customer satisfaction. Guidelines for complaints handling in organizations", published in July 2018 and confirmed as still current in 2023. It is guidance rather than a certifiable requirement, and it covers planning, operation, maintenance, and improvement of the complaints handling process, including management involvement, resourcing and training, and auditing and reviewing the process itself. Annex B addresses small businesses specifically.
In practice, quarterly is enough, and the agenda is short. Read the causes recorded at closing, grouped, and pick the top two to fix at source. Look at the cases that breached the first reply target and ask what they had in common, which is usually a specific handover rather than laziness. Check whether the category list still matches what arrives, and prune it. And look at the cases closed as waiting on the customer, because a rising number there often means the information being requested is something the business could have looked up itself.
What to change first
Put every complaint into one register regardless of how it arrived, and measure time from arrival to the first reply written by a person, as a median. Then delegate the common remedies with a stated limit, because most of the delay in a complaint process is waiting for permission rather than waiting for facts. Halict shows what a case looks like when the submission, the owner, the status, and the reply are on one screen.
Q1. What are the stages of a customer complaint process?
Logged, in progress, waiting on the customer, decided, and closed. What matters more than the names is that each transition has a condition and a named owner, and that decided and closed stay separate, so an internally agreed remedy is never counted as a resolved complaint before the customer has been told.
Q2. How quickly should a complaint be answered?
Acknowledge immediately and automatically, then aim for a substantive reply written by a person within one or two working days. Regulated sectors have their own deadlines set by a regulator or a contract, and those should be checked directly rather than taken from a general article.
Q3. How should complaints that arrive by phone be handled?
Logged in the same register as everything else, with a short staff intake route that takes under a minute. If logging a phone complaint is slower than resolving it informally, staff will resolve it informally, and the register will describe a fraction of reality that is biased toward the loudest cases.
Q4. What does escalation mean in a complaint process?
Reassigning the case to someone with more authority, with the history and state intact. Copying a manager into an email thread is notification, not escalation, and it changes nothing about who is responsible. Define the triggers in advance, including one based on how long the case has been open.
Q5. What should be recorded when a complaint is closed?
The outcome, the reason it was chosen, and one line describing what actually went wrong. The last of those is the field that makes the process reduce complaints rather than merely process them, and it is the one most often left blank because it takes a handler fifteen seconds and benefits somebody else.