Complaints Are a Conduct Metric, and Almost Nobody Uses Them That Way
Complaints Are a Conduct Metric, and Almost Nobody Uses Them That Way
Our dispute analysis argued that a dispute is a customer identifying a wrong record at their own expense — a free, specific data quality signal that most operations discard. Complaints are the same kind of signal about a different thing. A dispute says your records are wrong. A complaint says your behaviour is. They arrive through the same channels, get pooled in the same queue, and are managed to the same objective — closing items quickly — which loses nearly all of the information. And complaints have a property that makes them uniquely valuable in a collections operation: they are the only common metric that moves in the opposite direction to production.
What you'll learn
Two different signals
| Dispute | Complaint | |
|---|---|---|
| Contests | A fact | Conduct |
| Example | "The balance is wrong" | "You called me six times in a day" |
| Points at | Data and process | Behaviour and treatment |
| Fixed by | Correcting the record and the cause | Changing what the operation does |
| Can be resolved by | Investigation | Not by investigation alone |
The bottom row is why pooling them fails. An investigation can establish whether a balance is correct. It cannot establish whether calling someone six times was appropriate — that's a policy question, and routing it to a team whose job is verifying facts guarantees it gets answered as though it were one.
The practical consequence: a complaint investigated and found to involve no rule violation gets closed as unfounded, and the information that a customer found the treatment unacceptable is discarded. That information is the point. Conduct can be entirely permissible and still be generating complaints at a rate that tells you something.
Both frequently arrive together — a customer disputing a balance also objects to how they were treated — which is exactly why intake has to separate them rather than assigning one label to the interaction.
The counterweight property
The reason complaints deserve more attention than they get in collections specifically.
Nearly every metric in a collections operation moves the same direction: more contact, more promises, more recovery, higher cure rates. An operation optimizing all of them has no measure that says stop.
Our measurement analysis names this structure — a metric that improves monotonically with more of an action cannot evaluate that action. Complaint rate is the exception. It rises as pressure rises, which makes it the only common measure capable of indicating that recovery is being achieved through conduct the organization wouldn't defend.
Which produces a specific reporting requirement: complaint rate belongs on the same page as recovery, in the same review, owned by the same person. Reported separately to a compliance function, it functions as a control. Reported alongside recovery, it functions as a constraint — and the difference determines whether anyone weighs the tradeoff.
The failure mode it catches is documented in our promise analysis: collectors measured on commitments obtained will apply pressure to people who correctly said they couldn't pay. That produces promises and complaints simultaneously, and an operation seeing only the first reads it as performance.
The four categories
Categorize at intake, because each points at a different part of the operation:
| Category | The claim | Points at |
|---|---|---|
| Frequency | "You contact me too much" | Dialer configuration, campaign design, channel overlap |
| Tone and treatment | "They were aggressive" | Training, scripts, and incentive structure |
| Accuracy | "You're saying things that aren't true" | Data quality — overlaps with disputes |
| Process | "I couldn't get this resolved" | Routing, authority, and self-service gaps |
Two observations on what the distribution tells you.
Frequency complaints are usually a configuration problem, not a conduct problem. Nobody decided to call someone six times — a dialer campaign, an automated message sequence, and a manual queue all fired independently, and no system counted the total. Cross-channel contact counting is the fix, and it's a systems change rather than a training one.
Tone complaints correlate with incentive structure more than with individuals. A cluster among collectors on the same compensation plan is telling you about the plan. This is the mechanism our promise analysis identifies, appearing as a complaint pattern.
Process complaints are the most actionable and the least escalated, because they're rarely rule violations. Someone who couldn't reach anyone with authority, or couldn't set up an arrangement without a call, is describing friction — and friction is what the self-service paths in our collections framework exist to remove.
Reading the rate
Track complaints per contact attempt, not in absolute terms — absolute volume moves with portfolio size and tells you nothing about conduct.
What movement means:
- Rate rises with volume held constant — conduct or configuration changed. Investigate.
- Rate rises after a strategy change — the change did it, and the question is whether the recovery gain justifies it.
- Rate rises in one category only — points directly at the corresponding system.
- Rate rises for one team or agency — a comparison worth making regularly.
- Rate falls while recovery holds — the outcome to aim for, and it's achievable.
- Rate falls with recovery — you've reduced activity rather than improved conduct.
The comparison that matters most: plot complaint rate against recovery per contact over time. Both rising means you're buying recovery with conduct. Recovery rising while complaints hold means genuine improvement. Neither number alone distinguishes these, and most operations have only one of them.
Worth pairing with the segmentation in our cure analysis. If lift is concentrated in the uncertain middle, then heavy treatment of early-stage accounts is generating complaints in exchange for almost no incremental recovery — the worst available trade, and one that's invisible without both metrics.
Why speed is the wrong objective
Complaint handling is almost universally measured on resolution time. That matters to the person complaining and it's the wrong headline for the organization.
The reason is the same one from our dispute analysis: resolution addresses the instance and leaves the cause intact. An operation closing complaints in three days with no recorded cause will close the same complaints in three days indefinitely, and its rate will never fall.
What to record on every complaint alongside the outcome:
- Category, from the four above.
- Root cause, from a defined list — dialer configuration, script, training gap, data error, routing failure, policy, third party.
- The originating team or agency.
- Whether a policy or system change followed.
- Whether the account was also in dispute.
- Whether it escalated externally, which is a separate and more serious signal.
The monthly output that makes this worth doing: a ranked list of causes with volumes and owners. Speed then becomes a service standard rather than the objective — and the objective becomes a declining rate, which is measurable and which nobody currently measures.
The complaints you never receive
The limitation to hold alongside everything above.
Complaining requires effort, knowledge of how, and belief that it will matter. Which means the complaints you receive are a biased sample, and biased in a specific direction: toward people with the capacity to complain.
Who is systematically underrepresented:
- People in acute financial stress, who have less capacity for anything discretionary — the households in our liquidity analysis.
- People who don't know a complaint process exists, or where to find it.
- People who expect it to make things worse, which is a common and not unreasonable belief about complaining to a creditor.
- People facing language or access barriers.
- People who've disengaged entirely, who are frequently the ones treated worst.
Two consequences. A low complaint rate is weaker evidence of good conduct than it appears — it may reflect a population that doesn't complain. And the underrepresented groups overlap with the populations most likely to be treated poorly, which means the sample is biased away from exactly the cases you'd most want to see.
What to do about it: make complaining easy and visible, which will raise the rate and improve the signal. An operation that treats a rising complaint rate as a failure after making complaints easier has misread its own intervention — and monitoring outcomes by segment, per the approach in our governance framework, catches what complaints don't surface.
Third parties acting for you
An agency's conduct generates complaints attributed to your name, because the consumer recognizes the original relationship — the diffusion described in our chain analysis.
What oversight requires:
- Contractual access to their complaint data, negotiated at the outset.
- Complaint rate per contact by agency, compared across agencies and against your in-house operation.
- Category breakdown, not just volume.
- Monitoring of public complaint databases for complaints naming you.
- Call monitoring rights, exercised rather than merely held.
- Complaint rate as a placement criterion, alongside recovery.
- Defined thresholds that trigger review or removal.
The point that determines whether any of this happens: an agency selected purely on recovery rate will be an agency that recovers through whatever conduct achieves it. The comparison in our agency analysis holds — and per our benchmarking analysis, recovery rates aren't comparable across agencies anyway without matching placement age. Complaint rate per contact is more comparable than recovery rate is, which makes it unusually useful for agency selection despite being treated as secondary.
Building the process
- Single intake across every channel, including complaints arriving inside disputes or during calls.
- Separate complaints from disputes at intake, with both flags available.
- Mandatory categorization from the four types.
- Suppress collection activity on the account while a serious complaint is open.
- Investigate the conduct, including reviewing the actual contact record.
- Record root cause, mandatory before closure.
- Respond to the person, including where no rule was broken — "we reviewed this and here's what we found" is a better outcome than silence, and it costs nothing.
- Monthly review by cause, with owners assigned.
- Report complaint rate per contact alongside recovery, in the same review.
- Apply the same process to agencies and compare.
- Track whether the rate falls, which is the only test that matters.
Steps three, six, and nine are what distinguish this from ordinary complaint handling, and they add little effort to a process already running. Step nine is the one that changes behaviour, because a metric that appears next to recovery gets weighed against it and a metric that appears in a compliance appendix doesn't.
Fewer contacts, better targeted, fewer complaints
Most frequency complaints are configuration failures — separate systems contacting the same person with nobody counting the total. HL Hunt AI Debt Collection coordinates outreach across email, text, and voice with unified contact limits and self-service resolution paths, so pressure goes where it changes outcomes rather than everywhere.
Frequently asked questions
A dispute contests a fact; a complaint contests conduct. An investigation can establish whether a balance is right — it can't establish whether the treatment was appropriate.
Complaining costs effort and returns nothing, so the signal is weighted toward genuine problems. It's also the only common collections metric that moves opposite to production metrics.
Speed is a service standard, not the objective. Closing quickly without recording cause means the rate never falls — the useful measure is declining complaints per contact.
In practice yes, reputationally and in oversight terms. Outsourcing the activity while stopping the monitoring means you've outsourced the work and kept the consequence.
Key takeaways
- Disputes contest facts and complaints contest conduct — pooling them means conduct questions get answered as factual ones.
- Complaint rate is the only common collections metric that moves opposite to production, which makes it the natural constraint.
- Report it per contact and on the same page as recovery; in a compliance appendix it functions as a control rather than a constraint.
- Frequency complaints are usually a configuration failure — separate systems contacting one person with nobody counting the total.
- The complaints you receive are biased toward people with capacity to complain, so a low rate is weak evidence of good conduct.
- Complaint rate per contact is more comparable across agencies than recovery rate is, which makes it better for selection than it's treated as.
Track the metric that tells you when to stop
Get started with HL Hunt AI Debt Collection for coordinated outreach with complaint intake, categorization, and cause analytics reported alongside recovery — so the tradeoff is visible to the people making it.
This guide is educational and does not constitute legal or compliance advice. Complaint handling, response, and record-keeping obligations, and supervisory expectations for oversight of third parties, vary by institution type and jurisdiction. Consult qualified counsel about your obligations.