Ask an agency owner what their quality assurance programme costs and you will usually get a software number. That number is almost never the expensive part. The cost of QA on a collections floor is supervisor hours, and most floors have never worked out how many they are spending or what those hours are buying.
This is the arithmetic, done in units you can check against your own roster. We sell a tool in this category, so treat the last two sections as the argument they are — but the maths in the first half holds regardless of what you buy, including if you buy nothing.
Start with minutes per review, not licences per seat
One reviewed call is not one call length. A supervisor listening properly does five things: pulls the recording, listens (often at 1.25× with rewinds), scores it against the sheet, writes a note that will still make sense in six months, and files it somewhere retrievable.
On a collections floor where a real conversation runs four to eight minutes, a careful review lands somewhere between six and twelve minutes of supervisor time. Call it eight. That number is the hinge the entire programme swings on, and it is worth measuring on your own floor rather than accepting ours: time ten reviews with a stopwatch and take the median.
Now multiply. A collector who has 200 real conversations in a month generates, at various coverage levels, the following supervisor load:
| Review coverage | Calls reviewed per collector per month | Supervisor hours per collector | At 20 collectors | At 50 collectors |
|---|---|---|---|---|
| 2% | 4 | ~0.5 hr | ~11 hr | ~27 hr |
| 5% | 10 | ~1.3 hr | ~27 hr | ~67 hr |
| 10% | 20 | ~2.7 hr | ~53 hr | ~133 hr |
| 25% | 50 | ~6.7 hr | ~133 hr | ~333 hr |
| 100% | 200 | ~26.7 hr | ~533 hr | ~1,333 hr |
A working month is roughly 160 hours before holidays, meetings and everything else a supervisor does that is not QA. Read the table with that in mind and the shape of the industry explains itself. Two percent is not a considered risk position. It is the number that fits in the gaps of a supervisor's week, which is the same conclusion the maths reaches from a different direction in why two percent sampling cannot tell you what your floor is doing.
Multiply the hours column by your own loaded hourly cost for a supervisor — base plus payroll tax plus benefits, divided by productive hours — and you have your programme's real budget line. The Bureau of Labor Statistics publishes average wage data for first-line supervisors of office and administrative support workers if you want an external benchmark for that figure, though your own payroll number is the one that matters.
Where hiring stops working
The instinct when coverage looks thin is to add a reviewer. It works, briefly, and then it stops for three reasons that compound.
Reviewers do not scale linearly. A second reviewer does not double coverage, because two people scoring independently drift apart. Closing that gap takes calibration — a recurring session where the same call is scored by everyone and the disagreements are argued out. Budget an hour or two a month per reviewer for that, and treat it as unavoidable overhead rather than a nice-to-have; the scorecard piece covers why an uncalibrated second scorer can make the record worse rather than better.
The marginal hour is spent on clean calls. At 10 percent coverage a reviewer spends roughly nine of every ten reviews confirming that a fine call was fine. If the thing you are hunting is rare, most of the money you add goes into re-confirming the common case. The yield per hour falls as you scale, which is the opposite of how a headcount business case is usually written.
Coverage and depth trade against each other. Push a reviewer to 30 calls a day and the eight minutes becomes four. You have bought coverage by spending quality, and the notes get shorter exactly when you would most want them to be specific.
None of this means never hire. It means a reviewer is worth adding when you have a defined backlog of work only a person can do — disputes, escalations, coaching conversations — and is a poor instrument for buying raw coverage.
What automation actually moves, and what it does not
Automated review changes one term in the equation and leaves the others alone. It removes the listening-and-scoring minutes on the calls where nothing happened. It does not remove the supervisor.
Run the same floor with every call transcribed and scored against a collections rubric, and the supervisor's month looks like this instead:
| Activity | 20 collectors, manual at 5% | 20 collectors, full review with triage |
|---|---|---|
| Listening to unremarkable calls | ~24 hr | ~0 hr |
| Working flagged moments | folded into the above | ~8-14 hr |
| Coaching conversations | often skipped for time | ~6-10 hr |
| Calibration and disputes | ~2 hr | ~3 hr |
| Coverage the record supports | 5% of calls | every recorded call |
The hours do not go to zero. They move from listening to acting, which is the only part a supervisor is uniquely good at. That reallocation is the honest claim; anything stronger is a sales pitch. And a flag is a pointer to a moment, not a verdict — what a supervisor does with one is its own discipline, covered in a flag is not a finding.
The costs people forget to count
Three line items go missing from almost every QA budget we see.
Retrieval time. When a complaint arrives eight months after the call, somebody has to find the recording. On a floor with an unindexed archive that is an afternoon; with searchable transcripts it is minutes. It is invisible in the monthly budget and very visible in the month it happens.
The cost of not knowing. This one resists a number and it is real. A collector with a habit nobody has heard is a compounding liability, and every month at low coverage is another month of compounding. We are not going to put a dollar figure on it, because any figure we invented would be marketing rather than arithmetic. The Consumer Financial Protection Bureau publishes complaint volumes and supervisory observations for debt collection, and the Federal Trade Commission reports annually on collection complaints and enforcement — those are the public numbers worth reading before you decide the risk is theoretical.
Onboarding and the first month. Any new programme costs a baseline period where the numbers mean nothing yet, plus the time to tell your staff what is happening and why. That conversation is not optional and it is not free.
What this costs here, plainly
Our own pricing is per monitored collector per month: $39 for activity and screenshots, $99 with call recording, transcription and the FDCPA rubric, $149 with screen recording and three-year retention. Supervisor and administrator logins are free, because charging for the seat that does the reviewing would be charging you twice for the same work. The full breakdown and seat minimums are on pricing, and what each tier contains is on features.
Set that against the hours table above and the comparison is straightforward on a twenty-seat floor: the software is a line item, the 500-odd supervisor hours full manual review would take are not a line item you could staff at all. That is the actual case. It is not that the tool is cheap; it is that the alternative at equivalent coverage does not exist.
The honest objections
Automated review creates its own workload. True, and it is the most common surprise. A floor moving from 4 reviewed calls per collector to every call will see more flags in week one than the old programme produced in a quarter. Most of that is a backlog of things that were always happening. Plan for a heavy first fortnight and resist the urge to switch off categories to make the queue look better.
Not every flag is a problem. Also true. A rubric that never over-flags is a rubric that is missing things, so some proportion of what surfaces will be nothing on inspection. Budget for it, measure it, and tune. The alternative — a system tuned so tightly it stays quiet — is comfortable and useless.
The voice side has a hard prerequisite. Audio is captured from the collector's computer. That is what makes it work with any softphone, and it also means a desk handset that never touches the machine produces nothing to capture. If your floor is on physical phones, the activity side works and the call side does not. Settle that before any of this arithmetic applies to you; it is one of the constraints we publish on compliance rather than leave for you to discover.
Review is next-morning, by design. This is a supervision system, not a live-listen or barge console. If your business case depends on a supervisor intervening during a call in progress, the hours model above is the wrong one and so is this product.
As of September 2026
Prices quoted here are the published plan prices as of September 2026 and the seat minimums are on the pricing page. The regulatory material referenced — the Fair Debt Collection Practices Act and Regulation F — is public at the Electronic Code of Federal Regulations, with rulemaking history at the Federal Register. Nothing here is legal advice; it is arithmetic about staffing, and where a legal conclusion would begin we stop and point you at counsel.
If you want to sanity-check the model against your own floor, tell us your headcount and your dialer and we will run the hours with your numbers rather than ours, including the cases where the answer is that you do not need us yet.
Frequently asked questions
How many supervisor hours does a QA programme really take?
Roughly eight minutes per call reviewed, once you count retrieval, listening, scoring and writing a usable note. At 5 percent coverage on a twenty-seat floor that is about 27 hours a month; at 100 percent manual review it is over 500, which is why no floor does it that way. Measure your own median rather than trusting ours — time ten reviews and see.
Is it cheaper to hire another reviewer than to buy software?
For raw coverage, rarely, because reviewer capacity scales sublinearly: a second scorer needs calibration time and still spends most hours confirming that clean calls were clean. For work only a person can do — disputes, escalations, coaching conversations — a reviewer is the right spend and no tool replaces them.
Does automated scoring reduce headcount?
Not in our experience, and we would not sell it that way. It moves supervisor time out of listening and into acting on what surfaced. Floors that treat it as a headcount cut usually end up with a flag queue nobody works, which is a worse position than not having the flags.
What does CollectionsQA cost?
$39 per monitored collector per month for activity and screenshots, $99 with call recording, transcription and the FDCPA rubric, and $149 with screen recording and three-year retention. Supervisor and admin logins are free. Seat minimums and the full inclusions are on the pricing page.
What is the hidden cost people miss?
Retrieval. Finding one specific call eight months later is an afternoon in an unindexed archive and a couple of minutes against searchable transcripts, and it never shows up in a budget until the month a complaint forces it to.
We take card payments over the phone — does that change the maths?
It changes what you should switch on before it changes the cost. Card handling on the call path is a conversation to have before you start rather than after the first PCI question, and it is one of the constraints we state up front instead of burying.
Reviewing two percent of your calls?
CollectionsQA records, transcribes and scores every call your collectors take, and sends one supervisor report each morning naming the calls that need a human.