QA Scores 4 Calls Per Collector a Month. Reviewing All 9,000 Now Costs Less Than Postage.

A forty-seat collection floor records about 9,000 consumer conversations a month. A quality analyst working full time on call review, at roughly 25 minutes per call including the scorecard write-up and the coaching note, gets through about 160 of them. That is 1.8 percent.

Every agency owner in the country signs off on that number, because until this year the alternative did not exist at any price a mid-size agency could pay. Reviewing all 9,000 would have meant hiring 55 more analysts. So the industry standard became four calls per collector per month, and everyone agreed to call it a quality program.

What the other 98.2 percent contains

The scorecard your analyst uses is not vague. It checks whether the collector gave the required disclosure that this is an attempt to collect a debt and any information obtained will be used for that purpose. It checks the subsequent-call disclosure. It checks whether the recording notice was given in the states that require every party to consent. It checks that nothing was disclosed to a third party who answered the phone. It checks that no false urgency was created, that nothing resembling a threat of suit was said on a debt past its limitations period, that a cease request or a statement of attorney representation was honored on the spot, and that the card authorization was read back correctly before it was keyed.

Any one of those, missed, is a live problem. Missed repeatedly by one collector for six weeks before the analyst happens to pull one of their calls is a pattern — and a pattern is what a plaintiff's attorney and a state examiner are both looking for.

Reviewing every call means each recorded conversation is transcribed and scored against your own compliance scorecard within minutes of the hang-up, instead of being sampled by an analyst two weeks later.

The price change that made this ordinary

Between 2025 and 2026 the cost of running a capable AI model fell by roughly ten times. The going rate for a strong model is now around two dollars for something like three-quarters of a million words of reading and writing, and for high-volume repetitive work that runs on your own hardware it is roughly 90 percent cheaper again.

Put that in collection-floor terms. A six-minute call is about 900 spoken words. Turning that recording into text and scoring it against a twelve-point compliance scorecard costs on the order of a cent, all in. That is a fraction of what you pay to print and mail one validation notice — and you mail one of those on every single account, without ever asking whether it is worth it, because it is a cost of doing business. Reviewing every call is now cheaper than that. The thing that was uneconomic in 2025 is a rounding error in 2026.

flowchart TD
  A["Collector hangs up, recording lands in the dialer platform"] --> B["Transcript written within two minutes"]
  B --> C["Check: required disclosure given?"]
  B --> D["Check: recording notice in an all-party consent state?"]
  B --> E["Check: third party on the line, anything disclosed?"]
  B --> F["Check: card authorization read back before keying?"]
  C --> G["Scored and written to the collector's record"]
  D --> G
  E --> G
  F --> G
  G --> H["Flagged calls only into the supervisor's morning queue"]

7:50 a.m. on the supervisor's screen

The floor supervisor arrives before the 8:00 start and opens a list of six calls. Not 412 — six, out of the 412 recorded yesterday. Four are minor: two collectors trailing off the subsequent-call disclosure when the consumer starts talking over them. One is a card read-back that skipped the amount. One is genuinely serious: a collector kept talking after a woman said her husband was the account holder and he was not home.

The supervisor listens to that one clip — the flag points at the timestamp, so it is 40 seconds of listening, not six minutes — and pulls the collector before the shift starts. That conversation happens the next morning instead of eleven weeks later during the quarterly audit, which is the difference between a coaching note and a demand letter.

The compliance officer gets the other half of the value. When the hospital client's revenue cycle team runs their annual vendor audit and asks for evidence of call monitoring, the answer stops being "here are 25 sampled scorecards" and becomes "here is every call we took on your accounts, scored, with the exception rate by month and the coaching record attached." That is a different conversation about renewal.

The arithmetic, with assumptions on the table

So the same analyst now reviews 100 percent of calls by exception and has two-thirds of their month back for coaching, client audit packs and dispute research. On the other side of the ledger: the FDCPA allows statutory damages of up to $1,000 per action plus the consumer's attorney fees, and by the time a case is resolved the all-in number is usually several times the statutory amount. Call it $4,500 as an illustration. Preventing two of those a year covers the review cost roughly a hundred times over — and that ignores the placement stream you keep by passing the client audit cleanly.

How you would prove it: score last month's already-recorded calls, then compare the machine's scores against the 160 your analyst scored by hand on the same period. If they disagree, you want to know which one was right before you trust either.

What a scorecard cannot hear, and one risk nobody mentions

It hears words. It does not hear judgment. Whether a hardship claim is genuine, whether pressure crossed from firm into abusive, whether the person on the line is actually the right party or a relative covering for them — those are supervisor calls, and they will stay supervisor calls. Unfair and deceptive practice exposure is contextual by nature; a compliant script delivered coldly to a grieving spouse scores clean and is still a problem.

Second, do not wire the score straight into discipline or commission. The moment collectors believe a machine decides their pay, they will start reciting disclosures at the wall instead of to the consumer, and the numbers will look wonderful while the calls get worse. Use it to route a human to the right 60 conversations.

Third, the risk nobody puts in the sales deck: scoring every call creates a written record of every call, including the ones you failed. That record is discoverable. If your exception report shows the same collector missing the same disclosure for four straight months and nothing happened, you have handed a plaintiff's attorney the pattern argument. Before you turn this on, talk to your counsel about retention periods and — more importantly — commit to actually acting on the flags. A monitoring program you ignore is worse than no program at all.

Where to start, using calls you already have

Pull 500 recordings from last month out of your platform. Write your existing scorecard out as plain questions — the same ones on the sheet your analyst uses, in the same words. Have those 500 scored, then sit your analyst down with the 20 highest-flagged and the 20 cleanest and ask whether they agree. You will learn two things: how much your floor is actually drifting, and whether your scorecard questions are written clearly enough for anyone, human or otherwise, to answer consistently.

Frequently asked questions

Do I have to tell collectors every call is being scored?

Your employee handbook almost certainly already says calls are monitored and recorded for quality and compliance. Even so, tell them plainly, and tell them what happens with the results. Rolling it out quietly and surprising someone with a pile of flags reads as a trap and costs you more in turnover than it saves in exposure.

Does this replace my quality analyst?

No, and treating it that way wastes the money. The scoring is the cheap part; the coaching, the calibration sessions with the client, the dispute research and the exam preparation are the job. What changes is the ratio — an analyst who used to spend 80 percent of the month listening now spends most of it acting.

What about calls in Spanish?

Handled. Speech recognition and the models reading the transcripts work across dozens of languages now, and on many medical and utility portfolios a quarter or more of right-party contacts are in Spanish. Those calls have historically been the least reviewed in the building, because most floors have one bilingual supervisor and no bilingual analyst. This is often where the first real finding shows up.

Does scoring recordings create a problem in all-party consent states?

The recording is the part that carries the consent obligation, and you are already recording. Reading a recording you lawfully made does not change that. What it does do is finally let you verify, on every call rather than one in fifty, that the notice was actually given in the states that require it.

A note on the calls you never recorded

The calls your floor never gets to score are the ones nobody answered: the consumer returning your letter at 7:50 p.m., the disputed balance calling in on a Saturday, the payment arrangement someone wanted to set up while your office was closed. CallSphere builds AI voice and chat agents that answer inbound business lines and web chat around the clock, capture the caller's details and book a callback with the right person. Those conversations get recorded and transcribed like any other, which means the after-hours half of your inbound volume finally lands inside the same review process as the rest.