Last updated: September 12, 2026
What an auto fail is
In call center QA, an auto fail is a scorecard item that voids the entire score for a call when one specific behavior happens, whatever the agent did right everywhere else. Programs also call it a critical fail, a fatal error, or a zero tolerance item. Choosing which behaviors get that treatment is the hardest decision on the card.
Every other line on a scorecard trades against the rest. Eight points for the greeting sit next to ten for capture accuracy, and a strong call absorbs a weak moment in the middle. An auto fail says one behavior does not trade, so an agent who ran a clean, accurate call and skipped the recording disclosure has a zero, and the other criteria have nothing to say about it.
That is a judgment about what your program will never sell for points, and it changes floor behavior inside a week. It deserves more argument than the weights get and usually receives less.
What earns a spot on the list
Four kinds of behavior show up on almost every auto fail list we see, each carrying a consequence outside the QA program.
Regulatory breaches come first, because somebody other than you enforces them. A benefit discussed before the TPMO disclaimer was delivered is one of those, and a plan specific conversation with no Scope of Appointment on file is another. The Medicare sales call scorecard carries six of them across fourteen criteria, heavy for a general floor and ordinary under CMS rules.
Safety is the shortest list. An answering service that takes an emergency and never reaches the on-call schedule has produced an outcome no number describes, which is why the answering service scorecard puts a failed dispatch beside the wrong business name in the greeting. Data protection sits next to it, since account information released to a caller whose identity was never established stays absolute whether or not anything came of it.
Gross misconduct is the last of the four, and it is where the argument starts. Hostile tone reads like an obvious auto fail until two supervisors disagree about where firm became rude, and an absolute that depends on a reviewer's mood is one somebody talks their way out of. It stays on the general 100 point scorecard anyway, worded around what the agent said rather than how it sounded, because the alternative is a card with no answer for the worst call of the month.
Verticals add their own. Leasing calls carry fair housing exposures no other floor has, and an intake specialist who predicts a settlement range hands the firm a promise nobody made, which is why the legal intake card treats restraint as absolute.
The test that keeps a list honest is whether the call could still be called good with everything else perfect. A heavy deduction covers it when the answer is yes. If the call should never have been logged that way, the line is an auto fail, and the rubric page works through how that sits against the weights.
Where the flag sits in the arithmetic
An auto fail is a flag rather than a number, so it belongs outside the point total. Mark it before you total anything, because a reviewer working the other way around tends to soften the flag until the number looks reasonable.
Two conventions exist for the score itself. The call gets zeroed, or it gets scored normally with the flag carried beside it. Zeroing states the seriousness and costs you the coaching data, since a 94 point call with one disclosure miss and a genuinely sloppy 61 land in the same place. The templates here do both.
Watch what a flag does to an average. Say an agent takes 1,400 calls a month and you score 60 of them. One auto fail on a call that would otherwise have scored 94 pulls the monthly average down about 1.6 points, invisible next to the 85 percent SQM Group puts the industry average at. Count flagged calls per agent and per account instead, the same day they fire.
When an agent disputes one
Auto fail disputes run hotter than point disputes, since a flag usually triggers a documented coaching conversation and sometimes more than that. Three things belong on the table before that meeting: the timestamp where the criterion fired, its exact wording, and the account script or policy that wording points back to. A criterion written as an absolute and worded loosely loses the argument, and it should.
Outcomes match the three the dispute workflow describes for any criterion, and the rewrite is the valuable one here, since it settles every call after it.
Calibrate the auto fails on their own
Reviewers disagree about auto fails more than about anything else on the card, which makes sense given the item has no middle. A criterion scored 0 to 5 lets two supervisors land a point apart and call it close enough, while fail or no fail gives them nowhere to meet.
So run an hour that scores nothing but the flags. Five calls, every reviewer marks each auto fail item independently, then count the splits. Calibration variance on a binary item reads easily, since any disagreement means the wording is wrong.
Which ones a machine catches
Rule based auto fails hold up under automated scoring. Two timestamps on a speaker separated transcript answer whether a disclosure landed before the caller's first substantive line, and a callback number read back digit by digit works the same way.
The softer flags stay human. Hostile tone, or whether a benefit description left the beneficiary with the wrong impression, needs somebody who knows the material and can hear the call. Aircall and StackAI put the reliability line in the same place.
Every scored line in CallSniper carries a timestamp and a confidence value, so a high confidence flag on a missing disclosure goes to a named person and low confidence lines queue for review. None of that helps if nobody opens the queue.
Nine absolutes and a fail rate nobody works
The common failure is a list that grew. Somebody adds an item after a bad call, nobody removes one, and the card carries nine absolutes covering everything from a regulatory breach to a missed callback promise.
What that produces is a fail rate high enough to ignore. Say the list flags 12 percent of scored calls. Nobody opens all of those the same day, the queue ages, and the flag stops meaning much to the floor. Agents watching a regulatory breach and a soft one land in the same column learn the score is arbitrary, which takes a quarter to undo.
Keeping the list short is a filing question rather than a philosophy. Each item needs two things: a consequence you can name outside the QA program (a regulator, a client contract, a caller waiting on a dispatch, somebody's bar card) and a person who owns the same day review. Anything failing either half is not an absolute, and five or so is where most general cards belong.
That leaves behavior that genuinely matters and does not clear the bar. Make it the heaviest weighted criterion on the card instead, because twelve points off a hundred survives a coaching conversation and leaves the call scoreable. Counting the event without attaching points works too.
A review of your own recordings scores six months of your own historical audio against the card you run today, which is the cheapest way to see how many calls your current auto fail list actually catches, for a fixed pilot price.
Frequently asked questions
What is the difference between an auto fail and a critical fail?
Nothing, in practice. Critical fail, fatal error, and zero tolerance item describe the same mechanic, with the label varying by program rather than by meaning. CallSniper flags them as critical fails.
How many auto fails should a call center scorecard have?
Around five for a general floor, which is our editorial line rather than a published standard. Compliance heavy work runs higher, and a recorded Medicare sales call reasonably carries six. Past fifteen the list mostly records who won rubric meetings.
Can automated QA detect an auto fail reliably?
On rule based items, yes, since the transcript settles a disclosure or a read back number. Tone and judgment items need a reviewer, so treat a confidence value there as a routing decision rather than a verdict. More on how a call becomes a score, with the surrounding vocabulary in the call center QA glossary.
Sources
- Industry average QA score of 85 percent. SQM Group. Third party, unverified.
- Rule based criteria scoring more reliably than tone and empathy. Aircall and StackAI. Third party, unverified.
- Transcript timestamps, confidence values, scorecards with points per criterion, and critical fail flags. Trinsic feature sheet. Verified internally.
- TPMO disclaimer timing and the Scope of Appointment requirement. CMS marketing rules. Verify against the current text before publish.
- Auto fail counts on the templates cited here, and the recommendation of about five. Editorial judgment.