Last updated: September 12, 2026
What quality parameters are
Quality parameters are the scored criteria on a call center scorecard, grouped into compliance, process adherence, communication, resolution, and documentation. Each one is either fatal, where a breach voids the whole call, or non fatal and worth points. The general scorecard we publish runs seventeen scored parameters to a hundred points with five fatal items.
Parameter is the word the work uses. A QA analyst in Manila or Cebu writes parameters at the top of the form and walks a client deck parameter by parameter, while the same column reads criteria once the program runs out of Phoenix.
Each line gets worded so a reviewer can mark it pass or fail without arguing about what it meant, and its weight is the points it carries.
Fatal and non fatal parameters
Every scorecard here sorts into two piles before a single weight gets set. Non fatal parameters cost their points and leave the rest of the score standing, so an agent who misses the empathy line on a clean call lands at 95. Fatal parameters, called auto fails on one floor and zero tolerance items on the next, take the call to zero.
A fatal flag also moves differently through the program. It reaches a supervisor the day it lands rather than waiting on a monthly average, and it gets counted rather than averaged, since a mean buries one skipped disclosure in a month.
Which is why the list stays short. Ours has five: a required disclosure skipped, actionable information that contradicts current policy, profanity or a hostile tone, a live caller disconnected without cause, and account information released to an unverified caller.
Client pressure grows that list, usually right after an incident, and a fatal column at twelve items stops separating a bad call from a disaster. We argued the other side once with a program whose regulator leaves them no room, and lost. The test is whether you would genuinely zero a 97 point call over the item, and both lists sit on a 100 point QA scorecard template.
The five parameter groups
Grouping lets a score be read by section, which tells a team lead where the coaching hour goes. Compliance covers what the law or the client contract requires to be said and checked. Process adherence takes the steps the program itself defines, from the branded greeting to the approved sign off. Communication is the how: pace, listening, filler words, and the empathy line. Resolution is whether the caller got a correct answer and an accurate next step. Documentation is what reaches the system of record after the call.
Seventeen scored lines sit across the five, weighted off our own template rather than a standard.
Compliance parameters
Fifteen points over three lines: the disclosure read in full, the caller verified to the level the account policy sets, and the recording notice where the jurisdiction calls for one. The disclosure line passes when the current script wording arrives complete and before the part of the call it has to precede, and fails when it comes paraphrased or late. Skipping it outright is the fatal version, so the parameter grades delivery while the fatal item catches absence.
Process adherence parameters
Eighteen points, weighted toward probing questions. The branded greeting is mechanical, which is the reason it carries 5 rather than more. Probing questions carry 8 and pass at two questions before the agent proposes anything, and a call that fails that line usually sounds good, since the agent was fast and certain and the wrong answer arrived early.
Communication parameters
The largest group at 32 points, and the one where scoring gets hard. Letting the caller finish carries 6, and restating their issue accurately carries another 6, which catches an agent solving the problem they assumed rather than the one they heard.
The empathy statement takes 5 and deserves a warning. A pass is an acknowledgment where the caller expressed frustration, in words that fit what they said. Anybody scoring it is judging another person's tone, and two good reviewers split on the same call more often than either expects.
Resolution parameters
Twenty five points, carrying the heaviest single line on the form. Information that matches current policy is worth 10, because a wrong answer delivered well creates the callback and the complaint behind it. An accurate next step with a timeframe takes 8, and resolving without an avoidable transfer or hold takes 7, where the word doing the work is avoidable.
Documentation parameters
Ten points, and the group nothing in the audio can settle. Capturing the name, callback number, and reason for the call is checkable in part when the number was read back digit by digit. Documenting the interaction in the system of record sits outside the call: somebody opens the CRM next to the transcript, or the line goes unscored.
Weighting them across programs
Weights total 100 by convention, so the score reads as a percentage a client can set beside their own numbers. What sits on each line states what the program cares about, which means a scorecard where every parameter is worth 6 points has decided nothing.
Weight also moves between programs. A collections program puts its points on disclosure and verification, a retention program moves them to discovery, and copying one weight set onto the next is the standard failure. Put whatever you land on in writing with a version and a date, since scores from two weight sets cannot be compared and somebody will try it in a quarterly review. That is the argument for scoring against a rubric held in one place.
Which parameters a machine can verify
A parameter list is cheap to write and expensive to run, and what separates the two is whether the check can be made against the transcript or needs a person listening. Rule based parameters hold up under automated scoring and judgment parameters get shakier, which Aircall and StackAI both note.
| Group | Parameter | Points | How it gets checked | How well that holds |
|---|---|---|---|---|
| Compliance | Required disclosure lines read in full | 5 | Phrase match against the current script text, plus a timestamp against the first substantive exchange | Holds |
| Compliance | Caller identity verified to the level the account policy requires | 5 | The verification steps detected in the transcript in the required order | Holds on presence, needs the record for accuracy |
| Compliance | Recording or monitoring notice delivered where required | 5 | Phrase match, checked against the routing rule for the caller's jurisdiction | Holds |
| Process adherence | Answered with the approved branded greeting | 5 | Compared against the greeting string stored for the program | Holds |
| Process adherence | Asked at least two probing questions before proposing a solution | 8 | Question detection against the probing question library | Counts questions, cannot judge whether they were the right ones |
| Process adherence | Offered further assistance and used the approved sign off | 5 | Phrase match against the approved close | Holds |
| Communication | Let the caller finish without interrupting | 6 | Overlap between the two speakers on the timeline | Depends on speaker separation |
| Communication | Restated the caller's issue accurately before acting on it | 6 | A reviewer comparing the restatement against the caller's own words | Reviewer |
| Communication | Held an audible pace with clear articulation | 5 | Words per minute from the transcript timings | Pace measures, clarity does not |
| Communication | Stayed inside the filler word and slang tolerance | 5 | Counted against the vocabulary library for the program | Holds |
| Communication | Framed a limitation in positive terms with an alternative | 5 | A reviewer reading the exchange | Reviewer |
| Communication | Used an empathy statement where the caller expressed frustration | 5 | Phrase match against the empathy library finds the line | Finds the phrase, cannot tell you it landed |
| Resolution | Gave information that matches current policy | 10 | Compared against the program's own answer set where one exists | Holds only where the answers are fixed |
| Resolution | Resolved the call without an avoidable transfer or hold | 7 | Transfer and hold events from the call metadata | Events are exact, avoidable is a reviewer call |
| Resolution | Set an accurate next step with a timeframe | 8 | The timeframe phrase is detectable in the transcript | Detects the promise, not whether it was correct |
| Documentation | Captured caller name, callback number, and reason for the call | 5 | The read back detected in the transcript, digits compared against the record | Needs the record beside the call |
| Documentation | Documented the interaction in the system of record | 5 | Opening the system of record next to the transcript | Outside the audio entirely |
The list downloads as a CSV with the same five columns, so you can cut the rows your program does not run and hand the rest to a client.
Empathy is the row to sit with. A phrase match finds the sentence, and whether the caller felt heard is nowhere in the transcript, so a reviewer who listens will disagree with that flag often enough to cost you the floor's trust. Speaker separation is the other soft spot, since interruption and dead air depend on knowing who was talking, and a recording where both sides share one channel gives the model less to work with.
So machine scoring carries the rule based lines and the counting, every flag arrives with a timestamp and a confidence score, and reviewer hours go to the judgment parameters.
How parameters get audited, and by whom
Scoring and auditing get used as one word for two jobs. A score grades a call. An audit asks whether the score means anything, and four people ask it: your QA analyst, a QA lead running calibration, the client's quality team rescoring a sample of what you scored, and a governance review aimed at the program rather than the calls.
When a client asks for your call quality audit parameters, they mean the list above plus four checks on the machinery behind it: the sample and the rule you drew it by, the rubric version in force on the call date, the evidence behind each deduction, and whether every fatal flag matches an event they can hear.
Coverage is the uncomfortable part. Manual review reaches 1 to 3 percent of calls on the usual estimates from Verint, Gistly, and ViciStack, and SQM Group reports 60 percent of centers evaluating five or more calls per agent per month. Say a program runs 40 agents at 1,400 calls each: 56,000 calls a month, and five per agent covers 200 of them. Call Centre Helper puts a defensible read at 90 percent confidence closer to 65 evaluations per agent, which is 2,600 calls and thirteen times the review hours.
What a client tests is whether your evidence survives them rescoring it, which is the job of the client report these scores end up in.
Calibrating parameters across sites
A program running in two locations is two QA teams reading one list, and drift between them goes in a predictable direction: the rule based parameters hold and the judgment ones spread.
Calibration is the standing fix. Both sites score the same ten calls without talking, the scores get compared line by line, and the spread on any single parameter is the number worth chasing. Empathy comes back widest and disclosure comes back identical. Attach a written pass example and fail example to each judgment parameter, then treat stubborn variance as a wording problem, since a parameter two trained reviewers read differently is written badly rather than hard.
None of that converts a judgment parameter into a rule based one. The mechanics sit in calibrating reviewers on judgment parameters.
Frequently asked questions
What are quality parameters in BPO?
Quality parameters are the criteria on a call scorecard, each worded so a reviewer can mark it pass or fail, and each carrying a weight. They group into compliance, process adherence, communication, resolution, and documentation. Parameter is the standard word on a BPO floor, where a US contact center writes criteria in the same column.
How many parameters should a call center scorecard have?
No industry standard count exists, and anybody quoting one is quoting their own template. Ours runs seventeen scored parameters to a hundred points with five fatal items. Below about ten a score stops telling a team lead anything, and past twenty five reviewers score on autopilot, though both numbers are our judgment.
What is a fatal error in call center quality?
A fatal error, also called an auto fail or a zero tolerance item, voids the whole call score regardless of the points earned elsewhere. A skipped disclosure and account information released to an unverified caller are the two most common. A long fatal list stops separating a bad call from a serious one, so the list stays short.
How are quality parameters weighted?
Weights total 100 so the score reads as a percentage your client can compare with their own, and the number on each line says how much the program cares about that behavior. Versioning them with a date matters more than the weights themselves, since scores from two weight sets cannot be compared.
Can AI score call center quality parameters?
Rule based parameters score reliably, including disclosure wording, a greeting string, a filler word count, and a dead air threshold. Judgment parameters like empathy are weaker, and the gap is a property of the problem rather than a tuning issue. Machine scoring takes the rule based lines and reviewer time goes to the judgment ones and the low confidence flags.
The client's parameters and yours
A BPO floor is several programs at once, and each client arrives with its own definitions. One client's fatal list has four items on it. The next calls the same breach a 10 point deduction. A third hands you the form their previous vendor used and asks you to score against it, wording and all.
Blending that into one house standard so the floor produces a single number serves nobody. A client cannot find their own criteria in the blend, and the first real audit takes it apart.
What holds is a rubric per program, scored and reported per program, with a house layer underneath carrying only the parameters every program shares, usually disclosure, verification, and documentation. Which is more scorecards to maintain, and yes, that is the cost. How many programs you carry is usually what decides outsourcing QA against running it in software, and the vertical version sits on QA for BPO call centers.
Send a week of one program's audio through a call review and we will score it against that client's own parameter list, fatal items included, so the first report you hand them is built from their definitions instead of yours.
Sources
- Our own published scorecard templates: seventeen scored parameters to 100 points, five fatal items, and the per line weights quoted throughout. Verified internally, cited as our own.
- Trinsic feature sheet and call notes, 2026-08-27 and 2026-09-03. Verified internally.
- Aircall and StackAI on rule based criteria scoring more reliably than tone. Third party, unverified.
- Verint, Gistly, and ViciStack on manual QA reaching 1 to 3 percent of calls. Third party, unverified, setup only.
- SQM Group on 60 percent of centers evaluating five or more calls per agent per month. Third party, unverified.
- Call Centre Helper on roughly 65 evaluations per agent per month at 90 percent confidence. Third party, unverified.
- The 40 agent program at 1,400 calls per agent per month. Editorial hypothetical.
- Parameter weights, group totals, and the scorecard size range in the FAQ. Editorial judgment.