How Call Auditors Works
Call Auditors turns recorded phone calls into objective, repeatable quality scores. You upload a call, we transcribe it, and our scoring engine measures it against a model you control — the words and phrases that matter to your business, weighted the way you decide.
- ConsistentEvery call is judged against the same rules — no auditor fatigue, no drift.
- CustomizableYou define the words, phrases, and weights that drive the score.
- AffordableBuilt for smaller businesses that can't justify enterprise call-center suites.
The 5-Step Process
- Upload File Submit a recorded call in a common audio format.
- Speech to Text The audio is converted into a searchable transcript.
- Choose Scoring Model Pick the saved word-and-weight model that fits this call type.
- Detect Words Required, positive, and negative words and phrases are identified in the transcript.
- Calculate Score Your weights combine into a final, explainable score.
Step-by-Step Detail
1. Upload File
You start by uploading a recording of the call you want audited. The file is transferred securely and held only as long as needed to produce your transcript and score.
Technical detail
Common audio formats are accepted (such as MP3, WAV, and M4A). Files are uploaded over an encrypted connection. There is a per-file size cap to keep processing fast — long calls can be split into segments if needed.
2. Speech-to-Text Conversion
Once uploaded, the recording is run through a speech-to-text engine that produces a written transcript of everything said during the call. This transcript is what every later step works with — the scoring engine never "listens" to the audio directly.
Technical detail
Transcription accuracy depends on audio quality (clarity, background noise, microphone quality) and on speaker accent and pace. Where supported, speaker labels distinguish the agent from the customer so you can later restrict scoring to just one side of the conversation. Punctuation and capitalization are normalized so word-matching is consistent.
3. Select the Scoring Model
A "scoring model" in Call Auditors is simply a saved set of rules: which words and phrases count as required, which are positive, which are negative, and how much each one is worth. Different call types — sales, support, collections, appointment-setting — usually need different models, so you can save as many as you like and pick the right one when you score a call.
Technical detail
A model is a reusable bundle of three word/phrase lists (Required, Positive, Negative) plus a numeric weight assigned to each entry. Models can be edited at any time; re-scoring a previously uploaded transcript with a revised model produces an updated score without re-transcribing the audio.
4. Identify Required Words & Phrases, Positive Words, and Negative Words
The scoring engine searches the transcript for every entry in your model's three lists:
- Required — things the agent must say (for example, a greeting, a disclosure, a closing). Missing items are flagged.
- Positive — language you want to encourage (empathy phrases, confirmations, ownership statements). Each match adds to the score.
- Negative — language you want to discourage (dismissive phrases, profanity, prohibited promises). Each match subtracts from the score.
Technical detail
Matching is case-insensitive and ignores most punctuation. Multi-word phrases are matched as a single unit, so "thank you for calling" only counts when the full phrase appears in order. Each occurrence is recorded with its position in the transcript so you can review exactly where in the call the match (or miss) occurred.
5. Calculate the Score
Finally, the engine combines what it found into a single score. Positive matches add their assigned weights, negative matches subtract theirs, and missing required items apply a penalty defined by your model. Because every weight is set by you, two businesses can score the same call differently — and that's the point.
Technical detail
The score is computed as the weighted sum of all positive matches minus the weighted sum of all negative matches, adjusted by penalties for any missing required items. The result is presented alongside the transcript with each contributing word or phrase highlighted, so the score is fully explainable — you can see exactly which entries drove the number up or down.
Frequently Asked Questions
- Who actually scores the call?
- The scoring engine does — automatically, against the model you choose. There is no human auditor reviewing the recording.
- Can I edit my scoring model after I've used it?
- Yes. You can change words, phrases, and weights at any time, and re-score previously uploaded transcripts with the updated model.
- What happens if a required phrase is missing?
- The call is flagged for that requirement and a penalty defined by your model is applied to the final score.
- How accurate is the transcription?
- Accuracy depends on the recording quality. Clean audio with minimal background noise produces the most reliable scores.
- Is my call data kept private?
- Recordings and transcripts are tied to your account and are not shared with other customers. See our Terms and Conditions for details.
Ready to try it on one of your own calls?