How does Callyst score a call?
Short answer. Callyst transcribes your recording and scores the call on your own scorecard, as you wrote it. The call gets an overall score with a short written reason. Each item gets a score, with a quote from the call when one checks out. Each flag the AI raises comes with a word-for-word quote. A team of any size can try it free on up to 10 of its own calls.
What do you send to Callyst?
You send recordings, not transcripts. Upload .wav or .mp3 files, up to 50 MB each. For the free check, that is up to 10 calls on your private upload page. Callyst works with any phone system that can export recordings. Teams already using Callyst can drop recordings on our upload page or send them through our API.
Callyst scores calls in English. It sets aside voicemails, no-answers, exact duplicates, calls under 60 seconds (by default) and calls over 90 minutes, so those are not scored.
Step by step, see how to upload call recordings to Callyst. To get the files out of your phone system first, start at Use Callyst with your phone system.
How is your scorecard loaded?
Callyst loads your scorecard as written, in your words. On the free-check form, attach it as a JSON file (up to 200 KB). Or give Callyst a checklist (a PDF, DOCX or TXT file up to 2 MB, or up to 20,000 pasted characters) and it turns it into a draft scorecard for you to read. If you send neither, Callyst uses its standard sales scorecard.
A scorecard holds 1 to 20 items and up to 12 flags, on one whole-number scale such as 1 to 10. Each item is a score on that scale or a yes/no. It applies to all calls, inbound only or outbound only, and it carries your guidance on what a good call looks like. Each flag has a severity: critical, high, medium or low. The overall score has its own guidance too.
Here are two pieces of a made-up scorecard file, one item and one flag:
{
"id": "greeting",
"label": "Greeting",
"type": "scale",
"applies_to": "all",
"guidance": "Did the rep give their name and the company, and ask how they could help?"
}{
"code": "promised_refund",
"severity": "high",
"guidance": "The rep promised a refund or discount they are not allowed to offer."
}The whole format, with a complete example, is in how to use your own scorecard in Callyst. Don't have a JSON file? The free scorecard builder drafts a scorecard from your checklist, with no sign-up.
If your file doesn't fit the format or the limits, the form tells you what to fix, and your free check doesn't start until you do.
How is the call transcribed and scored?
Callyst transcribes the audio itself, on Cloudflare Workers AI. An AI model then scores the transcript against your scorecard. Our AI model provider gets text only, such as the transcript and your scorecard, never the audio.
What do item scores and reasons look like?
The overall score is on the same scale as your items. It comes with a short written reason, not a quote. It is judged against the overall guidance in your scorecard. It is not added up from the items.
Each item gets a whole-number score on your scale, or a yes or no. For each item it scores, the AI is asked for a quote from the call: 25 words or fewer, word for word. Item scores come with a quote from the call. If no quote checks out, the score shows without one. On a call's page in the Callyst app, that row reads "no verified evidence".
Callyst marks an item N/A when it doesn't apply to the call or there was nothing to judge. N/A items are left out of the averages. An item you set to inbound only is marked N/A on outbound calls, and the other way round. A call that is not a real two-way conversation, such as a recorded menu or a call where only one side is heard, gets every item N/A and no overall score.
Here is a simplified, made-up example of a call's page in the Callyst app. It is not a real call, and the real page looks different.
| Part of the call | Score | What backs it up |
|---|---|---|
| Overall | 8 out of 10 | Reason: a clear opening, and the rep moved a ready caller towards a booked next step. |
| Greeting (scale, 1 to 10) | 8 | Quote: "Thanks for calling Example Co, this is Sam. How can I help today?" |
| Booked a next step (yes/no) | Yes | No quote checked out, so the row reads "no verified evidence". |
What are flags, and why do they quote the call?
Flags are a separate list for a person to review. Raising a flag does not change the call's score. The behaviour behind a flag can still show in an item score when your scorecard has an item for it.
Each flag the AI raises comes with a quote from the call, word for word. A flag whose quote isn't in the transcript is dropped. A critical flag's quote must come from the rep or the customer, never a recorded prompt. Callyst may also add a low-severity "needs review" flag, with no quote, when a call's scores and notes don't agree, so a person can take a look.
Made-up example, same call as above: a high-severity flag, "Promised a refund or discount", with the quote "I'll refund that in full today, no problem." The call's overall score stays at 8.
On the flags page, each flag shows its severity, the rep, the time, the quote and a player for the call audio (until the audio is deleted). An admin marks each flag reviewed or dismissed, and Callyst keeps who did it and when.
How are quotes checked?
Each quote is checked word for word against the call's transcript. Case, punctuation and spacing may differ, and "..." may join pieces that come in order. A quote that is over 25 words or can't be found is removed. The item keeps its score, without the quote. A flag whose quote can't be found is dropped.
Callyst checks every quote again before the report is sent, and removes any quote that fails.
What does the report show?
When every file is uploaded, press Done, build my report. Your scored report arrives within 24 hours. It comes by email to the person who signed up, with the headline numbers and a link to the full report online. Sign in with the work email you used.
The report's full-call section lists each scored item with a one-sentence reason. Quotes appear on flags. To keep a copy as a PDF, open the report page and use your browser's Print, then Save as PDF. The sample report, written for a made-up company, shows the layout.
In the Callyst app, each call also has its own page: the overall score and reason, a coaching note, every item's score with a quote when one checks out, strengths, fixes with what to say instead, flags, the audio player and the transcript. Features shows the app pages.
FAQ
What if no quote checks out for an item?
The score stays and shows without a quote. On the call's page the row reads "no verified evidence". The overall score never has a quote: it has a short written reason.
Does a flag change a call's score?
No. Flags are a separate list for a person to review, and raising one does not change the call's score. The behaviour behind a flag can still show in an item score when your scorecard has an item for it.
Should I act on a score or flag straight away?
Review it first. Scores, flags and coaching notes are generated by AI and can be wrong. They are an aid, not a decision, as our Terms say. Review them before acting, especially before any step that affects a person's job or pay. The call's page has the transcript and, until it is deleted, the audio, so you can check a quote yourself. To compare Callyst with your own managers' scores, read how to check AI call scores against your managers.
What happens to my audio?
Audio is deleted 30 days after scoring, and no recording is kept longer than 35 days. Our AI model provider gets text only, such as the transcript and your scorecard, never the audio. Your data is not used to train AI models. The privacy policy has the detail.
Try it on your own calls
The free check scores up to 10 of your own calls and sends a scored report within 24 hours. Fill in a short form (name, work email and company). There is no card and no call with sales first, and it doesn't commit you to buy. Start the free check or see a sample report. For plans, see pricing.