Call recording metadata fields for quality analysis

Short answer

An audio file shows what was said in a call, but not the call's context: who spoke, on which shift, on which line, whether the call was inbound or outbound, whether it was transferred. Without that information you cannot group agent scores properly, compare agents fairly or join the two legs of a transferred call. The minimum set: a unique call ID, an agent ID, date and time, direction, shift or queue, a transfer link and the audio channel layout.

This article is not about how to configure a PBX. The question is different: which fields should accompany a recording for quality analysis, and what breaks when each is missing.

Why audio alone is not enough

Suppose a hundred call evaluations are ready. A manager asks: "Why is the score lower on the evening shift?" If the calls carry no shift information, the question has no answer. Or: "This agent's score is low" — but if the agent only works the complaints line and that is not recorded, the comparison is unfair.

• Without an agent ID, results cannot be collected per agent
• Without shift and queue, call types cannot be separated
• Without a transfer link, the two legs of a transferred call look like two separate, incomplete calls
• Without the channel layout, the reliability of speaker separation cannot be judged
• Without an outcome code, the link between quality and sales or resolution cannot be examined

Illustrative field map

The map below is an illustrative example. Field names and availability depend on the PBX and the contact center system. For each field: why it is needed and what happens without it.

1. 1Unique call IDThe key that identifies one call across all systems. Without it, avoiding analysing the same call twice and linking the result to the CRM record gets hard.
2. 2Agent IDA stable, non-name identifier. Per-agent results and coaching plans are built on it. Without it, results are matched by hand and errors grow.
3. 3Date, time and durationPeriod comparisons, the link to shifts, selecting calls by length.
4. 4DirectionInbound and outbound calls need different criteria. "Discovering the need" is judged differently on an outbound sales call.
5. 5Shift, line or queueShows the call type and difficulty. The key field for comparing agents fairly.
6. 6Transfer linkThe ID of the call's previous or next leg. Needed to check transfer quality from both sides.
7. 7Channel layoutStereo — agent and customer on separate channels — or mono, both on one. Directly affects how reliable speaker separation is.
8. 8Outcome codeThe result the agent or system gave the call: sale, callback, resolved. Read alongside the quality score, not instead of it.

Why channel layout matters on its own

When the agent and the customer are recorded on separate audio channels, the system can tell who said each line from the channel. When both voices are on one channel, speakers have to be separated by the characteristics of their voices — this is called speaker diarization. Diarization can make mistakes: if the customer's sentence is attributed to the agent, the agent is scored on words they never said.

So if your PBX can record on separate channels, that option is usually more reliable for quality analysis. It depends on what your PBX and recording system support, and before changing it, check storage volume and consent rules as well.

File names and a naming rule

When metadata cannot be passed separately, the simplest route is a fixed rule for file names. The naming rule acts as a bridge between systems.

1. 1Fixed orderFor example: date_time_direction_queue_callID. The same order in every file.
2. 2No personal dataThe customer's name and phone number are not written into the file name. For agents, use an internal code rather than a name.
3. 3Machine-readableUnderscores instead of spaces and special characters; dates in ISO format (2026-09-29).
4. 4A mapping tableCall ID to agent, shift and outcome code is kept in a separate table. Analysis results are joined to it by call ID.

Four analyses metadata makes possible

The value of the fields shows in what they let you ask. The four questions below can only be answered once quality results are joined with metadata:

1. 1By shift and hourAt which hours is "the need was clarified" missed most often? The answer is often the peak hours — agents cut corners as the shift wears on.
2. 2By queue and call typeOn which line are critical criteria broken most often? That shows which topic coaching or the knowledge base should target.
3. 3New versus experienced agentsJoined with start dates, it shows which steps are hard in the first weeks — direct material for an onboarding programme.
4. 4Crossed with the outcome codeOn calls where "a next step was agreed" is met, how often does a callback status appear? That shows association, not cause — but it helps check whether the criterion is meaningful.

Data quality: typical problems

• The agent ID shows the agent at the start of the call, not the one after a transfer
• The queue group changed, but older calls were stored under the new name
• Time zone mismatch: the PBX writes UTC, the CRM writes local time
• Very short calls — a few seconds, misdialled numbers — spoil the sample and should be filtered out first
• The same call was exported twice and analysed twice

Start with the minimum

Passing every field at once often delays a project by weeks. A practical order:

1. 1Stage oneCall ID, date and time, agent ID. Without these, results cannot be tied to anyone.
2. 2Stage twoDirection and queue. They let you separate call types and compare fairly.
3. 3Stage threeTransfer link, outcome code, channel layout. For deeper analysis.

Limits

• Metadata does not replace the call's content: an outcome code of "sale" does not prove the call was a good one
• A transcript does not confirm that a CRM record was written or an action was taken
• Aggregating results by agent ID without regard to call count and call type can create an unfair ranking
• Metadata can be personal data too; access and retention rules apply to it

How it works in Vexvon Audio Analyzer

Vexvon Audio Analyzer takes recordings as files; it has no automatic PBX import and no separate metadata fields such as agent, queue or shift. The mapping table above is therefore kept on the company's side. What the analysis does include:

• The file name is kept as the call's title and can be changed later — so a naming rule shows up directly in the list
• In a stereo recording the sides are separated by channel, in mono by diarization; a line the system is not sure about is kept as "unknown"
• The call's language is detected automatically
• Every call gets the call standard's steps, comments and transcript lines, and a 0–100 score

More: Vexvon Audio Analyzer. Fields written to the CRM after a call are a separate topic — call CRM fields. How to use these fields in a pilot is shown in the speech analytics pilot plan.

First step

Export one week's call list from your PBX and check it against the map above: which fields exist, which are missing, which are filled in wrongly. Then write your file-naming rule in one sentence. This article belongs to the audio analysis implementation and reliability section. To try it on your own recordings, get in touch.

Further reading on this topic: fair agent performance comparison, call transfer quality.