Recorded Calls Are the Obligation. Retrieval Is the Real Test.

Storing the call is the part firms have solved. Finding one specific conversation, months later, under a deadline, is the part that quietly fails.

Most regulated firms can say with confidence that client calls are recorded and kept for the required period. Ask the follow-up — how long does it take to produce one particular call and the sentence that matters inside it — and the confidence drops.

That gap is not a storage problem. It is a retrieval problem, and the two are easy to confuse because the same archive satisfies one while failing the other.

What “Retrievable” Means in Practice

Supervisors do not usually ask for an archive. They ask for a call: a named client, an approximate date, a topic. The firm that can produce it in an afternoon and the firm that needs three weeks are both technically compliant on storage. Only one of them looks in control.

The same is true internally. A complaint is easier to resolve fairly when both sides can see what was actually said, and harder when the evidence exists but nobody can reach it before the response deadline.

An Audio Archive Stores. It Does Not Answer.

A complaint arrives. It refers to something the client says was promised on a call in March. Nobody knows which call, and the client is not certain either. What the archive gives you is a list of files filtered by date and number. What the investigation needs is a sentence.

Closing that distance means listening. Audio is the one business format that cannot be skimmed: a thirty-minute call takes thirty minutes to review, and there is no way to know in advance which of the eleven candidate files contains the moment.

 

Where the Cost Actually Lands

The Sampling Problem Nobody Puts in the Risk Register

Manual review sets its own ceiling. If a reviewer can cover a handful of calls a day, the sample stays small regardless of how many calls the firm makes — and a small sample finds only the obvious problems.

The pattern that matters is usually not one bad call. It is the same phrase used by six advisers across four months. That is invisible to sampling and obvious to search.

A Transcript Changes the Unit of Search

With a transcript, the archive stops being a list of files and becomes a body of text. You search the words the client used, land on a timestamp, and play that passage to confirm it. The recording stays the evidence; the transcript is how you find it.

Speaker labels matter more than they first appear. A transcript showing who said what turns “this was discussed on the call” into “the adviser said this, and the client responded with that” — which is the distinction most complaints turn on.

The Multilingual Case Is Worse Than It Looks

Cross-border firms record in one language and review in another. Sending audio to a translator is slow, expensive, and creates another copy of sensitive material in another place. A transcript can be translated in place, and the reviewer works from text while the original recording never moves.

Coverage differs sharply between tools. Some handle a handful of major languages; others, including WhoScribe’s speech recognition, cover ninety or more and translate the finished transcript rather than the audio.

Retention and Erasure Pull in Opposite Directions

Regulation says keep the call. Data protection law gives the individual a right to erasure. The two are reconcilable, but only if you know precisely what you hold, where, and for how long — including any copy held by the processor that produced the transcript.

Settle this before the archive grows rather than after. Ask any transcription provider for its retention period in days and whether deletion reaches backups. Published figures are a good sign; vague reassurance is not.

A Workable Setup

1. Record as you do now. Nothing about the capture layer needs to change.
2. Transcribe on ingestion, not on demand. A backlog transcribed under deadline pressure is the same problem in a new place.
3. Keep speaker labels and timestamps. Without them the transcript answers “was it discussed” but not “who said it”.
4. Index the text and keep the audio as the record. The transcript is the finding aid; the recording is still the evidence.
5. Apply one retention schedule across both, and test an erasure request before you have to honour one.
6. Measure retrieval, not storage. Time how long it takes to produce a named call — that number is the one a supervisor is effectively testing.

Frequently Asked Questions

Is a Transcript Acceptable as the Record on Its Own?

Treat the recording as the record and the transcript as the means of finding and quoting it. Automatic transcription is accurate enough to search reliably; the audio remains the authority.

What Accuracy Is Good Enough for Compliance Search?

Higher than most teams assume for quoting, lower than most assume for finding. Even an imperfect transcript locates the passage; a human confirms the wording from the audio before it is relied on.

How Long Should Transcripts Be Kept Compared With the Audio?

The simplest defensible answer is the same period for both, deleted together. Divergent schedules create a window where one exists without the other, which is hard to explain to either a regulator or a data subject.

Does Transcription Create a New Data Protection Risk?

It creates a new copy, which is a real consideration. Weigh it against the alternative: an archive nobody can search is reviewed less, sampled less and understood less.

What About Accents, Jargon and Product Names?

Domain vocabulary is where automatic transcription is weakest. It rarely affects retrieval — you are searching around the term as much as for it — but confirm exact wording from the audio before quoting.

Can This Run Without Sending Audio to a Third Party?

On-premise and on-device transcription exist and remove the transfer entirely, usually at some cost to accuracy and language coverage. If transfer is acceptable, the questions to ask a provider are retention, sub-processors and whether your audio trains anyone’s model.

WhoScribe is a transcription tool for audio and video, operated by LingoScript Ltd in the United Kingdom (company no. 14794098). It produces speaker-labelled transcripts with timestamps, supports ninety or more spoken languages, and exports to TXT, SRT, VTT, CSV and DOCX.