There is little doubt that a large number of companies would find it useful to have their telephone calls transcribed. Just  consider how many people routinely search their email - why wouldn't they want to search their telephone calls?

30 years ago, finding correspondence would involve physically rummaging through filing cabinets. The advent of digital communications has meant we can now search correspondence instantly. Yet why should employees need to search historic emails and not historic phone calls? The reason that most companies do not transcribe their calls is because they think that either machines are not accurate enough, or the process is too expensive. This perception is not helped by user experience with digital assistants such as Alexa and Siri, or commercial BOTs typically provided by financial institutions to avoid paying for human operators.

However, the technology and human factors design to which the business and retail markets are exposed are by no means representative of the current state of the art. Machine transcription of spoken words can be as good, if not better than that of a human. Where it often fails is in the understanding of speech and without context, the human is no better. A machine will typically cost one tenth the cost of a human, and produce a transcription at least as fast as a human can speak.

But performance and cost aside, most companies who transcribe telephone calls only do so to investigate when things go wrong - for example in the event of a dispute with a customer. Given that disputes should not be the norm, there seems little point in transcribing all calls speculatively. Therefore, when call transcription is invoked, it is done on an ad hoc basis once the call recording has been retrieved and, given a company may be involved in hundreds of calls each day, finding a specific call can be very time consuming. It relies on knowing the exact time and date a call was made. Imagine that the only way to retrieve a historic email was to type in the exact date and time the email was sent or received? It is the fact users can search for keywords in emails that enables them to narrow down the search to something feasible, if not unique. To do that requires that every call be transcribed.

5 years ago, machine-transcribing every call would be neither cost effective nor accurate enough to be worth doing. Now with word recognition rates as good as humans and transcription costs less than £3 per hour, all that has changed. However, to benefit from machine call transcription, workflow has to change to match improvements in the underlying process of speech recognition.

Identifying individual phone calls then submitting them to a transcription service in an ad hoc manner is doomed to failure. Not only is there the problem of extracting the digitised speech for transcription, what is one to do with the transcribed speech once extracted? Many companies now run private branch exchanges (PBXs) either on-site or in the Cloud. Simply finding a call from a sequence of transfers is difficult enough, but ensuring it is recorded at the highest quality is just as important. Then, once extracted the recording needs to be indexed in a database so that it can be easily located in the future along with all its metadata such as party names, numbers, dates, times, etc. In short, ad hoc call transcription will be found wanting in every respect. Automatic call transcription is the only solution. In other words, the process of extracting calls, transcribing and storing them needs to be done automatically and companies which do this manually on an ad hoc basis will soon see their investment wasted.

Of course, the algorithms used to do the transcription are very important, but without the infrastructure to process them, these algorithms are useless. And as the number of employees using a single phone system increases, the process of isolating conversations and delivering specific transcriptions to the right staff can be as complex as the speech recognition process itself.

But there are many reasons to transcribe calls - even when you do not anticipate needing the transcription to resolve a dispute. Simply making notes on the outcome of a call is an essential practice in many professions and will often take as much time as the call itself. And the more the staff member is paid, the greater the overhead. But just like emails, telephone calls define how a company operates. The information contained in those calls is pure gold. It can tell about the way third parties are dealt with and the issues that they have. It can provide vital real-time information about how a company works without subjecting employees and third parties to surveys and questionnaires that become almost immediately out of date. The exponential growth of artificial intelligence means any company, no matter how small, can extract information that was previously totally out of reach.

That is what automatic call transcription is really about - discovering how an organisation is operating and how, using its own communications data, it can improve.