Ciara Ferrier

Broadly speaking, there are two "arms" of transcription

The creation of letters or reports has seen a definite evolution

It can be argued that rather than replacing secretaries and administrators, the advent of recording devices caused an evolution in their skills. I have never had to learn shorthand, however I am a skilled audiotypist.

We are seeing a similar situation now where, instead of audiotyping dictation, secretaries and administrators are proofing and editing the output of AI Scribes, writing prompts and training the technology to produce the letters and reports.

The evolution in our skillset follows a development in technology that improves the workflow of our clients.

Products such as Heidi AI are now able to populate clinical notes within a consultation so reliably that they give clinicians more time to focus on their patients, facilitate the smooth running of clinics and reduce paperwork outside of the consultation.

Assessing workflow and cost-efficiency - where is your data?!

As a note, we would still carefully consider cost-benefits and workflow when a client considers adopting an AI Scribe. Payment is by subscription and there are different plan levels offering different levels of functionality. However, the plan that is right for one client may not be right for another and may cost them more when considered over a full year.

Before adopting an AI solution, you should consider where your data is being processed, where it is being stored and also who can access it? It is essential to comply with GDPR and you may wish to protect your intellectual property (IP). Look for certification and compliance standards of any product.

Transcription of recorded conversations has not evolved to give the same level of results

AI Scribes do not need to work verbatim. They need to be fed very good prompts and given very good templates that tell them exactly which gaps we want them to fill. They can then pick up high level information and insert it into those gaps.

They are identifying facts, looking for what was said rather than how it was said. They have been given context already - a consultation - and as part of that, the clinician will also be asking questions of the patient looking to definitively establish facts on which they will base their decision making and advice.

AI transcription of recorded conversations is still disappointing, if what you want is a professionally-formatted, accurate representation of a conversation that does not require additional work on your part.

It does not cope well with:

• Strong accents
• Poor quality audio
• Place names
• People's names
• Acronyms
• Technical vocabulary.

It does not listen for the quiet person in a focus group trying to interject comments and it does not sensitively add punctuation - "Let's eat, Grandma" versus "Let's eat Grandma ." Essentially, it lacks emotional intelligence and so you lose the humanity from a conversation...between humans!

You can use this output as notes to refer back to, or from which to create meeting minutes, but you cannot give it a blank canvas and expect a perfect result.

Human-ear transcription is both evolving...and not!

So while the skillset for creating letters and reports is evolving in an exciting way, very best results for transcription of recorded conversations are still achieved by human-ear transcription.

Having previously worked in medical software development, I am genuinely impressed by some of the AI Scribes. Writing prompts and templates is like writing code - with actual words!

But, for now, I still staunchly defend human-ear transcription too. Providing you with an accurate, professional transcript takes time, skill and effort.

I do see projects using AI transcription for interviews where just notes is sufficient and then sending more detailed interviews to me for a full transcript but this needs to be done with care. The cost of AI transcription is attractive until it doesn't deliver what is needed. At this point, it becomes a costly mistake.

More expensive for a reason

This is the difficult bit. Having realised that you need a human-ear transcript, you also need to then understand you are paying for someone's skill and time.

Scissors and hair clippers can be purchased for about the cost of one haircut. But if you cut your hair yourself, could you expect a hairdresser to charge you less than their usual rate to put it right? You haven't helped them by cutting some of the hair off and have probably made their job harder.

Pharmacies now stock temporary self-fillings for teeth but could you expect a dentist to charge you less than their usual rate for a permanent solution?



Plan your project!

So, researchers, please plan transcription into your project carefully. All too often, I am asked to reduce my rate because teams have not understood the capabilities of AI transcription and have not budgeted enough for human-ear transcription. I simply cannot be expected to absorb the cost of that mistake.

Final thoughts

Research, understand the capabilities and plan the role in your workflow of an AI Scribe or AI transcription solution.

AI Scribes are driving an evolution of skills in terms of letter and report creation. Human-ear transcription still gives very best results for qualitative research analysis. 

Plan your budget. Plan your project.