There are more negative stories around relating to AI chatbots & voicebots than positive ones it seems. Maybe this is because the disaster stories are more interesting and chatbot miss-steps are always entertaining. Hostcomm has worked with AI, chatbots and voice bots for several years and it can confidently say it hasn’t been easy, however the new large language models (LLMs) that we have been introduced to in 2023 are, in our opinion, set to transform contact centres and customer service. So in this article we’re summarising some of the ways you get on the AI/LLM bandwagon in a quick but controlled way, and start to see tangible benefits and an actual ROI.
Call recording transcription
Telephone call recordings hold a great deal of valuable information relating to customer satisfaction, staff performance and issues relating to your services. The call recording files however are of limited use because they have to be listened to, which at normal speed takes a long time. Extracting information, such as a complaint or accidental data privacy breach is very time consuming because the incident has to be found on a call recording. This means that most organisations do not make best use of their call recordings because it is simply not practical to do so. By converting the call recording into a text file, using a transcription process, the information is released because it can be processed by artificial intelligence systems which can apply natural language processing techniques to it. AI has progressed very quickly over the last two years and call transcription accuracy has become more accurate and cheaper. It is now possible to transcribe calls for a fraction of the cost and perform task such an identifying speakers (diarisation), spotting keywords and redaction, with precision.
Call recording transcription is the first step towards using AI in customer service, if your organisations uses a telephone system or contact centre.
Auto-triage of messages
In the same way that call recordings ‘lock in’ valuable information, emails also do albeit they are in text form which means they can be processed by AI much more easily. It is very time consuming and inefficient to read through hundreds of customer service emails sequentially to ascertain a priority and then respond to them. This is where a large language model (LLM) can help because it is bad to read and understand the contents of an email very quickly and cost effectively. By setting up some simple training examples it can read emails and ‘auto-triage’ them into a priority order which saves a huge amount of time especially during very busy periods. This can be set up by confirming your emails to forward to a ‘collector’ which feeds the emails through a language model.
Message response automation
New applications like ChatGPT have shown us that LLMs are very good producing text derived from their immense neural networks. This can be applied to text message replies in order to speed up the response process. By suggesting the wording for a message reply which can then be edited by a person the time to response can be reduced considerably. The suggestion can be produced using previous replies so that consistency is maintained and time is reduced further.
This feature is available on many applications so is not particularly new however the next logical step is to configure the LLM to automatically respond to certain categories of customer service message without involving a person at all. By using a ‘crawl, walk, run’ approach AI can be introduced into the customer service system slowly and in a controlled way.
Below: Diagram showing how Hostcomm Interactive Analytics works.
Staff management
Staff performance can also be analysed through the same textual interactions using the same LLM system because it understands everything that is contained in the messages. Modern AI can distil an almost infinite amount of insights from textual conversations, some of which are useful for staff management such as:
- - The person’s mood and general attitude.
- - The professionalism of their language eg do they use slang or profanities?
- - listening skills.
- - Empathy
- - Developing rapport and using their personality.
- - Do they attempt objectives such as up-selling, marketing opt-in?
- - Their general score based on numerous preset factors.
Being able to extract this information on a call by call basis enables them to be better managed, which leads to increased professionalism and productivity. The customer experience will improve if the staff are receiving more frequent and accurate feedback. Hostcomm’s Interaction Analytics service is able to perform all of these tasks automatically for every interactions according to a set of objectives.
Business Intelligence reports from natural language instructions
Contact centres and telephone systems generate vast amounts of data which, as previously mentioned, contain valuable data which has traditionally been difficult to exploit. SQL statements would be used or an SQL client application that may help you produce reports. Producing good reports is technically difficult and costly because some Business Intelligence applications are very expensive and require coding. With the emergence of new AI and LLMs it is now possible to query your data directly using natural language. This is possible because the LLM converts your statement, such as “produce a heat map of the UK showing where geographically our complaints come from” into something that the data fetching algorithm can understand, whether the data is stored in a mysql database or AWS S3 bucket. It also means that anyone can query your data, they don’t have to be a technical expert, they just need to be able to describe what they want.
Here are two great examples of applications which can be used in this way:
- Noteable.io
- ChatGPT Code Interpreter (currently in Beta)
Quality Assurance (QA) & compliance monitoring
Traditional Quality Assurance (QA) typically involves gathering a set of 'objectives' that are checked off using a small subset of customer service interactions. This process can be likened to scooping a water sample from a flowing river - it captures just a snapshot of the whole picture. Other options include outsourcing QA where interaction data is released to a third party for analysis, this can be costly and time consuming and there is always the heightened risk of a data breach.
AI systems such as Interaction Analytics can now accurately confirm whether QA objectives have been achieved for every interaction in either real time or semi-real time across all communications channels. This is going to be exceptionally useful for any organisations that are regulated by bodies such as the FCA, which is introducing the FCA Consumer Duty which comes into force in the 31st July 2023. At long last there is a cost effective service which can tell you everything you need to know about every single customer interaction, and then classify, prioritise and alert when necessary.
Summary
AI has created a lot of interest and activity, much like NFTs and Crypto Currencies previously and in the same way the potential to waste money is also there. Our above suggestions are all free or pay-as-you go commercial options, some of which are available from Hostcomm. Our advice is to proceed carefully and in a controlled way, this doesn’t mean you have to move slowly. You can start easily with call transcription, then centralise your message storage and then move to analytics. This gives you a good foundation to take the next step - AI Automation and beyond.
