In a previous article, we talked about the importance of context in understanding speech and how this impacts the quality and accuracy of speech-to-text transcriptions. But having gone to the trouble of extracting context from email messages, there is so much more that can be done with the context than just improve the speech recognition performance.
Like it or not, a computer can only analyse messages a computer can read – which means those messages must be digital. In the case of emails or tweets, the messages are handed to us “on a plate”. With scanned documents and digitised speech, they are not. We need to use techniques such as automatic speech recognition (ASR) and optical character recognition (OCR) to convert that data into text strings suitable for analysis.
However, once this has been done, the resource it provides is enormous. Indeed, it is the reason we encourage our customers to start collecting messages before they know what to do with them and, better still, to go back and gather as much historic data as they can. The cost of storing it is small, yet the value of the history can be enormous.
But what exactly do you do with this data that you can turn it into a great return on your investment?
Well, most companies start by just wanting to find stuff when they need it. Of course, anyone can search their own emails, but what happens to the emails exchanged between their colleagues and outside parties – particularly the crucial ones sent to employees you can’t get hold of or have left the company. These days we store all our communications in our private email folders, yet it’s our experience that more than 90% of these messages are suitable for sharing within the rest of the company.
Finding stuff is very important, but it is only useful if you know what you are looking for. In our company, we operate an email transparency policy. This means we share internally all non-confidential messages – emails and phone calls. If you simply browse those messages not looking for anything in particular, I can almost guarantee I will find out something that I didn’t know and can likely result in a better outcome than had the messages been kept unnecessarily private in personal inboxes.
So if a human can do it, why not a computer. Something as simple as noticing the distribution of words used in messages, can alert you to changes that are happening that you might not otherwise have noticed. These might be new products, customers, issues, anything.
But with the available data we can go much further than that. It is not difficult to correlate the conversations of the staff that are the most productive, with how they interact with customers. The duration of call or length of email is not difficult to measure, but may well correlate (or inversely correlate) with sales performance. Tools, such as sentiment analysis are evolving all the time and can, for example, measure how positive someone is. There are literally hundreds of metrics we can measure, from the time taken to respond, to the collaboration apparent between staff, any one of which can be indicators of successful performance. Once we have the data we don’t need to worry about what aspect of behaviour gives the best performance, we can let the computer tell us that, but only if the data is there and there is enough of it, to see a trend.
So rather like the adage, “we are what we eat”, we behave like we talk, and in the work context, our messages can tell so much. Talking (verbally or in writing) is all about interacting with other people and it is those interactions that decide how successful we might be at doing our job.
Thanks to the cost of storage, “talk” really now is cheap, so hang onto it even if you are not yet ready to let it help you. At some stage you will be and you will want all the history you can get.