Written by our Head of Product, Jeremy Hamill-Keays
In the Rhondda Valley, in darkest deepest Wales, there are certain questions which have only one answer. One of them is ‘Who makes the best cawl?’ The answer, of course, is “my Mam”. As you’re probably wondering, cawl is a hearty stew beloved by the Welsh.
With winter drawing near here in Sweden, I started to long for a bowl, so I phoned her up and asked for the recipe. It’s not a secret so here it is; I highly recommend.
- Pot of water
- Chopped up neck of lamb
- Whatever root vegetables are in the cupboard
- Salt/pepper/herbs to taste
I was slightly confused; how could it be so vague? Shouldn’t it be exact weights and measures? My mother scoffed, and told me then it wouldn’t be cawl.
The Need for Training
At Freespee, we work heavily with providing a frictionless user experience including routing calls directly to a destination such as a high value sales queue. Freespee contains many services to optimise sales leads and one of these is predictive routing.
Developments in machine learning are making large impacts and have huge potential to drive customer experience improvements; that gold member needs to be looked after.
While machine learning has many advantages over rules-based systems, it suffers from a couple of major drawbacks. Before you can use a machine learning technique, the system needs to be taught, and obtaining training data can be difficult. It also requires that multiple sources of data are used to reach a level of sophistication required in the modern digital world.
There are generic machine learning engines which claim to do things like call categorisation “out-of-the-box”, however real-world testing often proves that to be inaccurate. They are not adapted to any one customers data set or business.
Good, accurate training set data is essential for commercial use and this is what holds back many machine learning projects. It is important to have multiple sources of data.
Your Recipe is Unique
So, what has this to do with Mam’s cawl?
Well, a little like her recipe, machine learning requires a mixture of ingredients all working in harmony; with the exact ratios changing from company to company, and over time.
Get it right and it works really well, which can make a significant impact on your business.
At Freespee, we have concentrated on ensuring that the data needed for machine learning engine implementation can be easily collected in many ways.
As an example, a speech analytics call categorization service can be trained with data derived from the pre-call Call To Action (CTA) widget, which asks a percentage of visitors what’s the reason for their call.
The CTA improves the visitor experience and is a low friction method of data collection. It can then be combined with data from CRM, IVR, or from “outcome” fields, completed by agents using Freespee softphones such as Freespee Talk and mobile apps.
Each company has unique needs, so we feel this holistic approach is better; technology adapts to people, not the other way around.
So if you are offered a speech analytics solution, ask – how difficult will it be for a training set to be built and does it match your process? How many ingredients or data sets are used to optimise the engine and provide an ability to adapt to an ever changing environment? Or better yet, get in touch with us to find out more.