Predictive analytics: The key to pre-empting customer service problems?

Star gazing
Share this content

As business professionals, we are continually bombarded with messages about the value of data. It is like the new gold rush, with many companies focusing on collecting data in more sophisticated ways from as many different sources as possible.

Sadly, relatively few organisations have worked out what to do with all this information (probably because the job of processing it is overwhelming). The fact is: all this data is useless unless it’s properly analysed and acted upon.

Speech analytics, where customer calls are automatically and retrospectively searched for key words and patterns, is already helping companies figure out where service issues are and to identify trends that can point to the sources of failure demand.

Predictive analytics, however, takes this to a whole new level, and can be used to actually pre-empt customer behaviour. If that sounds a bit science fiction, you're not wrong - there's plenty of science behind it. But far from fiction, using analytics in this way is very much becoming reality and is a tool with such potency that it has issues of morality attached to it. 

What is predictive analytics?

A simple explanation is that predictive analytics uses data extracted from past interactions to identify trends and patterns and use those patterns to predict future behaviour. Predictive analytics expert, Eric Siegel defines it further as: “Drawing on a combination of computer science, statistics, and operations research, predictive analytics reduces fraud and waste, automates manual processes, and drives smarter decisions by extracting actionable insights from the vast quantities of data within [organisations].”

How might it be used?

As Siegel adds: “The most actionable thing an organisation can get out of big data is learning from it how to make predictions per person, because those per person predictions derive the decisions organisations make.”

These individual predictions could be anything from whether or not a customer is likely to switch to a competitor, right through to what incentive would make an individual recommend you to others. Here are a few ways it could be used for customer service:

  • Working out likely customer churn: Predictive analytics uses data from a whole range of sources to work out a likely outcome. If we were to take the customer’s likelihood to leave, for example, the analytics software might examine the number of times that customer has complained, how many interactions they have with the company in a given timeframe, any reviews the customer has written. The predictive model will then compare those results with trend data to work out a score: e.g. that customer is 80% likely to leave within the next year.
  • Predicting levels of stock: A well-known example in the US is that of Walmart, which used analytics to identify that when a storm is approaching, customers stock up on batteries and torches. Nothing surprising there, but what they also noticed was that customers also stocked up on gingerbread. They now know to order in extra gingerbread when a storm is forecast, something they would never have known to do without analytics.
  • Location-specific service predictions: Theresa May (in her previous role as Home Secretary) suggested that predictive analytics could be used by police forces to target locations, communities and individuals that are most at risk of crime. In an increasingly digitised world, she argued, data should be utilised to make communities safer. Similarly, in the world of customer service, predictive analytics could be used by utility companies to identify where and when customers are most at risk of service outages.

Who are the key stakeholders?

Predictive analytics is not going to cut your wage bill. At least initially, it will take a lot of manpower to make it effective: you need board level management deciding exactly what they want to get out of it, experts to define and write the predictive models (there is a lot of science behind this, so you need experts), and people on the ground to tag interaction and data types. Once it is embedded, however, the insights you gain from the analytics will allow you to cut costs in a myriad of ways. Plus, tagging may be able to be done automatically once you have established the key phrases you always want to look for.

Predictive analytics really needs to be delivered by people who understand the science behind it and can use the mathematic and statistical principles to author effective predictive models. Several expert companies have sprung up to help organisations with this, for example WePredict, Tableau and Red Olive.

What are the benefits?

Clearly, the most powerful benefit of predictive analytics is that it provides clear pointers for operational investment. Rather than guessing where to spend time and resources for maximum benefit, organisations suddenly have a much better idea of the likely outcomes of focusing on different areas. Improved interaction analytics can also provide a much clearer view of where automation could have the biggest impact, as well as identifying areas of waste. But there are other benefits, too, including:

  • It provides a useful purpose for historical data: Customer contact centres have almost limitless amounts of data that has been amassed over time. “The more data the better,” asserts Siegel. “The more you have, the more from which there is to learn, to create predictive models that then create the per person individual predictions, which will be more accurate and more precise having been learned over a greater amount of data.”
  • It’s self-improving: The more frequently you use predictive analytics, the more reliable the results. This is because you will be able to check whether the predictions have turned out to be correct. You can then add these findings to your current data set. It means that the system will become smarter every day and the predictions more accurate.
  • It can give you a competitive edge: Interaction analytics software is currently very under-utilised in this country, so any company using it to successfully map customer journeys and identify risks and opportunities are way ahead of the game. Using predictive models on top of this could put you light years ahead of the competition.

What are the pitfalls?

While the benefits are manifold, there are also some pitfalls to avoid, especially at this early stage.

With great power comes great responsibility: Predictive analytics can be a hugely powerful tool, but it does come with a moral question. When does it start being too invasive? For example, if you learned a woman was pregnant, should you use that information to target them with specific messages? There is a great amount of responsibility with regards to individual privacy. Plus, there is a point where customers will find the amount of information you have on them unsettling. “It can get creepy,” says Nicola Millard, head of customer insight and futures at BT. “If my supermarket, or utility company knows more about me than I do, it's like having a strange stalker.”

You may have to incentivise customers to get the data you need. Predictive analytics is heavily reliant on customers agreeing to share information and they are unlikely to do so unless there is something in it for them. This, says Millard, “gives rise to a "me" economy. In other words, customers trading personal data for some advantage - whether it's free stuff, special offers, better service, or faster/ easier experiences.”

You need to have your data in order: To get a full enough picture, you need to be able to analyse data no matter where it has come from within the organisation, because a customer could have used any number of channels to resolve an issue. This usually means you need a single point of access - something many organisations are currently struggling to achieve.

“We need to figure out what data we have on customers, what might be useful to them in terms of telling them things they want to know and they will willingly opt-in to,” says Millard. “But proactive and predictive strategies make things easier and cheaper for both customers and companies when done well - especially in situations such as medical or engineering appointments, where missed slots cost money, and what won't come across as creepy. Ultimately, trust is important here.”

As mentioned above, it’s pointless to simply collect lots and lots of data. But it’s also no good to identify trends and issues and make predictions without actually acting upon them. This is your chance to really impress customers by taking care of a service need before they even know they need it.

About Claudia Thorpe

Claudia Thorpe

Claudia is a writer and expert in the customer service field, having previously held positions as editorial director for Call Centre Focus and director of SpeakEasy Communications.

Replies

Please login or register to join the discussion.

avatar
03rd Jul 2017 12:50

Thanks for article! It's really useful. I am using Big Data in my business and I really impressed with the greatness of this technology.

Thanks (0)
avatar
03rd Jul 2017 18:24

We actually spent some time looking for an easy way to do this for Zingtree. The goal was to take our customer signup data, and come up with a score that would predict if they were to become a paying customer.

Amazon's Machine Learning was the best solution we found, but still not very easy to implement.

We're also interested in adding this capability to our Interactive Decision Tree toolkit.

Thanks (0)
14th Jul 2017 01:56

Thanks Claudia - interesting article. At Thematic we also work on figuring out key drivers, but from the raw data collected using a standard NPS survey (Likelihood to recommend and two open-ends: Why, What can we improve).
http://www.getthematic.com/post/net-promoter-score-analysis/

Thanks (0)
avatar
28th Aug 2017 08:50

Claudia, I totally agree with the tactics you have shared.

Thanks (0)