As more companies look to compete on customer experience (CX), the number of opportunities to make large, game-changing innovations that set your company apart from the crowd naturally falls. To maintain a competitive advantage, businesses will increasingly need to identify areas for incremental and continuous improvement in CX rather than making regular wholesale changes.
In our previous post we started to look at how companies can make the most of popular CX KPIs and then leverage them more effectively using key driver analysis to prioritise those areas which most require intervention. This method tells us which touchpoints in our business are weakest and by how much, but it does not tell us what we need to do about it. Traditionally, if the problem was large enough, businesses would embark on large scale change programmes (at great expense) to redesign their failing touchpoints. Whilst this might well solve the problem, it is very difficult to see if it has, how well it has solved it (they often create new, unanticipated problems), and how much return was generated by the investment.
Returning to our metaphor of customer experience as a watch, in order to figure out how to move the hands (and not just tell the time), we need to understand the interactions of the cogs beneath the face. In other words, how experiences in different journeys and touchpoints affect how consumers feel. Using applications of machine learning, an unprecedented level of granularity can be reached to understand exactly this.
Painpoints: The cogs at the core of satisfaction
Let’s start by stating that overall satisfaction is the net result of all the positive and negative experiences a customer encountered on their customer journey. In order to build our model, we need first to understand the painpoints, by which we mean specific moments where a customer feels irritated or frustrated which reduces their overall level of satisfaction. Painpoints comprise the complete range of causes of poor satisfaction across a customer journey. For example, within the ‘Researching Information’ touchpoint when applying for a new banking product, painpoints such as ‘Lack of help when I had a question’ or ‘Slow website load speed’ may be identified.
The beauty of key driver analyses is that importance scores are represented as percentages summing to one hundred percent. This allows us to aggregate multiple importance scores into a higher-level category. While traditional methods of key driver analysis are limited in terms of the amount of data we can throw at them, machine learning algorithms such as random forests are ideal for this more data heavy approach. In particular they are:
- scalable both in terms of the number of customers and input variables
- computationally fast and efficient to run
- more accurate than traditional key driver analysis
- not impacted by highly correlated input variables (a core requirement of key driver analysis)
- decomposable, to calculate importance at an individual customer level
Therefore, using a random forest algorithm we can calculate the importance scores for hundreds of possible painpoints and then map the feasible painpoints to touchpoints. This gives us a hierarchical view of the association between touchpoints and painpoints.
While being able to calculate the importance of individual painpoints is highly valuable and a great step forward, we can go further. We can then decompose the resulting importance scores and derive the importance of touchpoints and painpoints for each individual customer.
Armed with this information, we have unlocked the ability to not only explore how individual painpoints have affected a customer’s overall satisfaction, we have also overcome one of the biggest limitations of key driver analysis. Knowing importance at an individual customer level, we can now slice and dice our model, looking at any subset of the population we like, without the need to re-run it. We can then filter the model to examine the CX of a customer segment, across a particular touchpoint, that completed a particular customer journey using a particular channel and having done that, we can then see which painpoints they experienced as well as by how much they impacted overall satisfaction.
Exciting stuff! Whilst we pause for breath, let’s momentarily recap and reflect on our journey so far. We have:
- used our KPIs to identify areas of interest that we would like to explore further
- prioritised which touchpoints within the customer journey require our attention
- quantified how impactful individual painpoints are within each touchpoint
- calculated the importance of each painpoint and touchpoint for each customer
We started with one number and now we have millions! So where do we go from here? We have identified all the underlying cogs in our watch and seen what impact they have on the position of the hands. We can start planning to move the hands.
More than one type of pain
The approach described above calculates the importance of individual painpoints and how frequently they were experienced by our customers. As such, we can simulate what our KPIs would be in the absence of a given painpoint. In other words, ‘how much satisfaction can you yield by resolving a painpoint?’. We call this uplift.
Although, on the face of it, this is the answer we were looking for, we need to hold-off on the victory parade just yet. As a metric, uplift in satisfaction only tells us half the story. Yes, it tells us a lot about some of the most widely experienced CX problems, but we run the risk of not picking up on the less common, and in some cases more extreme CX failures. For example, intuitively we understand that a mobile phone that sets on fire is a catastrophic CX issue. However, if it is quite rare then it will generate a smaller uplift in overall satisfaction compared to a software bug that is experienced by all. Likewise, a minor software bug experienced by all might be an annoyance, but our customers might be prepared to forgive us. Therefore, we need another measure to identify these painpoints that, when experienced, are most commonly associated with overall dissatisfaction.
We can achieve this by determining the overall ‘prevalence’ of a painpoint. We define this as the number of people who experienced the painpoint and were dissatisfied, minus the number of people who experienced the painpoint but were satisfied. In other words – how painful is the painpoint.
Calculating the impact of improvement
Using this understanding of how painpoints affect individual customers, we can start to simulate how resolving painpoints would affect the customer base at an aggregate level. This ultimately allows us to start prioritising individual and combinations of painpoints, allowing for the strategic design of change programmes.
Returning to our previous example, having identified that we want to focus incremental CX improvement within the ‘Customer Services’ touchpoint for our digital customers, we can identify the following group of related painpoints:
- being made to feel like a valued customer
- the length of the process
- experience over the phone
- the time taken to get through to the correct person
Using data collected from KAE’s ‘CX in US Retail Banking’ study, these painpoints were experienced by 22% of customers, of which, 39% were dissatisfied overall. If they were resolved, not only would satisfaction be improved for those customers affected, it would result in net 1% increase in overall satisfaction across the entire customer base. Taken together, one potential action to resolve those painpoints could be to introduce a call-back feature within the telephony service.
Finding clarity in noise
The use of machine learning algorithms has made it possible to drill deeper into the layers of customer experience in a way that was not previously possible. Prioritising those painpoints which have the greatest impact on overall customer satisfaction allows us to generate the greatest possible uplift in customer satisfaction through simple targeted actions. This additional granularity makes it easier for businesses to develop change programmes as well as knowing which customers it will benefit and by how much.
In the next post we will introduce some of the different strategies we can employ to algorithmically prioritise optimal combinations of painpoints. This systematic approach ensures that change programmes are not built on subjectivity or merely listening to customers who shout the loudest.