As the customer experience (CX) monitoring we have discussed in our previous posts becomes more and more complex, many companies will be (justifiably!) concerned about data blindness. In the face of a near-enough infinite combinations of analysable data points, how is anyone supposed to find the one that is right for them? The irony in all this is that the proliferation of improved data can lead people to fall back on to instinct, using the data selectively to ‘prove’ what they believed was true already.
Putting ourselves in this mindset, we considered the following brief:
Within CX, as in life, there is a temptation to prioritise the voice that shouts the loudest, but this is unlikely to be the most effective way to improve your situation.
In our previous post we introduced the analysis of painpoints – specific moments where a customer feels irritated or frustrated in a specific journey – using machine learning. We explored how we can use prevalence and uplift as metrics to measure the impact of resolving different combinations of painpoints.
This third and final post in our series will explore three ways to combine sets of painpoints to build the most effective possible change programmes. We will explicitly calculate the set of things to fix that will deliver the biggest benefit for your business with the minimum of cost and upheaval. To put some meat on the bones of our example, we’ll use a recent data set we developed for US consumer banking journeys.
Maximising satisfaction uplift
The easiest thing we can do is rank the painpoints by satisfaction uplift and select an arbitrary number of painpoints to resolve. In Figure 1 we have done just this and we can see that the incremental increase in satisfaction begins to slow from around painpoint 6 onwards. We could therefore elect that we should use this as our cut-off point for selecting which painpoints to resolve.
Whilst this is a reasonably effective way of maximising satisfaction, it does not consider the cost of resolving different painpoints. For example, some painpoints could require complete systems upgrades costing millions, whereas others could be resolved by tweaking a call centre script. By considering the costs of resolving different painpoints and uplift, businesses can start to understand the RoI of change programmes at much more granular level.
Which painpoints contribute the most to overall dissatisfaction?
Whilst this approach is intuitively simple, there are some downsides to it. Simply looking at painpoints by satisfaction uplift only shows us half the picture. Our previous blog post introduced the concept of prevalence, which is the difference between customers that experience a painpoint but are dissatisfied overall, and those that experience a painpoint but are satisfied overall. This means that if we just look at satisfaction uplift, we may not be resolving those painpoints that are the most painful.
Figure 2 shows two painpoints for consumer banking that have a similar uplift in satisfaction. However, whilst ‘Experience interacting with staff’ has a positive prevalence (when experienced, it is more commonly associated with dissatisfaction, rather than satisfaction), customers experiencing issues with ‘The security measures used’ are much more likely to still be satisfied. This provides an effective way of choosing which one to address first, despite the uplift from each being similar.
The second strategy involves fixing painpoints with the highest prevalance. Doing this is not as straightforward as satisfaction uplift because prevalence is not additive – the prevalence of fixing two painpoints is not the sum of the prevalence of the individual painpoints. To get around this, we can turn to a family of evolutionary optimisation algorithms, known as genetic algorithms. This is an algorithm that is inspired by Darwin’s theory of natural evolution, whereby the fittest individuals are selected for reproduction to produce fitter offspring for the next generation. In our model we go through the following steps to calculate the optimal painpoints:
- Initial population: Each individual or string is formed by a set of genes (painpoints) which are joined together to form a chromosome (solution).
- Fitness score: A fitness score is calculated for each individual. The probability that an individual will be selected for reproduction is based on its fitness score.
- Reproduce strings: Strings are selected to pass on their genes to the next generation. Individuals with higher fitness have a higher probability to be selected for reproduction. The offspring these strings produce contain genes (painpoints) from both parents.
- Termination: The first three steps are repeated numerous times, with each generation having a higher ‘fitness’ than the previous. In our model we repeated this process x times, but it may also be possible to repeat the process until the there is only a minor difference between a new generation and the previous.
Going back to the research, we can look at our example of the journey of ‘Applying for a savings account’. In Table 1, the difference between simply ordering our painpoints from lowest to highest and the top seven painpoints to maximise prevalence calculated from a total of potential combinations.
Combining both strategies
As with strategy one, there are a few drawbacks to this approach. If we just think about prevalence when choosing painpoints, we may be resolving issues that affect only a few customers. For example, we may be resolving a software issue that only affects select customers with specific out of date phones. The approach should be modified to accommodate for businesses that are looking to solve the most impactful painpoints that affect the majority of their customers.
The solution is to consider both prevalence and satisfaction uplift. The genetic algorithm approach above works equally well when combining both measures and allows us to locate painpoints that are most associated with overall dissatisfaction and also affect a lot of our customers.
Putting it all together
This is the final educational blog post on our solution to improve customer experience. We have discussed that simply measuring NPS, even at multiple points along a journey, does not provide sufficient detail or resolution to strategically improve a business. Our approach is highly technical, but the complexity under the surface is designed specifically to provide actionable insights at all levels within a business.
By defining painpoints and mapping these to experiences within a business, KAE has developed the means to spell out precisely what to fix and what to fix first. Further, our knowledge of machine learning allows us to calculate how the changes you make will affect consumers’ experience of your company and how these changes interact with each other. This greater visibility and granularity ultimately wraps up into a single purpose: to help our clients strategically prioritise change.