One of the main promises of Customer Experience (CX) Programs is their ability to help you save at risk customers.
In its simplest form, it works like this:
- Customer has a bad experience and is thinking of leaving
- Customer receives survey and vents their frustration
- CX software triggers an email alert
- Company “closes the loop” with the unhappy customer via a phone call, seeks to understand and resolve their issue
- Customer’s issue is resolved and furthermore they are impressed with the efforts of your company and become a more loyal customer (and in some cases will even tell a friend or two about your great service!)
Beyond this simple “closed loop” approach, many companies have gone further using statistical analytics to predict churn risk or lifetime value based on a customer’s survey score.
We see many of our clients performing this financial linkage analysis and demonstrating the link between survey scores and financial KPIs for their customers, for example:
- Detractors are 3x more likely than promoters to churn
- Customers scoring 10 on average will buy 3 additional products from our site, vs those scoring 0-6 who only by 1
- The lifetime value of a promoter is 120% higher than a detractor
And many more
Survey Scores as Predictors
The benefit from here, is that you can in theory assign a “likelihood” to churn score for each customer (similarly you can design predictive metrics for repeat purchases, lifetime value or other metrics).
Using this score, you can then design strategies to proactively reach out to customers with higher probabilities of churning with actions and offers which may ultimately save the relationship!
However, this focus solely on survey scores as predictors has its limitations – as a significant majority of your customers will likely never have survey scores you can base predictions on. This happens for a few reasons:
Firstly, most survey programs only sample a small % of your customer base. This could be due to customers not providing contact information or just gaps in your internal data systems.
Then depending on industry and survey type, response rates can be low, meaning even customers who are sampled may never respond, to be assigned a score in your model.
Many expect this to get worse as we gradually see “The Death of the Survey” with customers preferring to provide unstructured feedback, in the moment through voice or video channels, rather than responses confined to score based questions.
Including Additional Sources
How do we build a model which future-proofs against these factors? The key is to consider survey scores as one factor in predicting behaviour, but also look at what additional data could be included in a predictive model.
You are likely to have a wealth of different data sources which could be included in a predictive model, additionally you don’t need to limit yourself to just one (in fact multivariate models can provide greater accuracy).
For example in the retail world you could consider: transaction history, average transaction size, loyalty club membership, interactions online or with your mobile app.
Whereas in software, factors such as: product usage, webinars attended or meetings with account managers accepted, change in annual spend and support cases, can all provide insights.
Building a Model
At Higher Oak, we adopt a 6 step approach working with you and your analytics team to build a model to predict customer behavior.
1. Select Outcome KPI
Start by choosing the metric which you are trying to predict, this should be a key business KPI for which you internally have data available for a significant proportion of your customer base.
Examples include: Customer Lifetime Value, Retention, Average Order Size, Repeat Orders, Upsell, Cross Sell
2. Look for Existing Analysis
Before you reinvent the wheel, check what already exists. Analytics teams will often conduct similar analysis which they can provide as a starting point. Additionally they will be familiar with the datasets available and potential gaps, to help steer your approach.
3. Identify Data Sources
You then have to determine what data is actually available as inputs to your model (hint: it won’t always be the full list you’d like!) You can start with a list of desired fields and then mark where there are gaps, or ask your data team for a list of what is available in your main systems and highlight what could be useful.
Potential sources include: CRM, CDP, Transactional Systems, Loyalty Programs, Marketing Tools and Databases.
4. Perform Correlation Analysis
Then it’s time to start the analytics project. The analysis itself shouldn’t be too complex, however securing the resources to start your project is likely your main battle!
It may take a few attempts to find factors which influence your goal, but eventually it should identify some links between behaviour and your outcome KPI which you can then target. For example:
5 and 6. Data Integration and Orchestrate / Automate
The final two steps are linked. Based on the analysis you need to determine what action you want to take.
Perhaps the analysis shows that customers who shop through your app are of higher value than those who don’t – in this case could you send an email to incentivize downloading the app to those not already using it. Or use a digital personalization tool to customize your website landing page encouraging non-app users to download.
Or in software, you find that customers attending 1 webinar a quarter are less likely to churn. Can you add the non-attendees to a marketing campaign based around the benefits of webinars. Or have an alert in your CRM system for the account manager to reach out directly to encourage attendance.
There are many different tools available to help with this “orchestration” from marketing automation tools to website personalization. It is likely you already have a few in your stack which can be utilized.
Once you know your strategy and the tools involved, you can embark on the final steps of integrating data and configuring the proactive actions in your chosen tool.
Conclusion
From here, you hopefully start seeing some positive results, with at-risk customers being nudged in the right direction by these automated interventions, without being solely reliant on survey scores to predict the outcome.
As a final point, in addition to this orchestration, your analysis can be used internally to design more traditional, longer term initiatives to improve service and increase customer satisfaction.
Reach out to us below to learn more about how companies are adopting this proactive approach to customer experience management.