#102 – Using AI to predict customer loyalty – A useful NPS benchmark provider quality check

Welcome to the 102nd Customer Strategy newsletter. The five topics this week are:

1. A different approach to the use of AI: Early detection of customer disloyalty
2. A useful quality check if you use double-blind NPS benchmarks
3. Latest blog posts – Deep series on NPS continues
4. Notable items from other sites – Willie Pietersen on strategy – XM Institute research
5. Looking forward

A different approach to the use of AI: Early detection of customer disloyalty

I recently had the privilege of meeting Ilkka Huotelin, founder of Be Customer Smart. (Yes, the red color is deliberate.) He talked me through their fascinating approach to the use of Artificial Intelligence to do several important things that it would be difficult to do well another way.

1. Determine the relationship between customer satisfaction indicators and revenue trends and other financial indicators for your company. The use of AI means there are no preconceived notions about what metrics are most relevant. Everything is put into the mix.
2. Determine the relationship between every single operational metric you may have and customer satisfaction.
3. Determine the financial impact of a KPI improvement.
4. Predict revenue based on current and planned customer and operational metrics.

Ilkka’s core personal area of expertise is in telecommunications, though he and his colleagues have covered quite a broad range of industries. One of the most interesting things he showed me was how their software could analyze changes in mobile phone customer behavior and identify which ones predicted that a customer was likely to switch to another vendor at contract renewal time. Understanding this allows telcos to be proactive, intervening before contract renewal time to try to retain the customer. The use of AI means there is no preconceived notion about which customer behaviors matter, and the model is constantly improving to reflect actual customer actions.

As many of you know, the main area of AI that has interested me up to now is Natural Language Processing. Ilkka’s approach adds a new dimension. If you want to learn more, go to his website at http://www.becustomersmart.com/. Note in passing that I have no financial relationship whatsoever with Ilkka or his company. I simply like what they do. 

A useful quality check if you use double-blind NPS benchmarks

I spoke to the customer experience leader of an LA-based company a few days ago. (Let’s call him Ludwig for the sake of this article and for his love of classical music.) They had just received their first full set of double-blind NPS benchmarks, covering their company and their four main competitors. Ludwig was justifiably proud of the achievement. I found the way Ludwig and the vendor had compared the open text responses to the ‘Why?’ and ‘What could company X do better?’ questions particularly compelling. It certainly showed where each company’s strengths and weaknesses lay. I was surprised by how different improvement suggestions were for each company covered.

The journey to this result was far from painless. Some of this was because the vendor is quite new and had not worked in Ludwig’s industry before. However, the core issue is one that you are extremely likely to come across whenever you use double-blind benchmarking: the companies that provide the service contract in turn with companies that provide panels of representative customers for the relevant industry. Unfortunately, and I base this on my personal experience at HP while IPSOS was the overall service provider, the quality of the panels can be poor.

The issue arises when the benchmark vendor does not do a good enough job in checking the quality of the responses given by panel members. Ludwig and the new vendor found these issues after they provided the initial research results. The vendor had not noticed the problems. Note that panel members are paid, either in cash or with some sort of cash equivalent. If a respondent answers for more than one company, they make more money. This gives rise to fraud. The main areas of cheating I saw and that Ludwig also saw were:

• People who just hit the keyboard randomly to provide text answers. The results were answers like “élakshgéoahsg”. I was stunned that the survey vendors would not see this and insist the panel vendors remove such respondents from their panels.
• People who copied and pasted the same answers into different places in the survey, including supplying identical answers for all vendors.
• People who answered for multiple vendors, even for products or services where it is absolutely impossible for a client company to have more than one supplier.

I could easily give you a much longer list but I find these three items the most obvious. If you are considering benchmark NPS providers or indeed already use such services, I would encourage you to do this basic level of inspection yourselves.

Our latest blog posts

The posts on this list are part of my extensive and deep series about the Net Promoter System. Older posts are still available on the blog page.