Measuring Customer Loyalty using NPS
Customer loyalty has to be one of the key priorities of all businesses. Without a loyal and profitable customer base, most businesses will struggle to survive. So companies need a way to keep tabs on customer loyalty and do something to improve it. Undoubtedly, the most successful loyalty measurement system ever created is the "Net Promoter Score". This popular and deceptively simple measurement system first appeared in 2003 and has now reached near ubiquity having been adopted by more that two thirds of Fortune 1000 companies. In this article we'll take a look at some of the reasons for the phenomenal success of the technique, how it works and, of course, where coding can play a part.
It's fairly obvious that happy, loyal customers are more profitable in several ways. They will usually buy more stuff, make more repeat purchases, be more price insensitive and so on. But, as the saying goes: "you can't manage what you can't measure", so, it's important for companies to measure and track how their customers feel about them. In December 2003, Frederick F. Reichheld published an article in the Harvard Business Review suggesting an alternative to traditional, long, complicated customer satisfaction surveys. In collaboration with Bain & Company and Satmetrix, they set out to find a single, simple question that would serve as effective measure of customer loyalty.
Following this research, and various experiments, the question "How likely would you be to recommend (Product/Service X) to a friend or colleague?" was developed. This question has become so commonplace, it almost seems strange that companies used to simply ask people more direct questions like "How satisfied are you?" or "How likely are you to purchase from us again?".
So why, bother with this oblique approach? It turns out, likelihood to recommend offers up a subtly high bar. In order for a person to make an emphatic recommendation, two things need to happen. Firstly, they must believe that it offers something genuinely superior (e.g. price, features, quality etc.). Secondly, they must feel good about their relationship with the company in question. It stands to reason, that any company that is meeting both of these conditions, manifested as recommendations, must be doing something right. It follows then, that the long term prospects for such companies is likely to be fairly rosy.
It is for these reasons, that the original researchers managed to show a strong link between likelihood to recommend and the future growth of a company.
Hang on a minute, surely there are other ways to achieve growth other than through glowing customer recommendations?
In the short-term, yes there are. Advertising, special offers or simple entrapment are some of the ways you might achieve this - but it's unlikely to be sustainable in the long term if your customers are not happy. Consider the famous example of AOL, which during the mid-1990's widely distributed their software on free CDs (remember those?) in magazines, checkout stands and via direct mail. They managed to grow their customer base by millions, but provided such bad service that customers became frustrated and disillusioned. Furthermore, they gained a reputation for making it very hard for customers to cancel the service leading to a huge amount of bad press and negative feeling. Contrast this with, Amazon, for example. Amazon spend, relatively very little on advertising and gimmicks, choosing instead to invest heavily in infrastructure, service and delivery. The result is customers who are genuinely satisfied, loyal and over the long term has made Amazon one of the biggest companies in the world. Whilst it may be possible to make short-term "bad profits" from unhappy, disgruntled customers, long-term growth an profitability comes through "good profits" generated from customers happy enough to recommend a product or service to others.
When we ask the someone the NPS question, we find out (on a scale of 0 to 10) whether they are a Promoter (score 10 or 9) a Detractor (score 0 - 6) or a Passive (score 7 or 8). In order to calculate the overall NPS score for a company, you subtract the percentage of people who are Detractors from the percentage who are Promoters. In this way, the overall happiness of a companies customers can be neatly summarised on a simple scale from a maximum of 100 through to a minimum of -100. The simplicity of this score is one of its key strengths. It's an simple, understandable metric that the whole company can get behind. The message: "increase this number and the company will grow" is compelling, intuitive and easy to sell in to staff.
OK - you've set up your NPS survey, you've found out your NPS score is -12 (not good), what can you do to increase it?
The NPS approach is fundamentally about organisational change which is another key ingredient of its success. But to affect that change, you need to know "why" your customers are unhappy. One way to find out, is to get in touch with Detractors and ask them. Many NPS systems will feedback negative scores to frontline staff so they can make contact and discuss any issues at hand. This is a great way of instilling a culture of positive organisational change and customer focus.
Another approach is to ask a follow up question in the survey itself asking why the person gave the rating they did. This allows them, in their own words, to explain the reasons why the company is great and they rated it a 10, or why they're unhappy and rated it a zero. Clearly, this kind of feedback is invaluable for guiding change within the organisation at all levels. If the volume of feedback is sufficiently small, it may be reported/disseminated in its raw form - reading comments verbatim in this way can give staff a unique insight and connection to customers. However, if the volume of feedback is relatively high, this is impractical and we need to look for tools to help us summarise this type of data for analysis.
There are a number approaches we might want to consider to tackle this challenge. Our software "Codeit" supports all of these approaches - and there isn't a right or wrong option. The approach a company chooses depends on their specific constraints and goals, as well as the trade-offs they're willing to make.
It's common for companies to use automated tools to search NPS feedback for keyword or phrases to look for patterns and issues that require attention. The key advantage of this approach is it is fast and completely automatic. The Text Analytics available in Codeit can extract key phrases from thousands of verbatims in a matter of seconds. If the data contains some big, frequently occurring topics (e.g. "slow internet", "bad service" etc..) then these will naturally surface and rise to the top. The downside of this approach is that smaller, subtler issues may be missed - especially if they are expressed in lots of varying ways by customers - or drowned out by "noise" in the data.
To help understand this, think of a wordcloud. These are great at bringing the big, major themes to the fore, but usually these are surrounded by a sea of other smaller keywords which act as general noise around these main themes.
Whilst keyword analysis presumes no prior knowledge of the data or it's contents, an alternative approach is to start with a list of expected themes and phrases that you expect to find in the data. It's then possible to build a set of "rules" which can be applied to the data in search of expected themes. For example, you might create a rule such as: "(bad OR poor OR rude) AND (service OR staff)" to try and pick up on negative feedback around staff service. When new feedback is is collected, the performance of the rules can be reviewed and tweaked if necessary - for example, if a new theme arises in the data, then new rules can be added to capture this. The key advantage of this approach is that also automatic (once the rules have been created) and deterministic. The main disadvantage comes from the overhead and complexity of managing the set of rules in place.
This is the traditional approach to dealing with verbatim data - we pay coding staff to read through the customer comments looking for themes and categorising the data accordingly. A tool like Codeit makes this task as simple and efficient as possible, but there is still an inevitable cost associated with the manual labour involved. The upside, in return for this cost, is very high quality, actionable data. Given the importance and value of customer feedback outlined above, this cost is usually considered a worthwhile investment.
Lastly, a key feature of NPS programs is they tend to be on-going. It wouldn't make sense to measure your levels of customer loyalty once, then never do it again. This is something that needs to be continuously monitored. Therefore, if we take any of the approaches above, we find ourselves with a set of classified data gradually increasing over time. This provides excellent training data to teach a machine learning system, such as Codeit, to how feedback should be categorised into themes. Over time, the system can then automate a large part of the categorisation leading to major time and cost savings. The important thing to note about this technique is that we're not seeking to completely eliminate human involvement - NPS is fundamentally about keeping people in touch with feedback, after all. Instead, we're looking to the machines to take the burden of the repetitive work out of the hands of people, leaving them free to concentrate on the more detailed, items where new themes are emerging.
Try it for Free
Anything we can help you with? Ask us