Customer Retention: You already have enough segmentation, take action!

May 31, 2011 By 1 Comment

Think quick: to drive customer retention should you focus on a deeper understanding of your customer segmentation or take action with the data that you already have?

If you said segment your customer data base with greater accuracy you probably picked the wrong answer. According to research by Aberdeen Group (“How the Best in class use customer data to boost retention revenue in 2010″) best in class organizations focus on “doing” more than they do on analyzing. This certainly has the ring of truth to me.

It is a scene that we have observed many times in our customer experience management consulting practice. As we sit with the customer discussing what they would like to achieve we are met with a barrage of “can’t be done” because “our data is no good” or “we don’t know enough about our customers”. The desire to analyse, segment and target with ever greater depth is almost impossible to resist. More data piles on more data but we often seem to lose sight of the customer forest for the data trees.

As an example of this issue I’ll relate a recent customer experience. A few weeks ago I was running a workshop to design the high level segmentation fields that they wanted to use to describe their customers. After the initial flurry of segmentation style data was written up on the board I started to ask: “What will you do differently knowing that piece of information?” Applying this to each data field in turn, we ended up removing all but one key segmentation field and a couple of transactional indicators.

That one key field held the vast a majority of “how should we communicate with this person” information. The additional transaction indicators were then used to tweak the message to that person. All of the rest of the information was interesting but ultimately would never be used to change how they dealt with the customer.

Getting clarity on what drives customer purchases was good but the real value was to dramatically simplify their customer segmentation to the bare essentials.  At the same time they made their ability to execute, simple, straightforward and easy to implement. It is now clear how they should communicate with each customer based on one key segmentation variable.

I think that this is indicative of many customer data segmentation projects and discussions. The desire for an ever more complex, elegant, mathematically perfect segmentation model is almost overwhelming but at the cost of making almost impossible to implement customer communications. When you have 20 segmentation variables you feel the need to use them all in your communications plan but that is hard to do in practice.

Building a solid interactive and relevant communications strategy with just a few variables is complex enough. As you increase the number of variables the design complexity becomes exponentially more complex until it is just too hard to implement.

So have a quick look at your customer database. How many of those segmentation variables actually impact on the messaging that you send to customers and how many are molasses in your process — they are sweet but just slow you down?

If you’re looking to implement a customer experience management project why not start by downloading our free 4 Steps to Great Customer Experience Management report.

By Adam Ramshaw

Filed Under: Customer Analytics, Customer Experience Management, Customer Retention Tagged With: , , ,

Net Promoter Links to Recency-Frequency-Monetary (RFM)

July 9, 2010 By 1 Comment

From the early days of data-driven marketing, it has been known that marketers can predict which customers are most likely to respond to an offer by ranking them on the basis of;

  • how Recently they have transacted with you
  • how Frequently they have transacted with you
  • how much money (Monetary) they have spent with you.

It is also well known that of the 3: RFM, Recency is the best predictor of future business. My favourite database marketing guru, Arthur Hughes says;

“Frequency is often a powerful predictor of response, but it is seldom as powerful as Recency. We can easily illustrate the differences by comparing the response rates of the same group of people based on their recency and their frequency.”

Read the discussion here: http://bit.ly/aMST2R

So the customers most likely to be positive towards your product or services, those most likely to respond to your next offer, are those customers who have transacted with you most recently.

Nothing new there, but it has occurred to us that as NPS® also predicts future sales we should expect a recency effect in transactional (bottom up) NPS scores.

I was reading a thread in the LinkedIn Net Promoter Community where Paul Sherland asks the question; “…do you notice a time dependency in the [NPS] responses? My impression is that loyalty fades if it’s not reinforced by new engagement with the brand.”

John Abraham of Satmetrix responds (July 6, 2010) with;

“Your question about recent engagement is an interesting one. I have not analyzed this within our benchmark, but I have seen companies using NPS present findings at our conferences that match what you describe. Charles Schwab and Bain & Company did a joint presentation at our conference in San Francisco in January 2009 which mentioned the same phenomenon. Higher scores immediately after a branch visit, and then a drop off with time until another interaction occurs.”

So on the surface NPS and Recency cross-foot.

There is a lesson here for bottom up NPS.

Bottom up (also called transactional) NPS involves asking the NPS question close to a ‘Moment of Service Truth’ so you can translate customer feedback into tactical process improvements. If the score you receive partly depends on when you ask, make sure you are consistent in your timing. Preferably ask the NPS question as close in time as possible to the service experience you are improving, consistently, or your results will be distorted by recency differences.

For more information on Net Promoter Score and how/why it works download our free Introduction to Net Promoter Score (NPS).

If you are thinking about implementing Net Promoter Score (NPS) in your organisation give us a call. We can help you to measure word of mouth through an effective Net Promoter Score program for your business.

Net Promoter, Net Promoter Score and NPS are registered trademarks of Bain & Company, Inc., Satmetrix Systems, Inc., and Fred Reichheld.

Filed Under: General, Net Promoter Score Tagged With: , ,

How do you determine what is important to a customer?

May 9, 2010 By 7 Comments

I noticed the other day that I have touched on the subject of determining what is important to customers a few times in recent blogs (e.g Consumer Research: Poor research approaches give poor answers) but never given a full account of the different methods that you can use to do this.

This post fixes that oversight.

There are basically two approaches you can take:

1. Ask them (“Stated Importance”)

Under this heading there several methods that you can use but they all tend to suffer from the same drawbacks.

  • Customers don’t tell the truth: (intentionally and unintentionally).  In many situations customers either cannot or will not be honest with you.  A classic example of this problem is the question “How important is price”.  Very few customers will answer anything but very important for this aspect of your product or service, lest you raise the price.
  • Often socially or ego acceptable answers will arise: Few people want to answer questions in an anti-social way, even in a confidential survey.
  • Customers don’t really know: There are many times that a customer just doesn’t know how important an element of the product or service is to them.  They often purchase based on an ill-defined group of attributes and assigning specific importance to one attribute is very difficult.
  • “Industry requirements” can be misleading: Basic requirements, industry expectations and hygiene factors all fall into a category of attributes that just must be delivered for your product to be viable.  Think: bank statement accuracy — it’s not important until they make a mistake.

Approach 1: Ask for a rating

Simply ask the customer to rate how important a particular feature is to their purchase decision.  You’ve seen this sort of question before and it looks like this.

“From 10 to 1, How important is responsiveness to you?”

It almost always comes straight after a question that asks about how well the organisation is doing in performing the task.

Overall, this is the worst approach you can take.

It often leads to “ice skating” scores: 9.9, 9.9, 9.9, 9.9, 9.8.  Where everything is equally important — so you don’t know anything new.

While it is quick for customers to enter data, i.e. the survey doesn’t take long to do, there is also little to force respondent to take care in the evaluation.

Approach 2: Simple Ranking

You can also ask the customer to rank a list of attributes, forcing them to trade-off between each of the attributes.  For example:

Please rank the following in order of importance from highest to lowest

  • Delivering against your needs
  • Price
  • The accuracy and completeness the documentation
  • Technical competence of operational staff
  • Responsiveness in returning your call/email
  • Responsiveness in resolving you problem
  • Responsiveness in closing the loop after problem resolution

This is a better approach than the first.  For one, respondents cannot rank all of the attributes equally so you start to get some real information on importance.

However, it does take longer for the respondent to complete because they must think more carefully about their response.  Mind you that is generally a good thing, up to a point.

It also assumes that there is a difference in importance between all the different attributes and there may not be.

Lastly, this works well for short lists of maybe 4-6 items.  After that it can get difficult for the respondent to effectively rank the items.  In the worst case scenarios I’ve seen lists of 20 and 30 attributes, which are clearly impossible to rank effectively in this format.

If you have more than 6 items and enough respondents you can get around the problem by asking respondents to rank sub-sets of attributes.  Then use some fancy maths to combine all of the answers into one large ranking.

Approach 3: Best-Worst Ranking

This is a variation on the ranking idea above.  In this case you can have a large number of attributes (20 or more) and present them to respondents in groups of 5.  You then ask them to select the most important/least important attributes in each list.

This is a very powerful approach that can provide an accurately weighted and ranked list of key attributes.

It is relatively easy for the respondent to select just the most and least important item in each group of five.

There is one downside however, customers must answer 20 or more very similar questions, each with a slightly different group of five items.  This can cause survey fatigue and the dropout rate can be quite high so you need a larger number of potential respondents to get the required sample size.

Approach 4: Constant-Sum Allocation

This is also a better approach and asks the respondent to allocate points to different attributes. For instance:

“Please allocate 100 points between each of the following items where the more important an item the more points are allocated

  • Delivering against your needs
  • Price
  • The accuracy and completeness the documentation
  • Technical competence of operational staff
  • Responsiveness in returning your call/email
  • Responsiveness in Resolving you problem
  • Responsiveness in closing the loop after problem resolution”

This ensures that respondents weigh each attribute in the overall set and it allows them to give equal weighting to multiple attributes.  On the down-side this approach can often take the longest for respondents to complete.

You can also use this with quite large sets of attributes.  Respondents will tend to give the points to the really important items and ignore the other attributes in the list.

2. Derive it (“Derived Importance”)

This is the second key type of approach.  Instead of asking the respondent, you use statistics to infer what is important to them.  This gets around a lot of the issues cited above that occur when you ask outright.

The approach requires a “key outcome” measure.  This is a customer attribute or attributes that you want to influence.  You can use customer satisfaction but we would suggest Net Promoter Score.  If you can tie responses to customer data (revenue, revenue growth, gross margin, gross margin growth, etc) then that is very good as well.

In this approach you ask the respondent about the organisation’s performance in each attributes that you are investigating e.g.;

“How responsive are Company X in returning your call/email”

Then you use statistical analysis to calculate the relationship between the attribute and the outcome.

Using this approach you can infer the underlying drivers of whatever outcome measure you are trying to achieve without asking directly.  This is powerful becuase it can get at the importance levels of a basket of key attributes at the unconscious level.

The downside is that it requires higher levels of statistical competence than just graphing the numbers.

Adding this level of rigour to the process is not necessarily a bad thing as I’ve reviewed many customer survey reports over the years that make statements the numbers really can’t support.  All as a result of an incomplete understanding of the statistics being examined.

By Adam Ramshaw

Filed Under: Customer Feedback Process, General Tagged With: , , ,

How to evolve on-line customer advisory panels to add value to your business

March 28, 2010 By 2 Comments

On-line customer advisory panels are a relatively new feature in the market research business.  Often these panels are created by dedicated market research companies to provide ready access to a large number of respondents for a market research survey.  In that way they can be quite valuable to organisations wanting a quick way to get feedback on a new idea.

These types of panels can be very useful for two reasons.

  1. They have large sample sizes, ie. the entire online community group, not just small focus groups, online surveys and face to face interviews.
  2. You can receive feedback quickly on product and on-line design,  from customers and prospects; often within hours or a day or two.

However, they can also be quite limiting because they do not develop a relationship with the company for which the research is done.  The individuals on the panel have a relationship with the market research company that “owns” the panel.

Recently we created and ran a dedicated on-line customer advisory panel for a leading insurance company that wanted to launch a new direct (on-line) brand into the market place.  This panel was different to the panels noted above in that the insurance company concerned built and ran the panel from scratch.  This allowed the company to build a relationship with the panel participants and allowed a range of secondary areas of value to be created.

Initially the role of the panel was to gather market research style insights and segmentation information.  In this way it was similar to the market research company owned panels above. The similarity stops there however.

As the research process evolved, we were able to link internal company information with portal member profile type.  Then judicious use of questions, blog posts, quick polls and surveys enabled the client to extend their understanding of each segment’s needs and key drivers.

That’s not all however.  The role of the on-line portal evolved over time from a source of on-line segmentation insights to a forum for the co-creation of products and value propositions to a powerful advocate based launch platform for the business.  Each phase layered additional depth and value into the client’s relationship with the portal users.

This evolution of value elements does not happen immediately but over time the value of the panel far outweighs the initial market research function.  The final outcome of this process is, in the right circumstances, for customer lead support to occur.  This is where customers help each other regarding questions and approaches to using your product.

This customer lead support has been occurring in the technology/software space for many years but this approach brings it within the realms of more mainstream products.

By Adam Ramshaw

Filed Under: Customer Analytics, Customer Feedback Process Tagged With: ,

Consumer Research: Poor research approaches give poor answers

March 4, 2010 By 8 Comments

This post from the Harvard Business Review’s Daily Stat on 15 February 2010 shows a surprising lack of insight into how consumers actually respond to research questionnaires.  It reports a McKinsey survey where consumers were asked what they were more interested in: core benefits or bells and whistles.

If you believe the results, consumers are more interested in core benefits.

I don’t believe the results.

The issue here is the way the survey has been undertaken.  In this case, asking customers what they want is unlikely to give you an accurate answer to your question. Not because they will not tell you the truth but because they are unlikely to act in the same way, in practice, that they say they will act, in theory.

If you really want to know what consumers want then, in this case, it’s best to look at what they actually do.  On this issue (core benefits or bells and whistles) it would be a relatively easy piece of analysis to do: compare sales of the upmarket “bells and whistles” models with the downmarket “core benefits” models.  That way you can see what customers actually do rather than what they say they do.

Poor research design like this is a common problem.

For instance, we’ve seen loyalty program research that, when asking customers if they want points or discounts, found they overwhelmingly want discounts. Then those same customers go on to collect points, respond to points promotions, etc.

And there was the set of supermarket research that asked customers if they would buy “store brands”.  When you compared actual purchase data with customer responses, many customers with baskets full of the store brand product said they would never lower their standards that low.

If you naively believed what customers told you in either of these circumstances you could easily build the completely wrong customer experience.

Consumer research is fraught with this kind of problem and when designing your research approach you need to act defensively to guard against getting incorrect or inaccurate results.  If you don’t, you will get a pretty chart that looks nice in the report but the insights in it will be just plain wrong.

Another example of this type of not believing what customers say is the “how important is feature x” type of question.  Understanding how important a service feature is to the customer is critical in designing good service processes.  So questionnaires are often used to determine which features are most important.

The problem comes if the surveys are poorly implemented.  You’ve probably seen the bad ones; they look like this (paraphrasing):

Q1: How good is our price?

Q2: How important is price?

Q3: How good is our service responsiveness?

Q4: How important is responsiveness?

Q5: How good is our feature x?

Q6: How important is feature x?

Even before I see the results I know what the answers will be.  Everything is important: 9s or 10s out of 10.  So what do you know now?  Nothing more than you knew before because everything has the same high importance.  You’re back to square 1.

Even if the results are not all 9s and 10s they will be skewed by what customers want you to think.  For instance, no customer is going to tell you that price is unimportant, lest you decide to put up their price.

There are at least two better approaches than this:

  1. Some type of forced ranking: In this approach you force the customer to rank the importance of each feature using a points approach, ranking or a best-worst approach.
  2. Infer importance: If you design the survey in the right way you can infer what is important based on the answers you get from other questions or actual customer behaviour.  This takes a bit of additional analysis but is well worth the extra effort.

Both of these alternatives will deliver a much more accurate outcome.

So, when you next look at the survey questionnaire your agency has provided, act defensively, and think about whether the answers you get will be accurate.

Have you seen other poor research approaches?  Leave a comment and let me know.

By Adam Ramshaw

Filed Under: Customer Analytics, Customer Feedback Surveys Tagged With: , ,
« Older Posts