Customer Modeling for Finance Folks

Jim answers questions from fellow Drillers
(More questions with answers here, Work Overview here, Index of concepts here)


Q: My boss (VP of Phone Sales) is really looking to try out some new ideas and RFM is one he has latched onto. He actually has explored this concept for a few years but never acted upon it. Anyway, he just purchased your book and after finding that he did not have time to read it he gave it to me. My job was to read and understand at a high level and to lead a discussion with the marketing group to get them excited about the concept. I am a finance guy by trade so this concept was very interesting.

A: That’s funny, the people who really “get it” the most are Finance people and IT people, because my approach is very numbers driven.  Stuff either works or it doesn’t – did you make money or not?  Many marketing people seem to dislike the idea of accountability…..hmmm…

Q: Obviously I either did not do a good enough job explaining RFM, Latency Tripwires, etc. or they just are unwilling to have someone from their team tackle the concept. My feeling is they felt this is a sales tool. The question they always wanted answered was “Why did the customer behave the way they did? We find that out and make a sales call, not engage in ‘marketing air cover’ tactics.”

A: Not sure what you mean by this…in fact, depending on the value of the customer, a sales call might be exactly what is needed. If you have a formal “wall” between sales and marketing, usually the issue can be decided by “degree of pain” e.g. how painful will it be to lose the customer?  Generally, a personal call is more effective than Marketing but more costly, so you use those guns sparingly.

If you have a small number of very high value customers who look to be defecting then a sales call is triggered. If you have lots of medium to low value customers who look to be defecting, then a direct mail campaign is probably what you need, which is probably Marketing. Match the value of the effort to the value of the customer; this is how you get gigantic ROI’s (or since you are a finance guy, more accurately something like ROME’s – Return On Marketing Expense). The scoring approach to customer value is about allocating scarce resources to the highest and best use.

I think what Sales is saying is this: if you know a specific thing about a customer, we handle that “one to one” thing; Marketing does the “all customers” messaging. And this is precisely the point of customer models – they allow Marketing to do the “one to one” thing, as opposed to the “air cover” thing.

Q: So it has fallen upon me to develop a project plan and come up with some ideas to implement.  If we can not get marketing support we will run with it ourselves.

A: Good for you! A good old fashioned skunk works operation, I love that! And led by a Finance guy on top of that.  Bravo!

Q: I am now reading the book for a second time and I have a slight problem with how to best implement with our business. I can see how this concept could be used to radically change our sales channel, but I do not think I have that much pull.

A: Well, let’s take a look at it. Typically, and particularly since you are in Finance, what you do is look to prove out a high value concept, then share financial success up the chain. This builds momentum for the approach and gets people really interested in knowing more, which leads to taking concrete action.

So for example, find your very highest value potential defectors using either Recency or Latency. Then split them into two equal groups – test and control. Have sales call the people in the test group and find out what is causing the defection behavior, try to save the customer.

Then 90 or 180 days later, look at the number of test and control that stuck with the service. Subtract the control number from the test number, this is the “net” retained due to your calls. Multiply by value of the contracts, and you have sales due to your program.

Q: We are a subscription service in which customers pre-pay for the service they expect to use. Our sales (and I guess marketing to some extent) are responsible for driving customers to use their service throughout the year. Usually if a customer uses more than they committed to then they raise the commitment the following year. For us sales leads to higher revenues leads to higher sales, etc, one big circle. So I guess my question is this: Can RF scores be used for a pre-paid subscription service?

A: Sure, but perhaps not in the “classic” sense of transactional revenue. For many service biz, particularly subscription ones, you profile activity other than billing, since the billing tends to be static.? Sounds to me like what you want to profile is **usage** – the more Recently and Frequently a customer has used the service, the more likely they are to continue using it. I assume you are authenticating subscribers to the service on your web site, so this shouldn’t be a big deal.? Then your scores would rank customers by likelihood to “continue using the service” and their value.?

High value customers with falling or low likelihood (falling RF score) to continue using  the service get a sales call, mid to low value customers with low likelihood to continue get a direct mail piece from marketing.  Dramatic changes in score require the most urgent attention, in terms of allocating resources.

Q: As an FYI, we have customers who pay as they go and customers that sign a yearly commitment. Would it be best to segment the two groups individually for the RF model and Latency tripwires?

A:  Yes. Annual subscriptions and Pay As You Go are two fundamentally different behaviors and mindsets, so mixing them will confuse the scoring. You have a Long cycle (annual) and a Short cycle (PAYG) decision being made; both the models and the actions would be different. For example, PAYG will be a more sensitive model with action required more immediately. Also, these are probably low value customers so you’re talking about e-mail or direct mail.

And, your measurement cycle would be different. Taking the test example above, you would check for “net results” on PAYG probably at 60 days; annuals you would wait for renewal date unless the offer affected this date in some way.

Q:  We also have different size customers some spending more than $10K / year and some $1K, should we segment based upon dollar values as well since the more they committed to the higher their FM scores (you would expect)?

A:  You can make anything really complicated with segmentation if you want to! Just starting out, my answer is Segment in terms of message yes, but Segment in terms of scoring and triggering action, no.

Keep in mind the Current Value / Potential Value model; don’t confuse the two behavioral vectors and their meaning. Current Value – what they have paid so far – is about how valuable the customer is to the company and determines what action is taken. This is the “personal call” versus “send e-mail” part of the equation; the cost component.

The Potential Value (Recency, Latency) is about predicting the likelihood for future business, it’s about “when” to act. This is the risk of losing the business in the future.

So I would not segment by value in terms of predicting defection, because thelikelihood of losing the business is really unrelated to the Current Value of the customer.You can have High Value and Low Value customers with the same defection likelihood, whether “value” is measured as Sales, Page Views, Engagement, whatever. Value is largely independent of likelihood to defect. But once defection is predicted, you then segment between High Value and Low Value and take action based on the value of the customer or visitor segment.

The two primary rules of High ROI Customer Marketing are:

1. Don’t spend until you have to
2. When you spend, spend at the point of maximum impact

Current Value = What to do
Potential Value = When to do it

That’s why this approach is so much more profitable then dropping Marketing on a “batch and blast” calendar schedule (you called it “marketing air cover”). Right message, to the right person, at the right time.And itworks especially well online because Relevancy (right message, right time) is so important and switching costs are low.

Q: What kind of Marketing should we do?  Is there any other segmentation we should try?

A: Well, that’s a little tough without knowing more about the business, but there’s a good way for you to find out!

With a service, you hopefully know why people stop using it. To prepare for these campaigns from a Marketing perspective, find defected best customers (high value cancels) and look at why they stopped using it (or interview them if you don’t know, offer a free month or whatever to get them to talk to you). Create Sales / Marketing – pitches / materials / offersto address their issues.

Then when you see a client engaging in a defection pattern on usage (drop in RF score, Latency Tripwire), engage the appropriate response (Sales or Marketing) based on the value of the customer.

And sure, the more you segment your customer base, the better it works. You should start at the bottom, however. Don’t “out-think” the segmentation; let the data speak to you. Try something at a very basic level and look for the hands to be raised; this will tell you what works and put you on the right track for more complexity.

For example, let’s say (and I imagine it would be true) that SIC codes play a role in your sales and retention. Certain types of businesses are simply going to be more likely to realize value from the services. So you do a campaign (sales, marketing, or both) to *all* customers in a particular defection state and let the SIC data speak.

Let’s say for simplicity that you find if a PAYG subscriber doesn’t use the service for 10 days that’s a warning flag for defection. You prepare and drop the retention campaigns to any accounts that “trip” this trigger – right message, at the right time.

What you see when the data comes back is certain SIC codes had a very high response and “activation” and start using your database again, and others do not. The data has now spoken, told you which SIC’s it is worth spending time / money on.

Then you look at bit deeper, and find that within an SIC code that looks to be a “bad idea” overall, the results are pretty good as long as the offer is made by direct mail in the South.  So you keep this particular segment of the “direct mail” campaign and kill the rest of the marketing activity for that SIC code.

You can look for other segments by value, by region, by services subscribed to, by type of data they look up, whatever. As you subdivide segments, you will find new pockets of profitability. You could spend a LifeTime chasing down all the segments – I have never, ever finished this task on any particular engagement. In fact, clients call me years after they have stopped using my services to tell me they have discovered unique new segments that are extremely profitable.

I’m not saying you should?abandon traditional customer communications,the batch and blast that you do. What I am saying is there is a deeper, more Strategic Objective you can drive through either customization of current programs or by adding an additional layer -maybe cut back on a little of the blasting at the same time?

The basic idea is really no different than optimizing Campaigns – except you’re optimizing Customers by recognizing problems with individuals and offering solutions, instead of always being in their face asking for something – especially when the customer is already demonstrating to you there is a problem of some kind. A little “Is there something we’ve done wrong”? or “Can we help you use our product more efficiently?” or “Would you take a survey?” to specific customers could not hurt.

Sound like a good idea?

Jim

Get the book at Booklocker.com

Find Out Specifically What is in the Book

Learn Customer Marketing Concepts and Metrics (site article list)

Download the first 9 chapters of the Drilling Down book: PDF 

4 thoughts on “Customer Modeling for Finance Folks

  1. Great summarization of key tenets of RFM/Latency. The only comment I have is on segmentation.

    1. I agree that there is no point in segmenting to the nth degree prior to scoring, especially to someone who is just starting on things like RFM. However, just based on my experience I will stress on some segmentation “if there are distinct customer groups in terms of behavior”. As an example, it is possible that because of the nature of products or service a company offers, certain industries are more frequent users of the products/service. That does not [by default] make the customers in other industries any less valuable — purely on the basis of RFM. Just bringing this up as I have see many folks creating an RFM model without adequate thought and coming to wrong conclusions on who are ‘good’ customers and who are ‘risky/defecting’ customers.

    2. The second point is on current value. Again, I agree with you that current value is a measure of customer value. However, just as before I want to caution users that one has to look at current value keeping (a) the customer profile in mind (e.g industry as mentioned above), and (b) the customer’s past value. If one is doing a RFM scoring frequently then this is less of an issue. However, if someone creates a one-time RFM model they can wrongly classify a customer as ‘low-value’, when the truth is that the customer is a ‘high-value’ customer but on the declining life-cycle (giving their business to competition).

  2. Ned, thanks for the comments, agree completely.

    Example of #1 was in the advice to treat the annuals and the PAYG differently. Clearly these are two different behaviors and scoring them together is just going to muck the joint up.

    On #2, that’s why I’m not a very big fan of “traditional RFM” – there’s too much chance for people to screw it up *unless* they’re using it for a one-off purpose – like mailing a catalog. Much prefer the LifeCycle Grids approach, which is similar to keeping track of monthly RFM scoring at the customer level but a lot easier to understand – I think, anyway!

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.