Drilling Down

Now that a number of large scale CRM implementations have been completed and many more are underway, issues have surfaced regarding measuring the payback on these implementations.  Will the company ever experience a positive Return on Investment with CRM, given the complexity and cost of implementation?

Clearly, to correctly answer this question, standards or goals for payback and measurement should have been established up front, and companies should have established their own measurement criteria.  However, some of these companies are unfamiliar with the direct customer contact business model in general, and database marketing metrics in particular.  These companies may find it more difficult to accurately measure payback.

For example, there may be cost reductions or efficiencies created by the CRM implementation which are relatively easily measured (talk time, etc.) and used to offset the CRM costs.  But if the explicit theory of CRM – improving customer retention and loyalty will increase customer valuation – is accurate, then the payoff for a CRM implementation should also include the likelihood of increased customer value in the future.  To not take this factor into account would miss a potentially large offset to CRM implementation costs.

This article describes a universally applicable, simple to implement set of criteria for measuring the direct effects to future customer value of a CRM implementation in the B2C space.  It will be especially useful to companies without a long-term history of direct customer contact, who may be having difficulty accurately measuring the LifeTime Value of their various customer segments.  This method allows a company to start the measurement process at "today", meaning there is no requirement for a long operating history, intense analytical techniques, or direct surveys of customers.

For those more familiar with database marketing measurement techniques, the following is a derivation of RFM analysis, allowing a very simplistic scoring of the customer database, and monitoring positive or negative changes in future customer valuation using these scores.

For several decades now, the catalog industry has been studying direct customer contact strategies and tactics, and dealing with handling the customer service implications of conducting business in this way.  Over time, they have developed models directly linking the future value of a customer to the business with the customer’s behavior.  Recent work in the TV shopping industry, the first 24 x 7 "interactive" customer shopping experience, and on the Web, have confirmed these models developed for the catalog industry have direct application to interactive customer behavior in multi-channel direct contact environments.

The most powerful and simplest to implement of these models is called RFM – Recency, Frequency, and Monetary value.  RFM is a behavior-based model, meaning it is used to analyze the behavior a customer is engaging in, and make predictions based on this behavior. The RFM model states three conditions:

The more Recently someone has engaged in a behavior, the more likely they are to repeat the behavior.

The more Frequently someone has engaged in an behavior, the more likely they are to repeat the behavior

The more Monetary value someone has created by purchasing, the more likely they are to continue to purchase.

Classic RFM implementation ranks each customer on the Recency, Frequency, and Monetary value parameters against all the other customers, and creates an RFM "score" for each customer.  Customers with high scores are usually the most profitable, the most likely to repeat a behavior (visit or purchase, for example), and the most highly responsive to promotions.  The opposite is true for customers with low RFM scores.

Assuming the behavior being ranked using RFM has economic value, the higher the RFM score, the more profitable the customer is to the business now and in the future.  For this reason, RFM is closely related to another database marketing concept: LifeTime Value (LTV).  LTV is the expected net profit a customer will contribute to your business as long as the customer remains a customer.

Because of the linkage to LTV, RFM techniques can be used as a proxy for the future profitability of a business.  High RFM customers are most likely to continue to purchase and visit, AND they are most likely to respond to marketing promotions; these customers likely have the highest LifeTime Value.  The opposite is true for low RFM customers; they are the least likely to purchase or visit again AND the least likely to respond to marketing promotions, so these customers tend to have low LTV.

High RFM scores represent future business potential, because the customers are willing and interested in doing business with you, and have high LTV.  Low scores represent dwindling business opportunity, low LTV, and are a flag something needs to be done with those customers to increase their value.

One simple application of RFM is Hurdle Rate Analysis, where "hurdles" are selected for Recency, Frequency, and Monetary Value, and the entire customer base is evaluated against these hurdles as a group.

A Hurdle Rate is simply the percentage of your customers who have at least a certain activity level for Recency, Frequency, and Monetary value.  It’s the percentage of customers who have engaged in a behavior since a certain date (Recency), engaged in a behavior a certain number of times (Frequency), or have purchased a certain amount (Monetary value).

Because of the link between RFM and Lifetime Value, it can be concluded:

If the percentage of customers over each hurdle (Recency, Frequency, Monetary value) is growing, the business is healthy and thriving.  Customers are responding positively to the experience they receive, and as a group are more likely to engage in profit generating behavior in the future.

If the opposite is true, and the percentage of customers over each hurdle (Recency, Frequency, Monetary value) is falling over time, high value customers are defecting and the future value of your business is falling.  Customers as a group are responding negatively to the overall service they are receiving.

A business should expect a successful CRM implementation, because of all it implies for customer satisfaction and productivity (cross-selling, etc.), would result in rising Hurdle Rates.  An unsuccessful implementation would cause falling Hurdle Rates, and an implementation with no effect would drive no change in Hurdle Rates.

If the business has an understanding of customer LifeCycles, the logical Hurdle Rates to set for Recency, Frequency, and Monetary value would equate to customer behavior at primary changes in the customer LifeCycle.

For example, if it was known customers who have not purchased for 60 days rarely make another purchase, the logical hurdle to set for Recency is 60 days.  Sweep the database and determine the percentage of customers who engaged in purchase behavior in the past 60 days; this is the starting Hurdle Rate.  If 20% of customers have made a purchase in the last 60 days, 20% is the starting Hurdle Rate.  The same approach would hold true for Frequency and Monetary value of purchases.

If the business is very new or has never studied the customer LifeCycle, then a good default position to use is based on the 20/80 rule (20% of customers generally generate 80% of the behavior, be it sales, visits, etc.)  The analysis would default to a "starting Hurdle Rate" of 20% for each behavior, and examine the customer base to determine RFM values corresponding to the 20% hurdle.

In this case, the business would look at the top 20% of their customers for each of the Recency, Frequency, and Monetary value parameters, and examine the "tail end" customers – the bottom customers of the top 20%.  These values would become the hurdles the customer base is judged against.  This exercise is completed for each of the RFM parameters.

For example, in a database of 10,000 customers, to determine the Recency hurdle using the 20/80 rule:

1.  Select the behavior to be profiled – purchases, visits, etc.

2.  Sort customers by most Recent date of the behavior

3.  Starting at the most Recent customer, count down to customer
      number 2,000 (20% of 10,000) in this sorted database.  Examine
      the group of customers near this target level, perhaps from 
      customer 1,950 to customer 2,050.

4.  Determine how long ago these customers, on average, 
     engaged in the behavior you are profiling

5.  You find these customers last purchased an average of 60 days ago

6.  The Recency hurdle becomes 60 days for the "today" or 
      starting Hurdle Rate of 20%

Regardless of whether the Hurdle Rate is set using the customer Lifecycle or the 80/20 rule, the operational implementation is the same.  Each week or month, sweep the database and determine the percentage of customers who have engaged in the behavior within the hurdle definition.  For a 60-day hurdle, it would be the percentage of customers engaging in the behavior in the past 60 days.

If the percentage of customers "over the hurdle" (engaging in the behavior less than 60 days ago) grows over time, the Recency Hurdle rate is rising, and the future value of the customer base (LTV) is rising.  If the percentage of customers "over the hurdle" is falling, the Recency Hurdle Rate is falling and future value is falling as well.

This exercise can be completed on the same behavior for Frequency, and if there is a transactional value to the behavior (a purchase), for Monetary Value as well.  Additional behaviors can also be monitored simultaneously; on the web, tracking purchases and visits together would make sense.  Unless the business has a very clear understanding of revenue per visit across different areas of the site, it is unlikely tracking the Monetary Value of visits would be very useful.

The Hurdle Rate values can be graphed over time, and trends established.  Clearly there will be fluctuations up and down, and seasonality in retail or event oriented businesses.  But if solid trends in Hurdle Rates develop in either direction, or year over year comparisons are dramatically different for a seasonal business, the measurement should be judged to be significant and actionable.  Graphing Hurdle Rates over time provides an easy way to present a somewhat complex subject to management: line up = good, line down = bad.

Percentage of Customers
 Over Various Hurdles - 8 weeks

The lines in the chart show the percentage of customers over each Hurdle tends to be rising over time.  This particular chart is a combination of behaviors and RFM parameters – Recency of Visit (R), Frequency of Purchase (F), and Monetary value of Purchases (M).

The percent of customers who have visited in the last 30 days (Recency, broken heavy line) is rising.  The percent of customers who have purchased over 10 items (Frequency, heavy solid line) is rising.  The percent of customers who have spent over $500 in total (Monetary Value, light solid line) is also rising.  A business can mix and match tracking of behaviors and Hurdle Rates according to priorities in the business model.

This is the picture of "growing the share" of best customers in your customer database.  Your best customers are remaining with you, and other customers are "growing into" becoming best customers.  This is the effect desired in a positive CRM implementation.

Hurdle Rate analysis, like any good "lift" analysis, assumes there is only one significant variable being measured, in this case, the CRM implementation.  If the business makes radical changes to product offerings, or adopts a new business model during the CRM implementation, Hurdle Rates (and most other kinds of analysis) are not going to be able to tell which changes are really affecting the customer base, positively or negatively.  If direct cause and effect measurement is required, a basic assumption in a CRM implementation should be "keep everything else as stable as possible".

The direct linkage between RFM and LTV can break down for specific customers where costs related to the customer are much higher than the average customer.  This would include customers with extreme acquisition costs, customers with significant return rates, or customers with high after-the-sale maintenance costs.  For these customers, it is possible to have a high RFM score and low LifeTime Value, so the linkage breaks down.  In most businesses, it would not be true that the "best customers" from a sales or visit volume standpoint are also the least profitable, and that low volume customers are the most profitable.  If this is the case, there clearly is an inverse link between RFM and LTV, and the approach used above should be inverted; that is, falling Hurdle Rates are good for the business.  This is a business surely headed for bankruptcy, as less customer activity and satisfaction is more profitable to the business, meaning zero activity is maximum profitability.  For some business models, this is probably true.

For those not familiar with classic RFM analysis, the sequence of the characters R-F-M is intentional.  Recency is by far the strongest predictive variable, followed by Frequency, then Monetary Value.  Lacking the resources or ability to track all three on any particular behavior, start with Recency.  You can add the others later as resources become available. Recency of a behavior is the single most powerful predictor of the behavior repeating, and of response to a promotion.

A business with a lot of "noise" in the customer base may want to exclude this noise before calculating hurdles and Hurdle Rates.  For example, if there are a lot of one-time buyers, or visitors with only a few visits, excluding these customers completely (like they didn’t even exist) in the above process will lead to more accurate and meaningful analysis.  These customers tend to "dilute" the process by lowering the threshold for defining a good (top 20%) customer.

If you are looking for a "hard match" on incremental profits generated versus the cost of a CRM implementation, you will need to fully understand customer segmentation and valuation prior to the implementation.  Without this knowledge to serve as a "control" to measure later results against, the capability to definitively measure profit gains and match these with costs will be lost.  A fairly simple way to handle this is to look at customer profitability by decile before and after the CRM implementation.  This is extensive work, requires excellent customer history files, and embodies the kind of thinking a business should have a grasp on in order to successfully implement CRM in the first place.

What would you like to do now?