Context Parameters for Best Use of Recency Metric

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

Topic Overview

Hi again folks, Jim Novo here.

Time to take a look at some basic strategy framework ideas in a customer retention program. You have to know where you are first before you can decide what actions to take, and this initial analysis will prompt ideas for action. Trust me, finding out specifically what is happening in an actionable way is the most critical step to the design and execution of a customer retention program. Not doing this is why so many of the programs fail. Ready, Driller? Let’s do it.


Q: I’m reading some of your information you have on your web site, regarding Recency / Frequency. I’m curious about the statement that Recency is the number one most powerful predictor of future behavior – if you did some thing recently you’re more likely to do it again.

A: Yes. Funny thing about web sites, it’s hard to control what sequence people read things in. From the questions below, I believe I have failed to introduce you to the Recency metric in the right context. Shame on me!

Q:  With regards to purchases, how is this so?  I can think of numerous instances where this might not be true.  In fact, I would guess that price of purchase would be a more likely indicator of whether or not someone would purchase again.  If I’m running Best Buy, and someone comes and buys a washer / dryer, I would not expect they’d be buying another one anytime soon.  Ditto furniture, cars, travel bookings, etc.

A:  Two important “context” issues surrounding Recency.  First, Recency is a “relative” metric, it doesn’t exist by itself, but “relative” to other data points.  In the case of customers, Recency and the “likelihood” is a relative comparison of two customers, two customer segments, or a customer versus the average customer, for example.  So for a washer / dryer purchase, looking at the customer in question, Recency answers the question, “how likely is this person to purchase relative to another customer”.  It’s a scoring system, a ranking of likelihoods to (in this case) buy, or visit, or download, or whatever.

Second, Recency is a customer-based metric, not a product-based metric; it describes the behavior of the customer and likelihood to purchase, not likelihood to purchase a specific product.  I agree a customer who bought a washer / dryer Recently isn’t very likely to buy another one.  This doesn’t mean they are not likely to buy a stove or microwave though.

So putting these two context bits together:

Looking at a customer who just bought a washer / dryer and comparing them to a customer whose last purchase was a washer / dryer 6 months ago, the more Recent customer is more likely to purchase from Best Buy again relative to the customer who bought the washer / dryer 6 months ago, without regard to what they might purchase.

Q:  I would think that you’d really have to intersect the Recency with Frequency in order to truly predict the future behavior.  So if I bought 5 times on your site, and the most Recent was a week ago, I would think that person would be a higher value than someone who has only bought once, but 2 days ago.

A:  Well, value is a different story, Recency only predicts likelihood to buy, it speaks to potential value.  Frequency speaks to current value, this is a different concept.  What you said is true, the former person has a higher current value, but the latter person has higher potential value, is more likely to create value in the future, than the former person.  This is the customer value model, you can check it out graphically here; all customers have some mix of current and potential value.

In fact, you can create a two-digit score, in this case, Recency and Frequency or what I call an RF score, and rank all customers by a mix of current and potential value.  This can be used for many things, for example, predicting the response rate to promotions – the higher the score, the higher the response rate.

So adding Frequency to Recency does in fact make a Recency model even more powerful than Recency alone.  However, the reverse is not true.  Frequency alone is typically not a reliable predictor of likelihood to act in the future.

High Frequency and long Recency indicates an already defected best customer, again, relative to other customers with higher Recency.  Low Frequency / short Recency is a new customer who is a potential best customer.  All of this can be plotted on the customer value model grid to create a “map” for managing customer value.  Frequency by itself is not nearly as predictive as Recency by itself, though many people base segmentation strategies on Frequency because it is easier to count transactions than to measure “time since last activity”.  All Frequency tells you is what the customer is worth today, it does not speak to future value.  If you are talking about “likely to buy again”, you are talking about the future, not the present.

The best web-oriented story I have seen on this subject is from Amazon.  In the very beginning they used to announce the total number of customers who had purchased over 10 times (or was it 100?) each quarter.  It was the main number people focused on.  Then a retail analyst – who I’d guess was familiar with customer retention metrics – asked how many of these people had bought in the last 12 months (12 month Recency)?  The answer was a good deal less than 100%, and the stock absolutely tanked that day. Why? Because the retail analysts knew from experience that many of those 10x customers had very low future value, and the future outlook for the business is what drives stock prices. Oh, the metrics, the metrics …

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 

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.