THE RKGBLOG

AOV and Lifetime Value

Kevin Hillstrom wrote a great piece on the impact of purchase price on lifetime value, and I thought I’d pick up on that thread.

Direct marketers know that most sophisticated “scoring” systems for ranking the value of customers are based on “RFM” modeling. Studying data carefully for the last, oh 50 years, has taught the direct mailers that likelihood to purchase is best predicted by three variables:

  1. Recency: The more recent the customer’s last purchase the more likely they are to purchase again. This is a bit counter-intuitive at first blush; one might naturally assume that someone who just made a purchase is less likely to return than someone who hasn’t bought in a while, but the data says otherwise. The underlying reason likely relates to some combination of the following:
    • the recent customer feels relatively flush with cash at that time or they wouldn’t have made the last purchase
    • the recent customer just had a good experience with your brand (hopefully) so you are top of mind with them in that category at the moment; and
    • the recent customer is less likely to have loss a job, their life, or found a competitor of yours that they prefer.

    Whatever the case, recency is the strongest factor in predictive modeling.

  2. Frequency: the number of times the customer bought from you in the past. The more frequent, the more valuable.
  3. Money spent: how much did they spend on their most recent purchase? Customers who spend little on the first purchase tend to always spend less and vice versa. The big spenders are the most valuable.

Other factors play a role, but those are the big three.

While “Money spent” is the least predictive of the three, it is still quite meaningful and it gives us an interesting angle in paid search bidding.

When setting efficiency targets many smart marketers factor in some notion of lifetime value. That is, the value of the traffic isn’t simply the margin on the orders placed through a given keyword or cluster of keywords, some credit should be given for the value of future purchases from that customer as well as the likelihood that they’ll tell a friend about your company. Measuring the actual lifetime value by keyword may be theoretically ideal, but practically impossible.

What is NOT impossible is adjusting the lifetime value component of an efficiency target based on the average order value associated with a given keyword or cluster. Higher order values (by margin dollars) are more valuable in and of themselves, and a good bid management system will already take that into account, but it is also the case that those customers are likely to have a higher lifetime value — and wealthier friends!

How much of a factor is this will depend on the business. Likely if Pat’s order is twice the value of Taylor’s she is not twice as valuable as a customer, but perhaps a square-root of two factor may be a good rule of thumb.

This is not a game changing strategy, obviously, but when the core blocking and tackling is well under control, these are the types of tuning mechanisms that allow us to raise the bar.

Technorati Tags:

Comments
7 Responses to “AOV and Lifetime Value”
  1. George,
    Your post is the perfect set up for my next SEL writeup on measuring Customer Lifetime Value (CLV). I will be delving a bit into the math and the intuition behind it. The basic formula is quite straightforward and looks a lot like the discounted cash flow (DCF) models used to value stocks.

    Incidentally, there is a paper by Gupta, Hanssens et al that reviews different CLV models. One of the things it says is that “recent studies have compared CLV models with RFM models and found CLV models to be superior”. Definitely worth a read.

  2. Just don’t make me look bad, Sid, that’s all I ask :-)

  3. For those interested, the paper to which Sid refers is here: http://www.anderson.ucla.edu/faculty/dominique.hanssens/content/JSR2006.pdf, and it is indeed an excellent read.

    I would say they give RFM modeling a bit of a short-shrift. Models used by catelogers have advanced far beyond simple spreadsheets, to incorporate R, F and M into 15 or 20 variable neural networks. The tests they site are also a wee bit short of convincing given the tiny customer files used. There are many many ways to do RFM modeling, and they dismissed them rather abruptly imho.

    That said, they’re dead on that all these models are geared towards immediate gratification — who is most likely to respond to the next marketing blast — which may not address the more complex needs of a business. The whole issue of marketing mix and treatments adds a layer of complexity that we’re all wrestling with right now.

  4. Hi George,

    Can you post some links to good RFM papers ? I am not very familiar with advanced RFM and want to learn a bit more about the RFM modeling bit. So if you could be so kind…

    BTW,
    Happy New Year !

  5. Sid, unfortunately I can’t point you to a specific paper. They may be out there, but I haven’t really looked. We’ve done, and continue to do, a good bit of collaboration with academia, but I must confess, I don’t keep up with the literature like I should. It has always seemed to me that the practitioners tend to be pretty far ahead of the academics, so I haven’t paid much attention.

    I will say from experience that 10 years ago we were using pretty high-powered modeling techniques in catalog circulation, with R, F and M as inputs along with other interesting flags we’d noticed. We knew, for example that response rates in zip codes with competitor’s stores in them performed significantly worse than others — we drove business to them — so zip codes played a role, as did seasonality. Data showed that customers whose first purchase was at Christmas had significantly lower CLV than others, so that played a role as well.

    Perhaps folding 15 variables into a neural net no longer qualifies as RFM modeling, but R,F and M were clearly the strongest predictors of any that we examined.

Trackbacks
Check out what others are saying...
  1. [...] This post was mentioned on Twitter by Chad Summerhill, ana peña. ana peña said: AOV and Lifetime Value: Bigger orders aren't just better now, they're better later, too. http://bit.ly/i6IidR #fb [...]

  2. [...] Érdekes okfejtés arról, hogy a megtérülés számításakor érdemes figyelembe venni, hogy a kulcsszóhoz társítható rendelés milyen jövőbeni rendelésekhez vezethet [...]