Monday, April 02, 2007

Deltalytics' Lloyd Merriam Comments on LTV

My friend Lloyd Merriam has left a thoughtful comment on last week's post about Lifetime Value. It's worth treating as a post of its own. Here's Lloyd:

I completely agree that customer lifetime value (LTV) is the single metric against which all strategic business decisions should be evaluated. Although non-trivial, determining the current value of a customer isn’t particularly challenging. Calculating future LTV – which, as you know, is what really matters – is neither simple nor straight forward. LTV is driven by lifetime duration (LTD) and future purchases. How much a customer is likely to spend (on average per purchase), how often they’ll do so, and for how long will together determine their LTV. To the extent that these may have been poorly estimated, the accuracy of any subsequent analyses will be compromised.

That huge challenge aside, what’s even more difficult is to qualify and quantify the relationship between discreet business decisions (primarily strategic but sometimes tactical) and their back-end results. In other words, assessing which business drivers had what direct and specific impact on LTV. For example, even if we can reasonably estimate that LTV has increased, say, 15% overall, how do we tie this increase back to a particular driver when, in fact, many may be at work? Was it our redesigned website, product line expansion, more restrictive (or liberal) returns policies, or new factory that is primarily responsible? Whether a particular strategic driver had a positive or negative affect is difficult enough to discern. Assigning its quantitative score is typically next to impossible.

Therefore, while it’s perfectly valid to assert that it’s the impact of a given business driver on customer lifetime value that’s most important, it’s just as important to recognize that leveraging this principle is exceedingly difficult due to the sheer complexity of the numerous interactions taking place – especially internal, but also external as well (e.g. the actions of competitors).

Our approach (which we call “Deltalytics”) is to periodically estimate the average customer lifetime duration, the average customer spend, and subsequently track their change over time to expose trends that will ultimately govern future business performance. If we know that both are increasing, for example, it’s safe to say that the business is trending upwards. The rate at which this is occurring can, of course, be used to make specific predictions about future growth (or decline, as the case may be).

But these two metrics are just the tip of the iceberg. Others that can and should be used to gauge business performance include:

(1) Rate of new customer acquisition (and, conversely, attrition)

(2) Customer distribution by recency (the greater the proportion of recent buyers, the better the business will perform)

(3) Average latency (the sooner customers place subsequent orders the better)

(4) Customer distribution by frequency (the higher the better, although not nearly as predictive as recency)

(5) Multi-buyer conversion rate (the percentage of 1X buyers who become multi-buyers)

(6) Customer re-order rate by recency (the ratio of repeat buyers as a function of their recency segment, e.g. <30 days, 30-60 days, etc.)

(7) Customer reactivation rate (customers flagged as having lapsed but eventually reordered)

(8) RF Delta (the change in population density over time at the intersection of recency and frequency)

Although quite useful in themselves, the greater utility in each of the above metrics lies in evaluating and forecasting their deltas over time. Change, and the rate thereof, is far more meaningful and insightful in this context than the more common “static” approaches to predictive analytics.

Getting back to business drivers (and measuring their impact on the bottom line, viz. LTV) one must concede that no single solution or approach can effectively gauge them all. At some point, a seat-of-the-pants determination must be made based upon relevant, albeit inherently incomplete, data. Tests can be conducted to measure the impact of, say, introducing a new product line or instituting wide scale changes in pricing. But even then, other contributing factors that cannot be controlled for, and are likely to cloud the results, must be acknowledged (such as a new website, outsourcing the call center, and so on).

In a perfect world, strategic changes would be implemented in a linear and mostly piecemeal fashion to ensure that consistent and reliable analyses can be made. Because this is so rarely possible, however, some compromises must be made in terms of measuring and forecasting the impact of such changes.

It is our position that an optimal way to approach the problem is to analyze trends amongst the aforementioned business performance measures – more specifically, their change (and rate thereof) over time, and subsequently tying these back, as best we can, to their underlying business drivers. This, unfortunately, is much easier said than done.

Lloyd Merriam

No comments: