Tuesday, January 16, 2007

Types of Lifetime Value Models Part 2

Yesterday I listed four types of lifetime value models, based on combinations of complexity (simple/complex) and scope (partial/full). Each has its own characteristics:

- simple, partial. This is a simplest of all possibilities. Inputs describe just one type of activity—say, purchases—and the calculations themselves are simple. Such models are quite common. They include, for example, a typical formula that calculates lifetime value as revenue per year x years per customer. Because of their simplicity, they are largely limited to general strategic guidance on broad questions such as “what’s a new customer worth?” In addition, watching the input measures for changes can give hints of problems or opportunities worth exploring, but the measures are so general that significant shifts in underlying components are easily hidden. (For example, a small but important shift in product mix might not have much impact on the over-all numbers.) This sort of model is a reasonable place to start but doesn’t get you very far. To be of real use, it must be backed by a richer set of drill-downs into the formula components.

- simple, full. This uses a wider variety of inputs but still with simple calculations. Examples would be a lifetime value model built from a customer-level profit and loss statements, with different inputs for acquisition cost, revenue, bad debt, cost of goods, fulfillment, and so on. Again, such models are quite common, particularly among financial analysts who are used to the profit and loss statement format. The richer set of inputs allows more detailed outputs including cash flow projections and net present value estimates. To the extent that the input components map to specific experience types (prospecting, initial purchase, product use, service, etc.), these models can provide some insight into the relative cost and value of each experience. But the simplicity of the model’s formulas still prevents any exploration of subtle relationships among these experiences. For example, even if you know that customers who had a service problem renew at a lower rate than those who did not, this model cannot include that relationship. The use of these models is still therefore largely strategic, although they can extend beyond simple insights to include business planning.

- complex, partial. This uses more sophisticated formulas against a limited set of inputs. An example would be a system that uses detailed projections of sales by product line, married to fixed assumptions about acquisition, fulfillment and service costs. Such systems can help whatever department is providing the detailed input—perhaps, in the product sales example, to better understand cannibalization and cross-selling. They have less value to the company as a whole. In fact, they can be harmful if they miss important relationships: say, between a change in product mix and increased service costs. Without knowing that relationship, a revenue-based system might lead the sales people to push products that actually lose money because of higher service expense.

- complex, full: obviously this is the most comprehensive type of model, and also the most useful. By definition, it does capture the subtle relationships among different experiences. In Customer Experience Matrix terms, it fulfills the goal of “monetizing” each experience by calculating its net impact on customer value, including the downstream impacts on other experiences. Some forms of these models are built with deterministic formulas, although the best approach is probably agent based simulations. The real challenges with these models are assembling all the data inputs, standardizing them, integrating by customer, and analyzing them to uncover their relationships. Once built, the models can be used in many ways at both departmental and corporate levels: strategic planning; investment and resource allocation; forecasting; and tactical decision-making. Like any lifetime value model, they are most useful when frequently supplied with fresh data so they can be rerun to identify trends and deviations from expectations.

This bring us back to the question I started with yesterday: How detailed must a lifetime value model be to be useful? As we’ve just seen, each type of model is good for something, but my original question was in the context of helping with customer experience management. Perhaps disappointingly, the answer has to be that only complex, full-scope models can capture the long-term consequences of an experience. The top level result can still be a single, simple figure for lifetime value, but it must be backed up by sophisticated calculations on detailed data to be meaningful. Significant shortcuts are probably not going to work.

This should certainly not be an insurmountable barrier, either to lifetime value modeling or customer experience management. As we’ve already seen, companies can gain a great deal of utility from less demanding lifetime value models. Starting with those also helps them build understanding and expertise in creating the more advanced models. Even though the 2x2 matrix suggests a sharp delineation between the categories, there is really a continuum, so one type of model can evolve into another. In terms of customer experience management, simple lifetime value models can help train people throughout the enterprise to focus on lifetime value, even if they can’t yet connect it directly to experiences. Conversely, experience-oriented management provides many benefits even without lifetime value measurements. In short, companies can build both their experience management and value modeling skills in parallel, and connect them later when both have reached suitable stages of maturity. What’s important is to recognize that this connection must eventually occur, and plan for it during the early stages of the process.


Ron said...

"How detailed must a lifetime value model be to be useful?"

As detailed as it needs to be. (sorry -- hope you picked up on the sarcasm).

But seriously, my recommendation to any firm starting out on this path is that they need to LEARN their way to complexity.

They may decide that they NEED the complex model... but if they're going to spend 9-12 months developing the model...then they risk getting stuck in the mire.

Better to develop a quick and dirty model, see what it can (and can't) do for you.. and iterate to the next level.

David Raab said...

I always think of the line from The Simpsons: "A sarcasm detector...now there's a useful invention."

I'm all for doing things incrementally, but in this case there may actually be some significant discontinuities. The techniques used to build a simple model are quite different from those for a complex model and so are the applications. It's just not clear one can evolve into the other. As an analogy, think about predictive response modelling: companies may have lots of experience with conventional segmentation methods like RFM scores, but it's still a big leap to build that first regression model. You usually end up hiring an outside consultant to do it.

In other words, if you really NEED what a complex model does, building a simple model as an interim step won't move you very far in that direction. But as a practical matter, the things you need a complex model for are themselves complex, like enterprise-wide customer experience management. So you are probably going to have to evolve all the related people, process and technologies over time anyway. Building simple models in the interim does provide other benefits including familiarization with lifetime value concepts, which itself helps pave the way for more complex models even if the specific techniques must be replaced.