Build an Insight Engine: Impact Weightings

It’s nice to know how customers feel about things, but in isolation, this information tends to be quickly forgotten. To make it last longer, it helps to tie it to the things everyone cares about, like money. That’s what our Insight Engine is designed to do.

To build this connection end to end, it actually helps to start in the middle, between how customers feel and your bottom line. The relevant question here is, “What are the things that customers need to do to drive the value of your business?” Are your revenues generated directly from customer sales? Or do you generate revenues from advertisers that pay to sponsor your site because consumers visit it? How much does your business rely on people spreading the word?

It usually doesn’t take much to make a list of all these desired “behaviors.” For example, “buy product A,” “use service B,” “refer friends,” “pay on time,” etc. There are some challenges to this exercise when it comes time to isolate their independent impacts on financial performance, but I’ll save that for a later post. For now, a basic idea of customer behaviors will do.

In this glorious internet age, it’s become an increasingly easy (and cheap) exercise to gather reliable, multiple choice-type survey information from current and prospective customers.  If structured properly, we can get out of this one of the key ingredients in our model: measures of how the different aspects of mindset impact the likelihood of customers to behave in the ways we’d like them to. If we survey enough people, we can even look at how these levels of impact play out differently among different groups of customers (e.g., how different people are motivated in different ways).

I can’t state enough how important it is to get the right mindset measures in place. That story exercise I keep harping on really makes a difference! I did a bunch of work in the ‘90’s with one of the original contributors to the Service Profit Chain, an ambitious model that looked to quantitatively link investments made in employees to how “satisfied” employees were to how they performed their jobs to how “satisfied” this made customers to how customers then behaved to how this impacted the bottom line. Whew! It was always interesting to put this all together for companies, and we always found evidence that this chain of events does measurably exist. But sometimes we would get excited about relationships between factors that weren’t all that big (that would say, for instance, that one factor in the chain could be seen to account for <20% of the outcome of the next factor). A lot of studies done since have found that “satisfaction” isn’t generally the best predictor of behavior (although all of the suggested alternatives still suffer from being generic, rather than business-specific, mindset indicators). While using even weak metrics can still demonstrate important impacts, using good metrics can actually make the model useable.

From the research, we should be able to walk away with a decent set of starter values for our model that link how people feel to how they are likely to behave.  And if we’ve gathered information from enough people, we should have separate sets of impact factors for distinct groups of customers. The model will be strongest if we choose to distinguish customers (to segment them) based on these impact values*, as they take into account both their motivations and their behavioral tendencies. Just as getting the right mindset measures is a critical factor, defining customer groupings in just the right way is also a critical driver of our engine’s efficiency.

While the survey method is a great way to get us started, the strongest model will have evolved over time from measures of actual customer behavior. With a well-developed measurement framework in place (and, after some time up and running, the utility of the model clearly demonstrated), we then have reason to build out data collection and analysis processes aimed at updating and refining the measures of impact in the model.

_____________________________________________________________________

*Footnote: Technically speaking, if we’ve structured the research so as to be able to statistically derive impact coefficients (vs. asking respondents to provide them outright) then these variables don’t exist in a one-to-one relationship with respondents so can’t be used in a segmentation exercise. However, there are ways to cluster respondents based on their combined answers to individual mindset and behavior questions and thereby “sort” them according to impact similarities