Why Your Analytics Is Costing You Time and Money
Modeling (or analytics) for selection is way overrated and much of the time it doesn’t beat a simple RFM bucketing in Excel. What the sector really needs is a model for understanding but more on that in minute.
The modeling (or analytics) for selection is about using transactional – and perhaps demographic or lifestyle data from external providers – data to determine who is most likely to give again based, largely, on past giving.
Despite the circular logic and diminishing returns this does work; it can yield campaign level efficiencies with lower costs and higher net revenue. This is what all or almost all modeling and analytics in the non-profit sector is focused on answering – who do we select for this campaign to maximize a short term campaign metric?
However, there is a rather insidious (mis) application of this data and the “analytics” being applied – especially when combined with external demographic and lifestyle data – when we start thinking we are answering “why” or providing understanding. Said differently, it is dangerous at best to think differences between responders and non-responders using past giving or demos or lifestyle tells us why one group is in a given bucket.
So why is this so potentially dangerous and misguided?
Here is a point of fact: If you take any donor database and add demographic (age, gender, income) and lifestyle data (e.g. coin collectors, blue hat wearers, subscribers to this magazine or that one) to it and look for patterns that make responders and non-responders different you will find them, in reckless abundance.
The question to be asked is whether any of these patterns that show responders looking differently from non-responders is remotely useful in understanding why one group gave and the other did not. If the information was only used to further refine selection then all would be forgiven.
Unfortunately, many charities are taking data useful for selection and making huge assumptions about cause and effect. This leads them to, among other things, tailor a marketing message in acquisition based on a random difference – e.g. responders are more likely to be coin collectors than non-responders – and compound the error with a wild guess about how to act on it – e.g. offer coins as a premium, or stamps with a coin image or other twists of logic.
More examples:
- Finding that non-responders just happen to be of a lower socioeconomic status and deciding to adjust the ask-string on a subsequent mailing.
- Seeing that non-responders are much male and Republican and deciding to gin up some more “male and Republican” message.
- Finding that you can group responders into 12 (or 100 or 1000) different buckets of response rate and that if you look at these new groups you find that one of the best responding segments indexes very high for being educators and so you elect to focus on how your charity works with schools to educate kids on issue x.
These are all real examples. They share common traits of being,
- Incredibly random
- Being just a select few of the countless patterns one can find and
- All having nothing to do with the difference in response.
- They are merely coincidental and again, you can find piles and piles of coincidence in every donor database if you have misuse data.
So, what is the alternative? For starters, start issuing red flags anytime your internal data person or modeling “expert” or analytics person is trying to segment for “cause” or understanding using data on your CRM and/or that which can be purchased externally. This information can be used for selection and efficiency but not understanding. You can save your charity countless hours and money by not chasing these shadows.
Additionally, start looking into the role of attitudinal data, which is the only data that can tell you about needs, motivations and preferences. We can see the collective eye roll from at least part of the readership on this point. It is true not all survey data is created equal. Most surveys are really poorly constructed and yield information that is as dangerous to use as the stamp collector data append about our responders. However, this fact should not undermine the equally true fact that answering “why” is only doable with properly collected and analyzed attitudinal data.
Taking this one step further and by combining attitudinal data with transactional one can build a model that assigns financial value to all your touchpoints across all your channels. In short, an answer to what you do and provide and communicate that actually causes loyalty and the decision to stay or go.
This is the only type of model – attitudinal plus behavior with a theoretical framework guided by how the nonprofit to donor relationship works – that will provide you with understanding, Understanding what causes behavior is akin to navigating your organization using a GPS with turn by turn versus the stars in the sky