I’ve been working in some form of database marketing or analytics for about as long as I’ve been working. After I graduated from college with my degree in Philosophy, my most marketable skills were a pretty decent knowledge of programming micro-computers (you know – PCs) and an older brother with a genuine skill for managing political campaigns. Those combined to get me started in political direct marketing – taking voter registration databases, census data, and survey research and crafting direct mail campaigns.
How did we do it?
Let me start by saying that much of it was ridiculously simple.
Political campaigns are won in two basic ways: getting your voters to turn-out and persuading the small group of undecided voters to break your way.
When we targeted direct mail, we almost always had one of these two simple goals in mind and each started with data.
How do you identify “your” voters? It’s actually pretty easy - at least back when I was doing it. Your voter registration data (which, by the way, is public) tracks which party primary you vote in and how you declare your political affiliation.
In most voter registration files, we’d see a field called voting history – and it looked something like this:
This field recorded every year you voted, which primary – if any – you voted in, and whether you voted in the general election. If you have reliable history of voting in a particular party's primary, you’re extremely likely to vote for that party’s candidate in the general election. So the simplest and most basic analysis we did was to separate out voters into three basic segmentations: likely to vote our way, likely to vote the other way, and a third group that was hard to predict.
On the overwhelming majority of campaigns, we completely ignored the “likely to vote the other way” group. It’s nearly impossible to evangelize the enemy. That’s an important lesson in all marketing. You can’t afford to waste time and energy on people who aren’t interested in your message.
What about the “likely to vote our way group”?
The strategy for this group typically involved two phases. In the first phase, messaging concentrated on fundraising and encouraging volunteers. Money and time fuel political campaigns, and while most of this comes from human networks, it was possible to build and supplement those networks with aggressive marketing. Companies looking to build a social strategy should bear this in mind – and, in fact, it came up in an interesting fashion at a private Think Tank class I was teaching this past week for a client.
The Think Tank class was a new one I’ve developed on Database Marketing and Web analytics, and we were talking about creating and understanding Web site marketing cues. I described how companies can use social features on the site (reviews, comments, likes, etc.) to identify potential social interactors and target them to support viral campaigns. One of the class members then pointed out that it might make sense to add social functions to the site specifically to help identify this extremely valuable population.
It’s a great idea – once you start thinking about using site content to cue marketing decisions, you realize that you have considerable control over how many and which of those cues are actually available on your Web site. Identifying social interactors may be one of the most potentially impactful uses for social site functionality.
Note that we didn’t bother persuading this group, we enlisted them as part of the team.
The second phase of our messaging to this group was reserved for shortly before the election and concentrated on “getting out the vote.” Messaging strategies for this ranged from “patriotic duty”, to FUD and attack pieces, to candidate paeans and, my personal favorite, free doughnuts. Doughnut mailers have to be completely non-partisan and can mention no party or candidate. In them, you offer a free coffee and doughnut (in San Francisco I suppose it would be a Latte and Croissant) for voting. And, though you can’t mention anything the least partisan, you CAN send them only to people you know will vote your way. Doughnut mailers were, in my day, potentially decisive in off-year elections with small turnout.
Looking back on these experiences, they seem like an extraordinary precursor to social media strategy. Political campaigns have always been (and remain) far more social and viral than nearly any comparable corporate effort. It makes me think that companies looking to build social strategies might consider looking to political campaign talent for help.
Which brings me to the third and final group – the “hard to predict.” This is the group that got, on most well-run campaigns, the vast bulk of attention. Campaigns that can be decided are nearly always decided by which way the 10-14% of “hard to predicts” actually break.
How did we target this group? Let’s start with what little we knew about them. We knew their name, address (voter registration records) and we knew their voting history, their age and their date of registration.
By voting history, I mean we knew how often they voted and which party primaries they had selected. Our “hard to predict” population generally included all New Registrants, people who didn’t vote in primaries, and people who had voted in primaries from both parties.
The first step in our targeting process was data enrichment. There were (and there are probably more now) a myriad of clever techniques for this. We had vast dictionaries that, using a person’s name, would assign gender and, with pretty good accuracy, ethnicity. From address, we could take advantage of the consumer household databases that were then beginning to emerge. This often allowed us to add marital status, occupation, income range, and a host of other potentially interesting variables.
From address alone, however, we could do almost as well with that treasure trove of public data – the U.S. Census. Published at the BLOCK level, census data provides counts of ethnic makeup, income, education, family status, age, and much more. Where you live – when taken down to the block level – often says more about you than any other single targetable variable.
This date enrichment gave us the building blocks with which to target, but we still needed the data for driving the targeting strategy. That came from opinion research; classic, call-em and ask-em opinion research.
We pulled our opinion-research samples directly from the Voter Registration databases – appended phone numbers to them – and weighted our surveys to the “hard to predict” population.
Survey data driven from Voter Registration data gave us the critical link between the variables we had (New Registrant, Voting Pattern, and Individual and Census Demographics) and the key variables we cared about: intention to vote, partiality, and, most important, key attitudes. Though we asked (and cared about) the “horse-race” question – “Who are you going to vote for?” survey data was most important for providing the link between the targeting variables we had and the messaging that might work.
After data enrichment and survey collection, came – you guessed it – analysis. We used a range of fairly simple statistical techniques to create key segmentations. These might be as simple as a univariate selection: “all women”, “new registrants”, “voters under 25”, or “asian-americans”, but they were often two or three way cross-tabulations (“women under 25”, or “low-income, mixed-primary seniors”). These were typically created using either simple correlation tests or basic regression models. Nothing terribly fancy, but it didn’t need to be. The segments we created were VERY significant and, equally important, were understandable. That meant the writers could target creative appropriately.
At the conclusion of the analysis exercise, we presented a strategy to the campaign teams that described the key segments in the target population both in terms of their targeting demographics and the attitudes they had. In most cases, we were trying to identify the issue or attitude that was most important to them and that reflected our candidate’s general opinion. That issue would be the basis for messaging.
Writers then developed customized letters for each target population and we created coded Name/Address lists to match. The mailhouse added the operational component, and you had a database marketing political campaign.
For the really big campaigns, we were able to repeat these steps iteratively (survey being the most expensive and time-consuming step), tracking the actual impact of our messaging on the target population.
It was a thing of beauty when done well, and it worked. It definitely worked.
My intention is not, however, to sell political direct mail consulting. I described this process in great detail because I think it’s a nice illustration of the fundamental workings of traditional database marketing.
Distill it down, and you get this process:
- Start with a Population (Registered Voters)
- Enrich your knowledge of that population (if necessary) with data that can be tied to targetable properties such as attitudes, needs or desires (Name Lookup, Household data, Census Data)
- Macro-Segment your population into fundamental Groups (enemy, friend, undecided)
- Create a Macro-Strategy for each Macro-Segment (Enemy:ignore, Friend:Recruit, Undecided: Persuade)
- Create a link between the targetable properties, the behaviors you want to incent, and the messages most relevant to each combination of target group and desired behavior (Survey data in our case – most often with additional behavioral data in commerce targeting)
- Develop Creative
- Target and Message
- Measure impact
- Iterate from 5
This, in a nutshell, is traditional database marketing. What makes it hum, the places where all the magic happens, are Steps 2 and 5. Step 2 gives you the tools to target. Step 5 tells you how to use those tools.
So what does all this have to do with Web analytics?
I started off this series with a look at the fundamental principles that I believe underlie classic Web analytics. In this post, I’ve laid out a basic process for database marketing. In my next post, I’m going to show how the two overlap and where our distinctive online data presents both challenges and opportunities when it comes to creating a true digital database marketing strategy.
[If you're interested in this whole Digital Targeting topic, I'm going to be speaking directly to it at WebTrends' Engage 2011 (March 1-2) in San Francisco - it's a very enjoyable Conference - and fairly inexpensive. A good use of your Conference dollars if you're in the Bay Area or, most definitely, if you're a Webtrends client].