Tuesday, January 22, 2013

2012 Election: Big Data Makes It Easy to Shoot Yourself In The Foot

A recent post in MIT Technology Review about the use of Big Data by the parties in the 2012 election offers fascinating insights into what works – and what doesn’t.  Both sides had operations specifically designed to glean new insights about voters and apply them to helping their candidate win, and the article interviews the director of each effort shortly after the election.  As a result, we see long-term Big Data projects whose results have just been tested against the reality of the voter “market” and whose project reviews are relatively untainted by post-project “spin.”

I should note that I view the article’s author’s grandiose claims about the meaning of the Obama campaign’s relative success in Big Data analytics to be overblown.  Luckily, many of those claims take up only the first 20% of the article. The rest provides a fascinating insight into two approaches to Big Data, one of which demonstrably worked better than the other.

Problem #1: Big Data as a “Hot Topic” vs. as a Strategy

The first contrast between the campaigns was that the Obama one chose to bring the analytical systems in-house, whereas the Romney campaign chose to use existing software and hardware run by someone else.  In fact, the Obama campaign specifically committed major bucks to a Vertica solution way back in 2009.  This begs the question, however – why did the two campaigns take such different approaches? After all, the Romney campaign actually started quite early, around the same time, and had no real constraints on money spent.

The answer, I think, is that while Obama viewed in-depth analytics and trolling the Internet for more voter insights as a strategy, while for the Romney campaign, the concept of Big Data was fuzzier, more of a “hot topic” type of strategy.  We have all seen CEOs who have said, “Everyone’s talking about this new technology. We must therefore do it.”  The flavor of Romney-campaign thinking, as reported in the article, was, “Big Data is clearly effective at the corporations we look at; we must therefore get something that purports to do Big Data and use it somehow to understand voters better.”

The result was a cautionary tale.  The Romney system involved a third party merging two sets of legacy-application data repeatedly.   As a result, breakdowns occurred in the stress of the late campaign, and analytics was unnecessarily delayed, delaying candidate responses. The Obama campaign suffered from no such problems. It is not clear how much the Obama campaign’s faster response time helped (although it apparently helped change Obama’s style after the first debate).  However, it is clear that the Romney campaign’s fuzzy and delayed analytics helped lead to Problem #2.

Problem #2:  Hearing the “Customer of One” vs. Hearing What You Want To Hear

One of the great shocks of the campaign was that, in an unprecedented way, the Romney campaign and the Republican party as a whole deluded themselves about what was going to happen.  And yet, the campaign’s Big Data analysis mined a long tradition of understanding voter blocs, and could be supposedly cross-checked against a broad array of pollsters – most of whom were themselves deluded by their political biases.

As it turned out, even the closest estimate underestimated Obama’s margin by about 0.5%.  Moreover, on election day, the Romney campaign actually thought they had a better than even chance of winning. However, as it turned out, to win they would have had to amass more than 5.3% more of the vote in a couple of battleground states than they did, at minimum. To put it another way, they probably would have had to change the vote percentages by more than 5% nationwide to win.  They weren’t even in the same ballpark.

By the way, for Nate Silver fans, he was off by about 1.5%.

There appeared to be two key insights that made this election different.  First, perhaps one-third of people voting could only be reached for polling via cell phone – and they voted in a significantly different way.  Second, only perhaps 5% of the electorate reached the Labor Day start of the campaign with a changeable vote – again, a major difference from 20-40 years ago.  Gaffes, debates, Hurricane Sandy – it didn’t make that much difference. 

Given this, the Romney campaign’s one faint hope was voter turnout – unprecedented turnout for him in battleground states. And yet, there was not the slightest appreciation of this in the campaign.  Instead, the Romney campaign or related groups spent enormous sums on TV ads in the obvious markets, to the point where one station in Ohio was running more ads than programming. The crime wasn’t that this was wasted money; it was that relatively little money was spent on “get out our vote.”

In fact, most of the Romney campaign Big Data folks’ attention was on a series of seemingly baffling Obama ads in small markets targeting small demographics.  As a result, by election day, the Big Data folks were suffering from the “fog of war”, trying to counter something they did not understand.

Again, by contrast, the Obama campaign was focused on a new insight garnered from their use of Big Data to identify the “customer of one”.  More specifically, they realized that although most votes were unchangeable, smaller pockets of “conservative” voters could actually change their mind if policies they valued and prioritized were brought to their attention. These could be reached by, say, social-program-highlighting ads aimed at conservative women in more rural Ohio markets.

An under-appreciated part of these ads was that they contained real content.   After all, “conservative” voters were being saturated with generic Romney ads.  To succeed, the Obama ads needed to point to real, verifiable programs.

And now we come to Problem #3.

Problem #3, the Endgame: Big Data Tunnel Vision vs. a Loosely-Coupled Strategy

The end of every political campaign is “get out the vote” – maximize the number of ones supporters who actually vote on election day (and, of course, the increasing percentage who vote before then). And this is where the most startling difference between the campaigns’ approach to Big Data emerges.

It is apparent from various reports that on election day, the Romney campaign was prepared to focus its “turn out the vote” efforts on key demographics in battleground states, fine-tuned by what their Big Data analytics was telling them.  And then, right when voter turnout efforts were supposed to start, their feed from their system went down.  And it stayed down, for most of the day.

But what almost passes belief is what the campaign did about that.  Local offices were begging to go out and do “get out the vote” efforts anyway.  Instead, they were apparently told to wait until the system came up again.  Yes, theoretically, unfocused efforts could have done more harm than good.  Practically, however, it was overwhelmingly likely that “feet on the street” would have instead achieved slightly higher Romney turnout. And not only national but local campaign coordinators failed to realize that.

The problem, it appears, was that no one had autonomy, and all were focused on the Big Data part of the “get out the vote” strategy.  By contrast, the Obama campaign – partly due to an existing Internet-enabled strategy – granted local offices the ability to act proactively, and the Big Data “customer of one” focus was only part of a broader effort in battleground states to get out key Obama demographics, which were already well understood pre-Big Data.

Again, it is worth noting that this made little difference “on the day.” It is possible that it made a difference to several House seats; but not enough to overcome the effects of the 2010 election and its resulting Republican-dominated redistricting.  Note that there was apparently the biggest difference between the vote for each party (Democrats +1%) and the House seat allocation (Republicans +4%) ever recorded.

Implications for Organization Big-Data Use

We have seen, above, how the Romney organization used Big Data to shoot themselves in the foot – and yet, their strategies were superficially reasonable and well aligned with the practices in many businesses.  The immediate recommendations for IT and the enterprise are likewise relatively straightforward:

1.       Treat Big Data as a strategy, not as a “hot topic.”  Understand that for it to be successful, it must provide greater depth and a more accurate view of the customer and of one’s own organization, and those insights need to be translated to fine-tuning of strategies sooner rather than later.

2.       Focus on Big Data’s ability to understand customers not only more deeply as global “types” but in finer-grained groups, and ensure that the organization accepts a more realistic view of the customer.  Bluntly, the Romney campaign started believing their own propaganda; where have we seen that before?  And one reason was that Big Data was not telling them any different.

3.       Adopt an agile marketing strategy that does not hang on command-and-control top-down implementation. Agile marketing has a great respect for in-depth data.  However, it also has a great respect for the way that data reflects or fails to reflect actual customers, and customers who are constantly changing.

  And that brings us to our final point.  As I’ve noted, the success or failure of Big Data efforts turned out to matter surprisingly little to the outcome of this particular election.  However, Republican post-mortem efforts show that their failure to understand its implications has drastically delayed – and we are talking more than a decade here – their adaptation to changing American demographics.  They have created a party culture that makes it extremely difficult for them to move to anything more than permanent minority status nationally, because it carefully widened the divide with groups such as Hispanics in the name of older white-male “get out the vote.”

The point I am trying to make is that if Big Data cements an organization’s existing un-agile strategies in place, it is doing just as much harm as good – even though, for now, things are going better than ever.  The real value of Big Data, done well, is that it not only enables you to understand your customer better, but it also enables your organization to fit itself better to the customer – bottom-up as much as top-down – and both of them to understand how the customer is changing, not just what the customer is like now.

So what is it going to be?  Are you going to use Big Data to shoot yourselves in the foot, or to deliver better strategies, better implemented?  Inquiring Republicans want to know.

1 comment:

Jonathan L. Yarmis said...

I really think calling it "big data" has done us all a disservice. We need to turn "big data" into "timely knowledge." http://doctordisruptive.com/2013/01/18/big-data-is-wrong-on-so-many-levels/