You may not know the name Lee Sedol, but you can add him to the list, along with chess champion Gary Kasparov, of early casualties in the battle of Man vs. Machine.
Kasparov lost a chess match in 1997 to IBM’s Deep Blue, but experts in artificial intelligence and “deep learning” thought it could be another decade before a machine beat a human at the ancient Japanese strategy game of Go, which has an average of 250 possible moves in every turn, compared to 35 for chess.
It happened March 9, when a deep-learning AI called AlphaGo defeated Sedol, an 18-time international Go champion, four games out of five.
Machines aren’t only beating humans at games. We hear similar stories from almost every area surrounding our lives: finance, e-commerce, even music and sports. But there’s at least one area where machines continue to struggle.
B2B demand generation is still driven pretty much by gut instinct, sales intuition or a kind of herd mentality where marketers play it safe, investing in the same tools and tactics as everyone else. But those traditional tactics for creating market demand are showing rapidly-diminishing returns and marketers are under more pressure to deliver.
With machine learning—powered by big data—taking over in so many fields, B2B CMOs are being brainwashed to believe machines will get them out of this frustrating cycle of blind spending and chasing sales impulses.
Almost every day I meet and talk to early-adopter CMOs, demand-generation and marketing-operations experts who were willing to put their toes in the water and try predictive analytics. Some of them had big expectations and got carried away with the vision of being the first to truly revolutionize B2B marketing.
With all of those fancy new statistical terms suddenly becoming part of their lives, I hear mostly confusion, ambiguous results and decreasing faith that B2B marketing can go through the “data-science” revolution everyone is talking about.
Models that intend to predict who is likely to become your next client and when are 100 percent dependent on the data they are fed about those potential clients.
Or, in the words of a very smart guy I met when comparing predictive analytics solutions for one of the leading tech companies told me, “Math is math.”
If you have the right data and insights, math will do the work to correlate attributes, optimize, maximize and give you accurate predictions.
But if you don’t know much about your prospective client, or what you think you know is actually wrong, advanced mathematics, i.e. machine learning algorithms, can’t be effective—just the opposite. They can point you the wrong direction.
The good news is that B2B can still rely on sales to tell you when machines are wrong. And for now it seems the machines have minimal success in beating sales intuition in predicting who will buy and who won’t.
Here are some of the most common examples of how bad data biases the machines.
Machines don’t know the individual
Every sales rep knows that the likelihood to close a deal is dependent on the individual you deal with, and therefore they are obsessed with collecting professional, personal and psychological insights on their buyers.
It gives salespeople a huge information edge over machines that rely almost 100 percent on available company data. And if all of the individuals look the same, the machine will give them the same score and fail to differentiate between the true decision makers and the rest.
Self-fulfilling signals will predict the obvious
It’s common sense to assume that if a prospect is interacting with you a lot, he or she is more likely to eventually buy. Or on the contrary, if they don’t go through the phase of interacting with you they are much less likely to send you a signed contract.
But if a machine relies mostly on interaction data, it will just predict the obvious. They will score high the ones who are already on your radar and indeed will prove to convert. The prediction will be accurate but the value for demand creation will be minimal. Objective external data (“3rd party data”) is what truly provides an information edge.
Historical success is limited in predicting the future
All of the books trying to analyze what happened to Wall Street are trying to teach us the same lesson: blindly predicting the future with normally-distributed historical data misses the “black swans” out there and creates bubbles.
It’s the same for B2B marketing but even worse; historical data is extremely limited, especially when you try to predict big deals, while the markets are so dynamic with new competitors, products and buying trends being introduced every day.
Don’t assume machines can simplify a complex B2B buying process if they have limited data to learn from.
Cool, obscure signals vs accurate, “boring” data
Everyone is looking for an information edge—those hidden signals nobody thought about that will tell you before any one of your competitors that there is a potential client ready to spend now!
But while chasing after those signal “unicorns” that, to be frank, are often too noisy to rely on, we underestimate the value of getting the “boring” and “traditional” data right. Getting the basics wrong can dramatically skew the scores. Even worse, it makes sales lose faith in the machine. And when the sales team has lost faith, it’s very hard to win it back.
In future posts, I’ll go into more detail on each of these areas, and explore the ways that data and machine learning are transforming B2B demand generation. I’d like to hear your thoughts in the comments.