This article was originally published on MarTech Today by Barry Levine.

If you think self-driving marketing platforms are like self-driving cars, analyst David Raab disagrees. (He will be participating in a presentation on “Self-Driving Campaigns” at our MarTech Conference in October.)

Self-driving cars, safely maneuvering through every conceivable traffic condition the world can deliver, are further along and essentially exist, he told me recently.

But the day when you can say: “Computer, give me a campaign.” Ummm, nope. Not anywhere near there yet, Raab said.

The key reason is that they’re not doing the same thing.

Analyst David Raab

Oh, sure, self-driving cars are managing a ton or two of metal hurtling down roads with highly vulnerable humans inside, under all kinds of real-world conditions.

But that’s not as big a computational challenge as a self-driving marketing platform, Raab said.

Driven by increasingly powerful artificial intelligence of one flavor or another, marketing tools can intelligently present the right content, predict companies or individuals who are most likely to buy, carve out user segments ranging from millions to single individuals, recommend movies you might like, optimize the most engaging combination of approaches, and much more. AI-based platforms like Persado can even write marketing emails.

But Raab’s point: “These are all point solutions, [and]there’s nothing to knit it together.”

As he wrote in an email follow-up to our conversation:

When I think of “self driving campaigns,” I’m thinking of a campaign like a Marketo nurture flow — that is, multiple interactions with branches based on customer behavior and other things that might change. A customer journey could be considered several of those campaigns combined, or one really big campaign flow. Those are the things I doubt AI will be able to construct from scratch any time soon. Generating and optimizing one-message projects like a paid search headline (Persado) or email blast (Amplero) or product recommendation (lots of things) is vastly simpler.
Those point solutions might communicate with each other, but that’s a kind of signaling or data exchange between peers.

‘Models break a lot’
“Think of a marketing department, with specialists,” he said. The department also has a manager who conducts, coordinates and modifies the activities of the specialists, plus deals with the real world.

There are various kinds of orchestration engines, such as Usermind, or workflow automators, like Zapier. But they require you (the human) to define the journey and the campaign, and, when conditions change out there in the world of buyers and sellers, you have to oversee the redirection.

Raab noted that AI systems are trained on historical data, from which they generate models to predict and guide future actions.

“But the minute there is a change that makes the data obsolete,” he said, “the training is obsolete.”

“In marketing, models break a lot.”

Let’s say a flood shuts down a city, meaning a campaign has to shut down or dramatically recast itself. Or the spokesperson for the product in question is caught in a compromising situation, so the campaign needs to be diverted. Or there’s a recall on some of the product models, so the campaign needs to be rewritten quickly around a new product revision.

“At some point,” Raab told me, “things change beyond [the data that the system]was trained on.”

A meteor?
But, I countered, isn’t this the same as a meteor falling from space into the path of a self-driving car? Wouldn’t that car also be at a loss in the face of an unpredicted set of conditions?

To the car, he said, the meteor is just another obstacle to avoid. “But there is always a new condition in marketing that is not anticipated.”

The best expected setup in the foreseeable future, Raab said, would be a system that “fails gracefully” by handing a marketing effort off to a human manager when conditions change radically, the way some chatbots now hand off the conversation to a live operator when it gets too complex.

As he wrote in a recent blog post on the subject:

It’s easy — and fun — to envision a complex collection of AI-driven components collaborating to create fully automated, perfectly personalized customer experiences. But that system will be prone to frequent failures as one or another component finds itself facing conditions it wasn’t trained to handle. If the systems are well designed (and we’re lucky), the components will shut themselves down when that happens. If we’re not so lucky, they’ll keep running and return increasingly inappropriate results. Yikes.

This article was originally published on MarTech Today by Barry Levine.


About Author