Training A.I. to Understand Humans Better

It’s no secret that we are living at the brink of a new age of computing and artificial intelligence. A.I. and machine learning (M.L.) seem to surround us at every turn as more businesses tout their own A.I. technologies, signalling a market shift propelled by increase in both supply and demand of A.I. solutions.

But, as with any market craze, it would be wise for any prospective buyer to take a step back and really understand what these solutions are, what they aren’t, and what you need to know.

For starters, there is no single, all-encompassing “A.I.” The term more correctly refers to a field of research and an approach to computational and algorithmic design. Each “A.I. solution” needs to be designed to fit the problem at hand, meaning that each A.I. application will be unique. This not only means that companies need to invest in training their A.I. to solve for specific problems (i.e. vertical A.I. solutions), but also that failing to do so and relying on generic A.I. (horizontal A.I.) for every problem can result in misleading and even broken results.

To illustrate this, I’ll focus here on one of the most intrinsically human aspects of communication: emotional tone and language.

Detecting the emotional component in written text or spoken language can be useful in a range of applications. We have found that the emotional language in marketing messages accounts for as much as 60% of response across channels – email, Facebook, display, etc. This just means that emotions in language are the most important element in driving someone to engage and even buy (more so than product/offer features, calls to action, structure of the message, formatting of the message).

This is important, so let me be as clear as possible: The emotions in your message are more impactful than the actual features you describe in your top of the funnel communications.

Understanding this can be transformational for your business. If you remember the famous Mad Men scene about the Kodak Carousel (and if you don’t – watch it now), you would know that if you want to sell a carousel slide projector, you don’t talk about its technological features or how well it’s made. Instead, you try to create a sentimental bond with the product and what it can show you... you talk about nostalgia.

“It lets us travel the way a child travels,” Don Draper says. “Around and around and back home again to a place where we know we are loved.”

So how does A.I. fit into this? How can A.I. help me create powerful one-to-one emotional communication? And how can it help me convey them in personalised marketing messages to drive my business?

In the following case, IBM Watson illuminates what is possible in an interaction with a customer care agent. Let’s say the agent asked: “How can I help you today?” The customer might then respond in one of the following ways:

“Someone created an account using my email account."

Or…“This is not my account.”

Running responses through Watson can arm the agent with emotional cues from that customer wither on the call or for follow ups. This emotional intelligence enables the agent to deal with the situation more effectively and provide a more personal customer experience.

This is just one example of how A.I. can be a powerful tool. But when not trained towards a specific goal, it’s not nearly as powerful as you might be led to believe.

For instance, the main issue with current models is that they are not adapted to marketing. One of the holy grails in the field is the ability to create wording or propose images that will have the necessary emotional pull to motivate or inspire a person to take action. However, if you turn to a generic A.I., like IBM Watson’s horizontal API, and apply it to marketing messages, you’ll get little to no actionable insights; you may even get the wrong insights.

Here’s a simple example: Let’s assume that you wanted to evaluate the emotional orientation of the commonly used marketing message: “Attention please. Our offer ends today”. You’d get this:

Watson’s API qualifies this statement as Sadness. Does it sound sad to you? It doesn’t sound sad to me. In the context of your mailbox, this subject line’s emotional motivators are Anxiety (attention please) and Urgency (offer ends today). Watson is limited to the five generic human emotions and tries to understand what emotions are conveyed, but this simply does not work within the marketing context.

Let me give you another example: Let’s assume that your A.I. could analyse your customers’ written text via social channels and then use that to identify their personality propensity from a list like this:

What would you do if you knew that “John” (or someone similar based on segmentation attributes) was labelled with “Assertiveness”, “Imagination”, and “Openness”? How would you act on this information? What language would you use in your next Facebook ad to engage John? And how would you know that using that language is also how John likes to be talked to? And more generally, how can you be sure this is the right information to act on?

You might think you know these answers, but when it comes time to put pen to paper, you’ll quickly find that you’re still just guessing what to write.

If you want meaningful business results, you need is a different kind of A.I. You need one that’s been trained within the marketing context to detect what language and emotions inspire a given person. It may sound futuristic but it is possible today. Such an AI could analyse your customer responses to your communications, assign an emotional ID for each of your customers and then suggest the suitable language that better resonates with each of them.

It’s a bit like when you’re having a conversation with someone and adjust your language based on the signals you’re detect from them. A.I. can augment that capability so that you can communicate with millions of people at once and remember what works for each and one of them. If it did that correctly you would end up with something like this:

You’d first learn that this person (an ID, no need for personally identifiable information?) responds best to excitement, achievement, and challenge.

Your A.I. could then suggest the exact language to use if you targeted that customer, or type of customer, for a marketing communication. A display ad would change from something generic:

To one much more targeted and emotionally engaging to each individual:

One-to-one personally relevant communication is the holy grail of marketing. When correctly associating A.I. language and emotions to drive personalisation, the results can transform the way companies and their products interact with its prospects and customers – both in terms of experience and business results.

Oh and by the way, and to be fair to us humans, here’s something a machine is not likely to understand anytime soon: “Before LinkedIn, I didn’t know any strangers”.