Customer Intelligence and the Automated Digital Storefront

by Ben Gaines

posted on 04-18-2016

As I watched the news streaming in from F8, the Facebook Developer Conference, one announcement stood out to me above all the others: the introduction of chatbots (essentially, programs that use natural language processing to understand and respond to conversation with humans on the internet and take certain actions in response; more on chatbots here) in Facebook Messenger.

The Messenger platform now boasts 900 million Monthly Active Users, and will be open to businesses as a new channel for automated commerce, customer service interactions, and more. Chatbots represent a new “automated” digital storefront; from ordering a pizza, to scheduling a test drive, to getting help with a software product, the goal for marketers is to allow consumers to have conversations with brands anywhere in the world, from the comfort of their own fingers. Additionally, the emergence of chatbots on Messenger (and elsewhere) promises the ability to offer a personalized, consistent brand experience at scale. What a time to be alive, right?

The reason that interest in chatbots is exploding is that natural language processing (NLP) technology is getting good enough to make “automated conversation” viable. I got my first in-depth taste of NLP last year when I worked with the research team at Adobe to develop DataTone, an application demoed at the Sneaks session at Adobe Summit 2015 that takes analysis questions (e.g. “How much revenue did the spring campaign generate for women’s shoes?”) and turns them into Adobe Analytics queries, then returns an answer to the user (“The spring campaign drove $24,491 for the women’s shoes category.”). Imagine that same ease of use, but for everything else in life!

That’s far from all, however; in fact, the dramatic shift in customer experience might not even be the biggest opportunity for marketers resulting from this exciting new technology. Looking at it from the perspective of customer data, given the widespread use of messaging apps, what I see in the emergence of chatbots is a tremendous boon nearly on par with the emergence of social media last decade.

Why ISN’T This Just Social Data 2.0?

Messaging data will be fundamentally different from both social data and survey data in a few key ways that are important to understand, since you may have social data and survey data flowing into your customer intelligence efforts already. Chatbots certainly won’t replace social media or surveys in your analysis, but will rather enrich the picture of your customers to which these sources are already contributing.

Let’s start with social data. Unlike social media, chatbots represent a direct brand interaction. I’m not posting a picture of myself in the restaurant; I’m talking to the restaurant directly. As such, the way I talk—the things I reveal about myself, either explicitly or implicitly—will be very different. Another important difference between social data and messaging data is that social media allows me to broadcast an opinion to my network, whereas messaging is one-on-one (or perhaps a-few-on-one). Again, this will impact the signal that I, as a consumer, provide to a brand. A customer might try using snark to tweet-shame his or her ISP for having slower-than-advertised download speeds, but when that same customer is one-on-one with an agent (human or otherwise), the tone of the conversation might be calmer and more conciliatory—or, of course, it might be even angrier.

Now, let’s talk about surveys. Surveys are typically served to a cross-section of a population (e.g., your customer base), and rely on the customer to remember a previous interaction once it is complete (“How did you like the product you bought?”). Chatbots, on the other hand, allow a brand to scale infinitely (for all intents and purposes) to have real-time engagement with a customer or prospect. No sampling required—chatbots can engage with every single person who wants to talk to a brand for any reason. Additionally, engagement with a chatbot indicates a signal of interest—in buying a product, receiving help from customer support, etc.—whereas surveys are typically randomly served, often to customers or prospects whose intent at the time may or may not be related to the survey topic. Lastly, the ability to get feedback and insight in real time, as the interaction is occurring, will change the content and tone of the feedback. A customer may express dislike of a certain option in a live messaging context, but not even remember disliking the option when he or she receives a survey two weeks later.

The New ABCs of Customer Intelligence

Over the past 20 years, marketing has gotten really good at understanding customer behavior across channels. Knowing what your customers did—whether online, in store, on the phone, etc.—is, of course, key to understanding your customer. However, it is only part of the puzzle. Last year, we added the Customer Attributes capability to Adobe Analytics, which allows you to merge attribute data from a CRM or other customer data source with behavioral data. Knowing demographics, psychographics, lifetime value, propensity to purchase, etc. tells you a lot about who your customers are and how to market to them—especially when combined with behavioral data.

Now let’s consider chatbots. These are brand-specific interactions. You’re not tweeting about the brand. You’re not walking through a crowded mall. You’re literally talking to the brand about what you, as a customer, want or need. Unlike on-site chat, there’s no friction; the engagement happens within a brand-agnostic app where users spend a ton of time already. There’s no waiting for someone to respond to your tweet; it’s nearly instantaneous. And, unlike in-store or call center touchpoints, it’s already digital. With each message, your customer is giving you direct, decipherable signal about who they are.

As it did with analytics in DataTone, natural language processing in a chatbot takes what can be an overwhelming task (“Uggghhhh, I can’t figure out how to add more toppings to my pizza on this coupon”) and makes it as simple as sending a text message or talking to Siri (who is, herself, an NLP application). In a messaging context with a brand, a simple chatbot conversation might look like this:

Brand: Hi, how can I help you today?
Me: I’m taking a trip to Hawaii and I need snorkeling gear for my family.
Brand: Sounds fun! We’ve got a handful of great options for kids and adults. How heavy duty do you want to go?
Me: It’s our first time in Hawaii; I’ve never really snorkeled before. What do you recommend for novices?
Brand: Hmmm. We’ve got three snorkels for surface-level exploration. Do any of these look good?
Me: I think my wife would like these blue ones a lot. How are the reviews on them?
Brand: Very positive. An average of 4.6 out of 5, with 438 reviews. Would you like to buy them?
Me: Sure, these seem great.
Brand: Awesome! Click over to our web site where I’ve placed this product in your cart already.

I’m not actually going to Hawaii—I only wish I were. But look how much signal about me is contained in just the three responses! In a direct conversation with this brand, I made it clear that I have a family. We’re going on vacation to Hawaii. I’m fairly new at sea exploration. I’m reasonably articulate (thank you, thank you). I care about product reviews. My wife likes the color blue. Imagine adding all of that quantitative and qualitative insight to my customer profile, overlaid on top of the attributes you might already have for me, and any behaviors I might have performed on any of your other digital or non-digital channels.

It remains to be seen exactly how data from chatbot interactions will be made available through Facebook and other platforms. One of the big concerns, of course, will be privacy; chatbot conversations are meant to be private messages like any message you would send through these platforms. However, there is likely a happy medium to be found, where brands can capture tremendous marketing value from NLP data, and where consumers can maintain trust both in the brands themselves and in the platforms through which they are interacting.

The ABC combination of Attributes + Behavior + Chat is going to make one amazing triumvirate for customer intelligence.

Bring On the Bots!

In the quest for scalable customer intelligence, the development and productization of reliable NLP technology—the automated digital storefront—stands to become a huge boon to marketers, potentially changing the quality and depth of customer understanding forever. While we don’t yet know how/when/where messaging data will be fully available to marketers by various platforms, including Facebook Messenger, the combination of attribute data, behavioral data, and chat data stands to enhance marketers’ views of their customers in ways that we’re only beginning to imagine. So I say bring on the bots! “Hello, Delta Airlines chatbot. I think I’ll have that trip to Hawaii now. . .”

Topics: Analytics, Retail