AI meets consumer insights

So, with a conceptual framework in place, let’s look at a tangible example showing how AI can analyze social media data in ways that are either much faster than manual analysis or were simply not possible with traditional research and data gathering methods.

Uncovering hidden opportunities

Based in Chicago, Levy Restaurants offers premium vending and food services to major entertainment and sports venues, such as the Scottrade Center in downtown St. Louis — home to the NHL’s St. Louis Blues. The company wanted to refine this venue’s restaurant concepts to appeal to the changing preferences of the city’s sports fans, especially in the coveted 20 to 30 age brackets.

AI-powered analysis of social conversations revealed a strong interest in fusion cuisine, which combines elements of different culinary traditions. Levy’s modified its restaurants to meet this demand, and within half a season, the same locations generated more revenue than in the entire previous year.

Only the convergence of AI, analytics, and social media made this possible for several reasons. Focus groups and surveys could never provide the breadth and volume of consumer opinion and preference that social media can. In-venue interviews might seem promising until you consider the logistics and the fact that you’d be interrupting the game experience, which could easily provoke some very negative social conversations.

Finally, these traditional research methods are time-consuming and resource intensive. In the time it takes to set up a focus group — let alone conduct it and analyze the results — AI technologies could analyze and classify thousands of conversations about food preferences at sporting events. It’s just no contest.

Using AI to customize data analysis

It’s easy to understand how AI can deliver speed, accuracy, and scalability. After all, artificial intelligence is machine intelligence. But what’s often lost in the discussion about AI’s power is its ability to help us understand nuance and context. Very important characteristics of human communication.

Understanding those characteristics means you can answer more complex questions. You can dig into why people behave the way they do and what subtle drivers shape their preferences.

AI can analyze qualitative data, usually in the form of text, by training algorithms to analyze sentiment at a very granular level. And the more data you have, the more sensitive the analysis becomes to the nuances you wish to surface. Three examples illustrate this capability.

1. If an auto manufacturer wants to associate certain values with its brand — exclusive, innovative, technologically advanced — analysts can teach an algorithm to recognize these qualities in text. Not just those keywords but language that invokes them. When you train an algorithm using a large volume of unstructured data such as tweets, forum posts or blogs, you can continually fine-tune the results.
2. Brands often need to train social listening tools to distinguish between words or phrases that have different meanings based on context. Let’s take the phrase, “My vacuum really sucks!” Does that mean the vacuum isn’t working properly or that it’s very powerful? Of course, it depends on context. Similarly, a company like HP — Hewlett Packard —  needs to distinguish between many potential meanings of those initials. These include abbreviations, acronyms, and slang like horsepower, Harry Potter, Hindustan Petroleum, high performance, high precision, House of Parliament, and many more. With text analytics, machine learning, and well-chosen training data, this can be achieved across many social channels.
3. Nuance doesn’t just exist in text. It’s also part of images. Every day, three billion photos are shared on social media. There are more than 800 million hours of video on YouTube. And camera-enabled smartphones have slashed the barrier to entry in terms of communicating via images. Deep learning analysis of images can recognize and classify logos, objects, actions, scenes and facial attributes. A consumer products company might use image analysis to not only find images that show its product, but also find items associated with the product, locales where it’s used, and the facial expressions of those using it.

Images also affect how we process information. In 2003, a Harvard student worked with a South African bank, sending 50,000 letters with offers for short-term loans. The letters varied the interest rate and included other psychologically-influential cues. It turned out that adding a picture of a happy female to the letter had as much positive impact on the response rate as dropping the interest rate by four percentage points!

The best of both worlds

As AI advances, it’s fair to ask whether it will eventually replace human analysts. Not anytime soon because algorithms still need the “human touch.” Algorithms work best on closed systems — a Go board, for example. The algorithm needs a human analyst to describe the environment in which it will run.

And, the consumer insights generated via AI-enabled tools will still be applied by humans. These tools will become smart assistants to analysts as their roles evolve. The speed, power, and pattern recognition capabilities of AI coupled with people pointing them to the right questions will create the best of both worlds.