This is part of this year’s Next Big Thing series, tracking the latest technological developments set to be a bigger part of marketers’ lives. Check out what’s new and emerging in other areas here.

This story originally appeared in the October 2019 issue of strategy.

In recent years, every sector, from finance to QSR, has amassed and aggregated data from various sources to uncover themes around unmet consumer needs, says Meghan Nameth, managing director for customer marketing and product analytics at PwC.

But identifying patterns within defined sets of structured data is “actually the easy part of AI,” notes Anne-Marie Castonguay, who heads up data and insights at Lg2. Now, the next big thing is deciphering unstructured data, such as documents, audio, video, social chatter and product reviews, which are usually qualitative in nature and far more difficult to parse.

Machine learning algorithms typically use structured data – clearly defined data types that can be easily categorized, stored and searched in a database – to function. But recently, Hugo Thibault, VP at Stradigi AI, says the industry has moved away from using tables of columns categorizing the data it thought was important to track to using “unstructured data sets that captures a whole lot more.”

In 2017, Unilever showed the potential of exploiting new data sources in pursuit of product innovation when it launched cereal-flavoured ice creams under the Ben & Jerry’s label. Using AI to analyze song lyrics in the public domain, the CPG giant discovered as many as 50 songs that associated ice-cream with breakfast. Consumers’ existing affinity for early-morning frozen desserts was backed up with other research, and Ben & Jerry’s soon had a product others have since imitated.

As AI becomes more sophisticated, many companies are beginning to understand the potential of unstructured data to inform their decisions. Advancements in natural language processing has enabled AI to utilize these complex, qualitative data sets to more easily “read, understand, classify and find clusters of information,” Thibault says. This should then influence how products are developed and communicated to their customers.

Last year, Tommy Hilfiger partnered with IBM and the Fashion Institute of Technology on a tool that analyzed customer responses to its clothing line, as well as fashion trends around patterns, silhouettes, colours and styles – using 15,000 Tommy Hilfiger product images, 600,000 publicly available photos of fashion shows, and 100,000 patterns from various fabric sites – to help inform the design process.

Similarly, Avon announced in August the launch of a new mascara based on the findings of its Genius Algorithm, an AI-based tool created in partnership with agency Media Monks that enabled it to read, filter, process and rank thousands of social media comments to determine the features consumers want from a mascara, such as “volume, length, lift, definition and dramatic black colour.” While those tend to be the characteristics most people look for in a mascara, Avon has said the big finding was that customers don’t want to choose one feature over another.

Early on, AI required setting clear search parameters, such as keywords, to help machine learning algorithms pull relevant insights from data, according to Line Atallah, SVP at Weber Shandwick and a former VP of marketing at Keatext, a Montreal-based AI company that specializes in unstructured text-based data for consumer applications. But advancements in natural language processing has enabled it to make sense of information outside of established parameters, helping to strip away some of the human bias. This change, she says, has enabled the discovery of potential “blind spots” in product development processes.

Atallah, for example, has worked with a brand in the recreational vehicle space that discovered a disproportionate amount of negative online comments about the company that contained the word “beer.” After digging deeper, the brand discovered its vehicles were poorly equipped to keep beer cold during transportation. Unable to share specifics, she says the brand has since incorporated that insight into product design. “No human in the world would have put beer as a keyword when looking for a defect in the product,” she says. “That’s one example of how AI can help discover the unknown.”

To date, few companies outside of Google and Amazon have been able to apply what is known as deep learning – a more advanced subset of machine learning and one that powers self-driving cars, for example – into their business practices, Castonguay notes. Unlike machine learning, deep learning learns through its own trial and error. Few companies currently have access to the amount of data required to make deep learning a reality, except of course Google, Amazon, Apple and other data titans.

Still, Thibault believes the future potential of deep learning is “tremendous” and will increase significantly over the next few years, thanks in part to the proliferation of AI platforms, such as Stradigi AI’s own Kepler, which can forecast accuracy, make inventory predictions and product recommendations, as well as analyze social media.

What’s more, Brent Chaters, managing director of digital customer at Accenture, says the future will see different AI platforms continue to integrate with the goal of enhancing consumer experiences. For instance, IBM has worked with Salesforce to combine its Watson AI platform into Einstein, the CRM company’s own AI offering. And Google and Amazon have made some of their AI capabilities available to smaller businesses.

One of the biggest challenges facing brands is correlating their own data with external ones to create Big Data sets that can be fed into AI, notes Castonguay. “Some people are pretty good at finding insights,” she says. “But using those insights on a daily basis [to] then inform your business – we’re not quite there yet.”

PwC’s Nameth adds that doing so enables companies to understand their customers before they even become customers, because AI tools can help “merge structured and unstructured data to identify clues as to who might be a potential customer… based on the experiences you’ve had [with] your current customer base.” Combining various data sets is where the value of AI rests for marketers, she adds.

Thibault sees a recent series of AI acquisitions this year as evidence of the potential of AI when it comes to solving business challenges. He points to a Gartner survey that predicts 85% of all customer interactions in retail will be managed by AI come 2020.

In September, McDonald’s acquired Apprente, a voice-based, conversational AI system focused on fast-food ordering that aims to increase the speed of transactions at the drive-thru window. Apprente, which will be folded into the QSR’s Silicon Valley-based McD Tech Labs, says it can handle “complex, multilingual, multi-accent and multi-item conversational ordering.”

News of the deal came six months after McDonald’s acquired Dynamic Yield – a machine learning company focused on personalized experiences – for more than $300 million. McDonald’s is expected to use the company’s decision-making algorithms to help tailor its drive-thru menu based on weather, real-time restaurant traffic and trending menu items. And in an interview with tech publication Wired, McDonald’s CEO Steve Easterbrook suggested similar predictive tech could eventually connect back to the kitchen, driving efficiencies right through the supply chain.

Outside of QSR, in February, Walmart absorbed AI startup Aspectiva – which specializes in personalized product recommendations based on a combination of reviews and shopper browsing behaviour – in a bid to enhance its ecommerce experience to better compete with Amazon. And in August, Nike acquired Celect, a player in ecommerce AI, to help boost its predictive analytics and better anticipate inventory needs using structured and unstructured data, as part of its ongoing “consumer direct offense strategy.”

To Thibault, examples like these suggest brands are “leveraging the whole journey from production to post-consumption to collect data to ultimately optimize the overall experience.” Companies often start with two or three use cases to prove ROI, he says, before moving towards a more complete AI roadmap.

“You’re seeing a lot of these organizations going out and they’re not just looking to buy tools. They’re actually looking to embed capabilities into their organization, which is a smart way of thinking about this,” Chaters says. “The organizations that are exploring this as part of their DNA are probably building for the future.”