How AI can prescribe new opportunities in retail

The audience at the eTail Asia 2023 conference heard how social data can be turned into consumer insights for generating new retail ideas.

• Social media’s visual-heavy medium with picture and video formats is a goldmine for retail marketers.
• Consumer data from social media can deduce trends, understand what type of offers will work and know how to amplify them.
• Generative AI makes it easy to create content to publish on social channels and trigger consumers emotionally.

Why it matters

Predictive analytics give marketers an idea what is going to happen, but prescriptive analytics tell them what to do about it, with improved social intelligence as large language models go beyond demographic markers to look at aspirations and behaviours instead.

Takeaways

• External social data can be leveraged beyond internal data from point of sales transactions for consumer intelligence.

• Social media can now be used for real-world segmentation in a way that naturally reflects real-world groups.

• Large language models (or LLMs) are able to see data without languages and understand it as if it’s in the same language.

Retail is one industry that is big on visuals, as Mimrah Mahmood, VP and partner at social and consumer intelligence provider Meltwater, sees it. This explains why social media, a visual-heavy medium with picture and video formats, is such a goldmine for retail marketers and why Meltwater currently mines it to the tune of 1.3bn social posts daily.

“When our system sees the data, we deduce people, brands, scenes and colours”, explained Mahmood at the recent eTail Asia 2023 conference in Singapore.

“For example, when you type in our system (to ask), ‘What is the hottest blouse or outfit of the day?’, we can tell you which colour is the hottest on Instagram. This is the power that is allowing us to unlock consumer data in a completely new way.”

Captured in real time, social data can quickly be turned into consumer insights for generating new retail ideas.

“For retail especially, one thing we have been using is a seven-step process where we use consumer data available in social media to deduce what the trend is, understand what type of offer is going to work in the market at that time and how to amplify it.”

But it doesn’t end at trendspotting.

“The cost of content is going down. With generative AI, the creation of content becomes much easier. Banner ads can be created on the spot.”

He said this is where Meltwater sees the biggest opportunity.

“We can look at the trend, create the content, publish on social channels and really excite and let that emotional trigger happen for your consumers.”

Planning for demand

Mahmood singled out the case of how 7-Eleven Japan’s planning team was able to anticipate sudden surges in near real-time demand for specific product categories through social data.

This was done through extensively tracking “every single thing you can think of that might go viral in Japan” – from sweets and characters, to trending items across over 100 categories and over 10,000 sub-categories.

“What this allows us to do is to trigger a signal at the right time”, said Mahmood.

As such, 7-Eleven Japan stores could beef up their inventory ahead of stock order deadlines to prepare for the surge in demand.

“The retail franchises can be told, ‘Hey, you need to stock this item for tomorrow because there is a large conversation gathering and for the next three days, there will be people coming into your store looking for it.’”

This allowed 7-Eleven Japan to capture demand that they otherwise would have missed.

“With this, we have been able to unlock new categories for 7-Eleven, categories they have never thought were possible. This is purely because of using content available online to really help understand the trend, how long this trend is going to be and how they can connect it to their own data.”

A total of 12 or so key trends were captured, leading to an expansion in new sales categories for franchise stores.

Besides higher order volumes, increased sales, optimised inventory and reduced stock shortages, the exercise proved to have one invaluable advantage – C-level buy-in, as the benefits of leveraging external social data beyond internal data from 7-Eleven’s point of sales transactions for consumer intelligence became apparent.

From predictive to prescriptive

While the use of big data technologies traditionally honed in on the diagnostics aspect, things are changing with prescriptive AI-enabled recommendations at scale.

“We started rolling out predictive analytics in the last four years”, said Mahmood. “Predictive analytics does give you some type of idea what is going to happen. But we are starting to go into the prescriptive type of analytics and this is the sweet spot. Prescriptive analytics tells you not only what it is about but what to do about it.”

Social intelligence looks set to get even better with large language models (LLMs) in particular. Mahmood predicts that what will unlock opportunities for most companies is not the OpenAI-type of platforms but local LLMs.

“Organisations can run it (LLM) on your own data, for or with a large amount of internal data. We help overlay that data. We are talking about bringing unstructured external data into the organisation.”

He added that this data can “sit within your local server”.

“There is a lot of talk about government policies and privacy but to run it through a neural network to create a better file, you are not touching the data. The data never leaves. You are looking at a better file.”

One emerging opportunity, for instance, is a much more sophisticated understanding of consumer segments that goes beyond demographic markers and looks at aspirations and behaviours instead.

“For the first time, social media will allow you to do real-world segmentation”, said Mahmood, explaining that this psychographic manner of approaching consumer segmentation naturally reflects real-world groups.

“If you look at Star Wars fans, it doesn’t matter what age, what gender or what location they are in. Star Wars fans are across the board. That is where you can use social media. You can look at historical data, look at every single mention of Star Wars and see for the past five years what kind of breakfast they had. We can tell you what type, what kind and what colour that breakfast is.”

Breaking down language barriers on social

Emerging markets in Asia represent a huge growth opportunity. But capturing high quality data in these markets is also notoriously difficult due to language barriers.

Herein lies a golden opportunity for fast retail: by picking up weak signals with potential to turn into huge trends through social posts, brands can develop products with relevant attributes for different markets in line with customer preferences – even before the trend happens.

LLMs lend a powerful advantage as they see “the data without languages and work on all languages equally well and understand it as if it’s the same language”.

“For the first time, it doesn’t matter if we look at this data in Japan or Jakarta, we can look at the data instantaneously. The team in Japan gets it in Japanese but the trigger is in the local language (Bahasa). That’s the beauty of it.”

Traditionally, the analysts would read through the data in the local language, translate it to English and send the report to Japan where it then gets translated to Japanese.

“By the time this all happens, the trend has already gone. But now we can do it in real time.”

Ultimately, while LLMs might be the answer to overcome many data challenges in data or online engagement, there remains a caveat.

“Some people ask me, ‘Does this mean I can let go of all my analysts? The answer is no.” Mahmood added that LLMs require a lot of computing power.

“You are running on huge datasets, so it’s not cheap to run this. But it can find the needle in the haystack in a couple of seconds. It doesn’t have the barrier of languages. It allows us to do user segmentation and integrate much more seamlessly. It’s more expensive, but it can get the job done.”