Using AI to put the best plate forward

Subscribe:

Be honest. The last time you ate at a restaurant, did you photograph your plate before eating? If you’re a millennial, probably. But don’t be ashamed. Post it to a restaurant review website and you could be doing the business a solid. 90 per cent of customers consult an online review platform before visiting a business. And if you leave a review with a beautiful picture, people are more likely to engage with it and become customers themselves.

User-submitted photos are critical to restaurants’ success on online review platforms. But when hundreds of people submit their savoury snaps, they won’t all look equally appetising. So how can restaurants use these photos to put their best plate forward?

Enter AI.

Hyunji So is an assistant professor of Information Systems at the Desautels Faculty of Management. In a recent paper, she and her co-authors (Warut Khern-am-nuai, Maxime Cohen, and Yossiri Adulyasak) tested whether AI could outperform humans at picking good photos to advertise for restaurants on review platforms like Yelp or OpenTable.

The answer was a resounding “yes.” AI-selected images were up to 16 per cent more effective than humans at picking photos that generated user engagement.

“It’s important for restaurants and the review platform to consider what kinds of images will keep users engaged,” said So.

She and her colleagues studied a restaurant review website in Asia. Businesses on the platform have a page where users can leave reviews and submit images from their visit. If other users find a review useful, they can upvote it. By and large, the most popular reviews also include the most aesthetically pleasing photos and the ones most likely to win over new customers. The most upvoted review photos are featured on the restaurant’s page as cover images and preview photos in search results.

However, user votes may no longer be the best way to choose featured images.

Ending the popularity contest

Popular restaurants benefit the most from this voting system. Because they already have a large client base, they receive more user-generated reviews, and more votes for those reviews. This increases the odds that the user images will contain appealing qualities. Smaller restaurants, though, don’t enjoy this luxury. Fewer users mean fewer people to submit and select good-looking feature photos.

This is where AI could help. During the study, So and her colleagues deployed an AI that analyzed all uploaded images on a restaurant page and picked the best ones to use as featured images. The AI picked photos based on qualities widely associated with “good” photography: lightness, colourfulness, saturation, colour harmony, among other characteristics. The AI then quantified these elements into an aesthetic score. The five images with the highest aesthetic score were selected as featured images for a restaurant.

The AI did its job. Users were more likely to click on AI-selected photos than crowd-selected ones when searching for places to eat. This was especially useful to smaller, less popular businesses, who might not have otherwise attracted those viewers.

An AI photo revolution

This might seem like a narrow use of AI, but the implications of these findings are far-reaching.

Researchers have been aware for a while that AI can outperform a single human at processing large swaths of information. But these findings go a step further. They suggest that AI can also outperform a crowd when making aesthetic decisions, explained So.

So’s application of AI on restaurant review platforms can also be transposed into other areas of marketing. Advertisers could use similar technology to select the most engaging photos to share on social media, pamphlets, or billboards.

“These are just some of the most obvious examples,” said So.

As with most things related to AI, the future will probably surprise us.