UK supermarkets to trial age estimation tech for alcohol purchases

Digital identity platform Yoti is in talks with major UK supermarkets to pilot its biometric age estimation technology for the sale of alcohol.

The trials form part of a government initiative – led by the Home Office and the Office for Product Safety and Standards – to expand the use of digital identities in the UK economy, and will allow the organisations involved to develop a “regulatory sandbox” for the sale of alcohol using age verification technologies.  

Regulatory sandboxes, such as those being developed by the UK’s information commissioner, are test environments that allow software to be trialled in real-life situations under the close supervision of regulators or other oversight bodies.

Yoti is in talks with the UK’s 10 largest supermarkets, as well as a number of e-commerce and convenience stores, about providing its biometric age estimation software in self-checkouts. All participants in the trial will need to work alongside local licensing authorities and police to ensure the deployments are legal and safe.

Although Yoti is unable to confirm which supermarkets are involved in talks at this stage, it is partnered and integrated with NCR Corporation, a US manufacturer of checkout machines that supplies Tesco, Sainsbury’s, Asda and Morrisons, among others.

Proposals on the development of the alcohol sandbox can be made until 31 May 2021, with trials running until April 2022.  

Using cameras embedded in the checkout machines, Yoti’s facial analysis technology works by comparing a customer’s scanned face against its artificial intelligence (AI)-powered algorithm to determine whether the shopper is old enough to buy alcohol.

Unlike facial recognition systems, which establish a person’s identity by comparing a real-time scan of their face with a pre-existing photo, Yoti’s facial analysis system does not store any biometric information, either locally or in the cloud, and immediately deletes the scan once a person’s age has been verified.

John Abbott, chief business officer at Yoti, told Computer Weekly that the technology could help both supermarket staff and customers save time, which will not have to be wasted checking IDs.

He said that although each individual customer may not have to wait that long for an ID check, for major supermarkets with 20 million-plus customers, the time spent on these checks adds up. “The way we tend to structure the business case with the retailers is around reclaimed time for the staff to use, and it should also provide queue saving, in other words a time saving, for the customer because they are not waiting around,” he added.

“If you’re in a big supermarket, you are likely to wait, on average, two or three minutes for someone to come along and check your ID, and that’s a bit compounded because the reason it takes that long is because they’re probably checking someone else’s.”

Abbott said that, in terms of accessibility and inclusion, the tech could also be beneficial to people who do not have physical forms of ID, such as a driving licence or passport, who are otherwise unable to buy alcohol, as well as staff having to deal with abuse from customers for declining a sale of alcohol.

To train its facial analysis algorithm to estimate age, Yoti used anonymised passport data, such as images and birth months – which were taken with consent from the 10 million or so people globally who have installed its digital identity app – alongside user-provided selfies.

By running this dataset against its algorithm, said Abbott, Yoti was initially able to predict someone’s age to within an average of four years, but can now do so within two and a half years, and this number is falling as the technology becomes more accurate.

He said the error rate was even smaller for people aged 16 to 24, to within 1.8 years, because it is easier for the algorithm to detect changes in a younger person’s face due to fewer variations having occurred in their bone structure or skin up to that point in their lives.

“It’s never going to be even for every skin tone and gender, and so forth, but it’s getting the right level of data, the right level of quality, and then being able to explain what some of those differences are,” he said. “We publish it by age group, by skin tone and by gender, so you can start to see the conclusions and where the pockets of bias lie.”

Abbott said that although the technology is least accurate when it comes to estimating the ages of 40-plus women with very dark skin, this is because of good complexions and fewer wrinkles, rather than their actual skin tone. “We need to be very clear in publishing that sort of data,” he added.

According to Yoti’s whitepaper, its technology has the highest “mean absolute error” when it comes to estimating the ages of dark-skinned women aged between 50 and 60, estimating to within 5.6 years of their actual age.

In the UK, the Retail of Alcohol Standards Group requires retailers to check whether customers appear at least 25 years old, which is where the threshold for the system would be set, said Abbott.

“What regulators say is, ‘right, for an 18-year-old, we’re going to certify you for 25’ – in other words, we are confident that you could set a threshold of 25, and you have a statistically very low likelihood of serving 17-year-olds,” he said.

But if a customer aged over 18 is not recognised as meeting the threshold age of 25, there would also be an option for people to onboard their driving licence or passport to the Yoti app, which would allow them to share anonymised date-of-birth information by scanning a QR code, said Abbott.

For making purchases in corner shops or pubs where customers can buy alcohol but there are no checkouts, the app can also be used to save a digital ID card with its hologram, something UK law currently requires customers to present when buying alcohol.

Both of these could be used as alternatives for people who do not consent to having their faces scanned at checkouts, and customers will still have the option to get their IDs checked manually.