The Times reports on How AI helped frame designer

After a shaky start, Broughton, now 42, had to borrow money from his parents to keep the company afloat in 2014 but then he was offered £100,000 in investment from a happy customer and opened his first shop in Soho in 2015. Today the company has 16 UK shops, an AI-powered app and celebrity customers including Ariana Grande, Stanley Tucci, Paul Thomas Anderson and Idris Elba. Cubitts employs 140 people and expects sales of £15 million this fiscal year, up from £12.5 million in 2022.

How Cubitts pivoted to an AI and machine-learning-first model

Broughton did not start out planning to run a tech-led company but he now credits focusing on building in-house AI and machine-learning technology with helping to streamline operations and prepare for upcoming overseas openings. It began in 2016 after the company won a £32,425 grant from the government agency Innovate UK to develop a “cephalometrics” machine able to scan heads in “true depth”. This machine, although later superseded by other technologies, provided a great data starting point for a company innovating with glasses.

Cubitts has since invested between £500,000 and £750,000 in data, machine learning and AI technology and employs one full-time developer who works with four freelance contractors based around the world, from France to Argentina.

It has established a data warehouse and built up a dataset of more than 10,000 head scans that the team has used to train machine-learning algorithms. Today these inform Cubitts’ “virtual try-on” tool and determine how much stock it orders. The idea now is to build up this dataset while developing its algorithms and AI tools, and the company has also started using Generative AI (GenAI) to speed up content creation and marketing efforts.

“There have been two broad phases in how we’ve tried to incorporate AI and machine learning, or ‘deep learning’, into what we do day-to-day,” Broughton said. “The first phase was more about problem-solving and then more recently it’s been about business efficiency.

“It allows us to make better business decisions … We’ve found that a combination of big datasets, machine learning and AI can help make better, more accurate predictions than a human being could. And that’s important because the challenge we’ve set ourselves is: ‘How can we give a person who’s completely remote the exact same optical dispensing experiences as if they’re in a store?’ ”

Using AI to deliver efficiencies and drive growth

The “virtual try-on tool” in the Cubitts app superimposes different frames on users’ faces. Picture taking a selfie, but with a different pairs of glasses appearing on your face in your reflection. Broughton said Cubitts’ data shows that people who use the tool are five times more likely to buy a product than a standard customer and the company has a patent pending for its technology.The team is also testing a new AI tool for the app that can measure heads “to fractions of a millimetre”, which it hopes will help customers to choose better-fitting frames and result in lower return rates.

Cubitts is also using AI to inform how many pairs of glasses it makes in each size in new ranges, cutting waste and driving efficiencies.

Broughton said: “Historically if we were going to make a thousand frames, you would make 400 medium, 200 small, 200 large, etc. But what we can do now is actually map the product on to people’s faces and it can tell us we should make 32 extra-small and 117 small, so it means that it makes our production much more efficient.”

The company has also produced targeted ranges based on AI analysis of statistical probability models, which reveals potentially under-served customers. For example, Cubitts now has a range for people with narrow nose bridges and people with larger heads. “We wouldn’t have been able to do this without the AI, which led to an accurate dataset,” Broughton said.

Using technology to prepare for international expansion

Cubitts is opening a shop in New York in April 2024. It has been using AI and GenAI in the preparation process, which has already produced significant cost savings, it said. Broughton said that being able to use a ChatGPT-equivalent has eliminated the need to hire an additional US-based customer-facing team member, for example.

The Cubitts marketing team is using the GenAI tool to quickly write guides with standardised responses for a US audience. These will power chatbots operating on Cubitts’ US website and Broughton is confident that the chatbots will have learnt Cubitts’ tone of voice by the time of the US opening. “We’ll be able to offer a UK-like service but without having to build a localised customer service team,” he said.

The company’s marketing and sales teams are also using GenAI to create more than a hundred pages of search engine optimised content for a US audience. Broughton said that this process, which would previously have taken his team months, should be completed in weeks. “A year ago we would have sat down with our content team, thought about all of the titles and written content over many months. But now, once we’ve trained the ChatGPT-like engine, we can start producing that content very quickly.”

Advice to others considering investing in machine-learning software and AI

Broughton would advise anyone embarking on a similar journey to start gathering data as early as possible. “All of these tools are only as good as the data that goes into them and the bigger the dataset the more accurate and more effective it can be,” he said. “Even if you don’t know how you’re going to use the data, having really, really clean data is important.

“We ended up building our own data warehouse, and if I were doing this again I would have done that much earlier so that we would still have that data from those early years that we could be making use of now.”

He also recommended taking a similar freelancer-based approach to building in-house tech capabilities, as it is more flexible and allows companies without a huge budget to invest. “We’ve found that the work is very spiky,” he said. “There are times where we need lots of resource and times where we’re just doing user testing, we don’t need that much.”

Cubitts has come a long way since 2013 and Broughton is planning for its tech-led future. “The next ten years are really about how we can move from being a small, independent, UK and London-centric retail business into being a tech-driven global brand that’s genuinely innovating in this 300-year-old, very slow-moving industry.”