Online photo printing company Photobox stores a lot of photos—around six and a half billion of them. That’s around nine petabytes of data, and they ingest anywhere between three and five million more every day.

Richard Orme, CTO, Photobox Group

That gives the group, which comprises Moonpig, Photobox, Hoffman and Poster XXL, an abundance of data that can be used to improve the customer experience using AI.

Photobox’s initial use of AI is simply identifying what’s in the content. Essentially, AI is used to trawl through photos uploaded by customers and identify markers that can be categorised. A long white dress, for example, could indicate that someone is creating a wedding album.

"While that's interesting to some degree, that doesn't really help our customers,” said Richard Orme, Photobox Group’s chief technology officer.

“What we then do is say, well, if we understand the content, can we look at a historical dataset and compare the themes of the content that have been uploaded with what we have from other customers at other times and understand the context.”

Orme says that AI goes further than basic categorisation, though. It feeds in all of the information from its dataset and the demographics of the logged-in users to determine what your role is in the album’s story, such as who you might be at the wedding.

While Orme admits that might sound creepy, AI is not comparing customers’ photos with other customers, they are comparing them to patterns from previous uploads.

“We're in a reasonable enough shape to say this is a set of wedding photos or this is a set of travel photos, or this is a set of new baby photos, for example."

AI that tells your story with you, not for you

The underlying concept of Photobox’s AI is for it to tell the customer’s story with them, not for them. When customers were presented with an option that can create an entire album automatically in a matter of seconds, they rejected it.

"They actually want to spend the time doing it,” said Orme. “That's surprising to a technologist who is like 'quicker, faster, faster', but what it helped us to realise is that actually we should use the AI to solve the emotional challenges our customers feel."

One such emotional challenge is the anxiety customers reported when first starting a photo album. This anxiety stemmed from the daunting prospect of 50 empty pages or the overwhelming choices available in design.

“It's up to us then to build a technology to be sympathetic to that and actually to help them be creative, so make suggestions rather than fulfil the customer journey."

“We should use the AI to solve the emotional challenges our customers feel.”

Photobox instead used AI for the low-value, monotonous work such as removing duplicates and blurred photos. AI also helps to overcome the first hurdle by suggesting photographs and layouts, which Orme describes as allowing the customer to be the 'director of the movie', having a say on the final cut.

However, research by Photobox found that 45% of 16-24-year-olds believe that AI will be better at finding a gift for a loved one within the next two years compared to just 13% of over-55s, suggesting that future generations may be happier to relinquish more control to AI.

Can AI be emotionally intelligent?

AI can automatically assemble a photo album, but its decision-making process is going to be very different to that of a human. Where an AI would create an album based on categories, such as a colour or theme, a human would make emotional decisions that make no sense to an AI.

But could that change in the future? Orme thinks it unlikely.

“Human beings are very complex and our emotions are driven by an incalculable number of inputs on any given day.

“So do I think AI will truly love, or hate or fear? I think probably not. But I think we'll probably be able to do a reasonable job of making you think that it is. Whether or not we should do that, I think is the interesting question.”

“Without techniques like machine learning we couldn’t recommend products other users have bought but we will be able to do that with our technology later this year.”

However, Orme does believe that AI can be used to determine someone’s emotional state, which could help brands understand the impact that the technology is having. This could be used to provide feedback to brands and allow them to modify its approach in real-time.

The ability to do so is some way off, though, and many obstacles remain. It is commonly acknowledged, for example, that diversity is needed in the technology industry. Diversity becomes increasingly important when designing AI.

"We have to get more opinions, we have to get diverse perspectives as we're building these technologies. We have to get more diverse datasets to train them on to truly represent what is culturally correct and what's ethically correct."

Image courtesy of Abd. Halim Hadi /

Share this article