Photobox Group CTO Richard Orme spoke to CIO UK about how the online photo printing company is developing emotionally intelligent AI to help its clients create photo books; the decision to migrate its nine petabytes of photo data onto Amazon Web Services; and how the company's work culture facilitates horizontal and open channels of collaboration across teams to foster creative approaches to problems.
Beginning his career as a developer for Virgin Media, Orme has since held a number of senior tech roles, most recently acting as CTO at delivery management firm Metapack, before joining the Photobox Group - which comprises of brands, Moonpig, Photobox, Hofmann and posterXXL.
"Our customers on Photobox come to us with a very different motivation than the standard ecommerce side," Orme said. "Most of the time they're thinking about somebody else when they come to visit, or they're thinking about an occasion."
In the age of the smartphone, we take more photographs than ever - whether it's holiday snaps of crimson sunsets, jokey selfies with our partners or 'Gram-worthy' food pics. But for most of us, even though these pictures vary wildly in value, they sit bunched together, undifferentiated in our mobile photo libraries or gathering dust in our hard drives. Photobox helps people to turn these collections of pixels on our screens into physical, glossy photos we can hold in our hands, pin up on our walls or flip through in custom photo books.
As a result, this company handles billions of the digital records of our most cherished memories - a mammoth task for its servers, and online customer experience. Given the highly emotional nature of the content involved, the company has carried out extensive user research to identify the motivations and desires of their customers - with some surprising results.
"Our photo books on average take customers about two weeks to build," Orme said. "When we first looked at that, we were really taken aback by how long that was, and thought, 'Wow, shouldn't we be doing something to make that shorter and easier for customers?' When we tried that, what we found was they absolutely didn't want us to do that."
To work out why this was the case, the company developed a user research team made up of people with psychological training to help understand the motivation of customers arriving on the site.
"It turns out our customers not only come to us because they're thinking about somebody else, but as they start making these books, their motivation flips," Orme said. "They move into a very creative space, where all of a sudden they get to express themselves. This, for many of our customers, is the only time that they can sit down and be creative, and they get very, very protective of that. They're very conscious of the fact that this is their project - it's their output."
But the pressure to create something amazing for their loved one can end up becoming stressful to their clients. "They get nervous about, 'Am I making the best thing I could make? Am I on trend?' The most daunting time for our customers is when they open up this book, and it's just a series of 50 empty pages," Orme said.
Consumer research carried out by the brand confirmed that 31% of respondents experience anxiety around trying to find a present for a loved one - and turn to technology to help. Armed with this insight, Orme's team came to the realisation that it wasn't about reducing the time the client takes to build a photo book, "It's, 'how do you give the customer that creative control?' That allows them to still be the storyteller and the director of the movie, but helps them with all the low-value work that they need to do to get to that point," he said.
Some of Orme's tasks focus on how the company can help customers select photos for their books, as well as layout, themes and captions. Right now, customers are offered different options whenever they turn a page, known as 'smart fill' features.
"Ultimately, our goal is to produce an AI that's helping the customer tell the story. That's what we mean by emotional intelligent AIs," Orme said. "What is the story I'm trying to tell? Is it a wedding story? Is it a travel story? Who are the heroes in the story? How do I position those heroes in each of the photos, on each of the pages, for maximum effect? How do we learn from what everybody else is doing at scale?"
To do this, they can draw from the tens of millions of data sets they already have, following the client's journey through which photos make to cut to the executive decisions regarding the final product. Machine learning helps them to identify and verify the patterns they see in this data.
"When we bring that kind of insight and that kind of assistance to the creation process, we actually see a higher conversion rate from customers," Orme said. "They do go on to buy more from us. We believe that there's a great opportunity for us here, to create AI that doesn't shorten that experience. It still takes you two weeks, but at the end of it, you feel like you've told the best story you can."
AI is integrated into some of Photobox's products today, asking questions such as why certain photos were left out. The firm's 'smart fill' features are built on the patterns exhibited by previous customers. "As you build every page, our intent is that the AI gets smarter and smarter," Orme said. "Because it's learning from you, whether or not you're accepting the suggestions that it's making, or rejecting them."
The goal is to launch what is internally dubbed 'the magic book,' which will anticipate client's wishes and effectively tell the story alongside them. This will start being used commercially on the site at some point this summer.
Migrating data to AWS
One of the major IT challenges for Photobox is the sheer number of small files the company handles. This prompted the company to migrate this 9PB of data from its data centres to the cloud.
"The reason for us deciding to migrate to the cloud was fundamentally to do with the fact that we were starting to hit the edges as to what we could do, in terms of storage," Orme said. "We were starting to reach a point where we were spending a lot of time just maintaining the hardware and the equipment within the data centres that we had.
"That really then takes us away from the investments that we like to make, and the customer journeys, and the ability to innovate in the physical product space."
Orme notes that it was the personal touch that helped AWS win Photobox's business. "Just in terms of their support, in terms of their willingness to engage with us," he says. "I mean, they brought their CEO to meet us. It was a real sense of partnership that we were getting with them, and that made it a very straightforward choice for us."
The migration has gone smoothly, with the project on track to finish around five months ahead of schedule. However, there were some early teething problems.
"We found some constraints with the hardware devices that Amazon had given us," Orme said. But AWS quickly responded. "They worked around the clock with us, so night and day we were speaking to engineering teams in Seattle, at what must have been 4am their time. They changed their product roadmap in order to support us. Very quickly, within the space of three weeks, we got past that problem."
Photobox has recently moved into new offices in London, located at Herbal House in Clerkenwell. The office is open plan, with myriad meeting rooms and breakout spaces, structured around a spacious, central atrium. In the centre of the ground floor is a seating area facing a large screen. Orme says staff are encouraged to use this as a social space. He himself had brought two of his friends here at the weekend, and screenings of Game of Thrones and the Championship League will take place here on occasion after work.
The fluid boundaries between work and play also apply to the traditional segmenting of different departments, with the company recently adopting a DevOps approach to running the day to day business.
"Eight months ago, we completely re-imagined the operating model for the business," Orme said. "We moved to a squad model, which was a really big jump for us to make. The idea of a squad model is a squad of people who have very diverse skill sets, who would traditionally either sit in marketing, or software engineering, or testing, or infrastructure, or security, or commercial and pricing. Now, all working on one goal."
Squads are given a common goal, such as increasing traffic to the site.
"All that they are measured on and supported on is, how do you get to the goal?" Orme said. "That is exciting and scary in equal measure, depending on what you're used to, but I think that creates an environment that allows us to bring the very best talent in. Because it's a challenge that says, okay, this isn't just about how good are you at writing code, or what's your experience within the infrastructure world, or how good are you at DevOps. It's about how much of a challenge you want, in terms of solving a problem - because any idea within the squad is a valid idea that the squad debates, so we've removed hierarchy to some degree."
So far, the squad model is working well, with squads exceeding expectations. In the office there are spaces devoted to squad work, with their own breakout and whiteboarding areas. The squad model was first piloted within Photobox's sister brand, Moonpig, and resulted in £7 million in incremental revenue being made in the last financial year, as well as boosting employee engagement, the quality and speed of delivery of products.
Orme says that the squad model is one way to attract new talent, another is simply the engaging nature of Photobox's business.
"Number one, we as a group of brands have a fantastic customer proposition," Orme said. "We provide a service for our customers that has so much meaning for them. That when you go and tell somebody about it, they start to really light up themselves. They start to want to engage in that. So for a mission statement, as a group of companies, that was a big tick, right? In terms of going out and finding people, and saying, 'Hey, do you want to come and join us on this journey?'"