AI in Retail Conference: success down to retailers cleaning up their act with data
Retailers are in an advantageous position when it comes to deploying artificial intelligence (AI) technology because they generate massive amounts of data but they have to be extremely careful that it is in a format and a location that it can deliver meaningful results when processed.
Speaking at last week’s Retail Bulletin AI in Retail Conference, Milos Milojevic, consultant on future of technology at Capgemini, told delegates: “Retailers are probably at the forefront of innovations in digital – and AI – and have been for some years. They have lots of data. As many as 40% of retailers in the UK are deploying AI at different stages. They are in a good place as they have lots of data already available.”
No shortage of data in retail
This is certainly the case with Karl Boyce, head of digital & CRM at Domino’s Pizza Group, who highlighted how his business is almost more of a technology company than a food service business, such is its use of data to drive its decisions and fuel its AI developments.
“AI suits our business as we trade in data. We have shedloads of data on our customers. There is also tons of operational data. Unlike Just Eat and Deliveroo we have customer data and operational data. AI is nothing if you do not have the data,” he suggests.
For Domino’s, Boyce says the objective of using AI is to ensure the company is at the “front of mind at the point of consideration”. “Over the last 18 months we’ve built platforms to ensure the data is in the right place and have used it to develop the relevant communications. We’ve integrated technology and AI into the most relevant places,” he explains.
AI powering all parts of the pie
This involves using AI to power a more intelligent strategy around PPC (Pay Per Click), which is important as search makes up over 50% of all the website transactions at Domino’s – worth £250 million per year. It is also powering social media, which has helped it to reduce the cost of hyperlocal marketing activity for customer acquisition from £2 to 50p through the use of dynamic video.
Boyce also points out that AI will be powering the future plans of the company through working with franchisees, which could enable them to drive sales to certain units by sending offers to their top customers in the area as well as diverting demand from a busy restaurant to a less utilised outlet nearby.
The capability of AI is also proving increasingly valuable to Michael Mrini, director of IT at hotel company Edwardian Group London, who says the need to engage with customers in the most efficient way possible led to the development of a ‘virtual host’ app called Edward that deals with customers – beginning with a welcome message ahead of arrival.
As well as dealing with the regular activities of check-in and check-out it also has the capability to handle myriad questions and demands from customers – including housekeeping requests, dealing with room service requirements, and taking restaurant bookings. “At first Edward did not understand certain words and misspellings but these have all been added as part of his learning,” he says.
Bringing together the data
The power of Edward is derived from the work Mrini did to bring together the many systems operated by the hotel whereby the data is in a single location. “We’ve brought the systems into a central location and made them talk together. Previously we had to run off printed reports for the relevant people from all the systems. It was a drudgery for employees to get this data. I thought – let’s free them of the paper. Let’s get all the info onto mobile devices and give them live data. We now have 30 mobile apps with every department having at least one,” he explains.
By using AI on this data it has been possible to bring in prioritised room cleaning, know who has extended their stay, and who has changed their check-in time. Providing the link between this capability and the customer is Edward that has become an increasingly valuable part of the customer engagement process.
Such a system has clearly added much value to Edwardian Group but for many organisations it is a challenge to determine what resources to commit to AI projects. Chris McGrath, IT and digital programme manager at Marks & Spencer, says ROI measures can be difficult with AI and at M&S he has looked at focusing on high volume, low complexity use cases.
Rules-based solutions might be best option
“With our inbound call centre we had a simple use case and it was about the better use of our staff to enable them to add value. There is no point in putting in AI when a rules-based system would be right 90% of the time. Is it worth the expense for the extra 10%?”
Natalie Konstantinova, lead software engineer in R&D at Shell Energy Retail, agrees: “Do you need something fancy or is it available as an API out-of-the-box? Most probably small retailers won’t achieve these capabilities with two data scientists. If out-of-the-box does 90% of what you need then does it matter about the 10%?”
In many cases it undoubtedly does not but what is certainly the case for all AI activities is the absolute requirement for quality data being fed into the system in the first place. Konstantinova suggests that 90% of data science is about cleaning data and organising it into a usable state.
Mark Terry, tech lead data platform at Holiday Extras, very much agrees that this is fundamental to any AI implementation: “The hardest part is the structure of the data. Structuring the data at the lowest level means the next level is easier and the analysis is better.”
Maximising value of data scientists
He acknowledges that this can be particularly challenging and with only a small team at Holiday Extras he brought in a system to overcome this problem. “We’ve legacy databases and practices and we had a small number of data engineers. They were always firefighting – fixing bad data. We moved the validation of the data down to the provider of the data and not the data engineer. They could then move from fixing to value add. The data engineers no longer own the incoming data, they own the outgoing data,” he explains.
Despite the challenges there are myriad opportunities to be potentially enjoyed from deploying AI but the reality is that retailers need to begin experimenting and proving specific use cases. They must also be willing to accept that not all their initiatives will be successful. McGrath suggests that the supposed failures could arguably be defined as successes.
“We ran an experiment at a cost of £25,000 and we proved that what we were looking to do was wrong. But that was not a failure as it probably saved us some millions of pounds. We need to rethink about the idea of fail fast. I don’t think these are failures.”
Words by Glynn Davis. Photo by Colin Fielder.
Make sure you join us at next year’s AI in Retail Conference on 14 October. Use the discount code RETAILER50 to qualify for the early bird discount of 50%.
