At June’s Masters of AI event, Peak’s in-house retail experts, Tom Summerfield (Retail Director and former Head of Commerce at Footasylum) and Manjit Curtis (Retail Customer Success Manager) talked us through AI in retail and the ways the industry is adapting and changing as we enter the new normal.
After the event, our retail dream team fielded a number of questions from attendees who were considering implementing AI into their businesses, but weren’t sure where to start. Here are some of the most interesting Q&As from the event, with Peak’s dynamic duo providing you with their cutting-edge insights and highlighting the main do’s and don’ts of implementing AI in retail effectively.
Who needs to care about AI and machine learning in a retail business?
Tom: I think it depends on the solution. Leadership buy-in isn’t crucial, in my experience, but you’ve got to be able to take people on the journey with you and educate continually. It spreads then. What I found was, if there’s people who don’t get it straight away, the business case and the numbers do the talking. Like anything, backing things up soon starts generating interest internally, which I think is the case for any new tech adoption in the business.
How would you take the customer on the AI journey? How do you get buy-in from teams?
Tom: It’s all about empathy. Everyone says “I understand you can optimize with AI,” but how does it come into my world?” There’s the vision piece, which is important, but you need to break it down with end users and how it’ll super-power their processes. When I was at Footasylum, it was about giving people ownership of elements of the AI journey. It can often feel a little ambiguous at first in its exact application, but get clarity early and people will start to see the value and embrace it. You can then start to see that culture shift across the business, that AI is a cool thing that adds value – especially when it’s backed up by results.
What is the explainability of an AI system like, and how important is that part of it?
Tom: It’s a black box for us. That’s where we win at Peak, because there is full visibility of the data sources and all the metadata that lives in there.
Manjit: I think sometimes it’s always building towards it, so as you build your AI models to work with people and to work with some of that data. I worked with somebody that was a non-retail customer but they loved seeing that analysis before we started to build the models. It really helps them to go. “are we looking at the right problem? Have we got the right data? Is it showing us the things we expected to see?” Sometimes, taking people on that journey is really important and then you get to the point, working with some of the customers I have, where they are more interested in the output, and the AI is thought of as a given – but people need to go on that journey for themselves.
Tom: When that explainability exists, it allows for more trading strategy to come from it because it kicks the retailer on to go “if we can do that, then what about this?” and so on. This is really exciting and adds that extra value to the explainability, the evolution of it and the collaborative approach that we use. The whole ambiguity around AI as a topic is resolved by its explainability.
Any advice around good use cases to start your AI journey and areas to avoid?
Tom: This depends on the business. I have found that starting small really pays off in the long run, and trying to take a lot on straight away is not that practical – and a reason why many companies fail when deploying an AI system. I think it’s important that you don’t become distracted and pulled onto a new project, but instead take small steps on a longer journey, where you see results pull through. This, in turn, will make it easier to get things moving throughout the company.
Right now it would be project specific to the demands that are present in your business. But, starting too big is not productive, whereas breaking the process down into smaller pieces is advisable. Then, culturally in a business, an AI system does draw attention and you can ride momentum through the business as people become more interested in what it is you are doing. That would be my one main piece of advice.
We have challenges across our retail business that AI could solve – but where do we start? What’s your advice for taking those first steps?
Tom: At Footasylum, we started small. We were interested in personalization and were ready to learn about it. But once we started building the customer piece, we ended up with a kind of flywheel analogy, with the customer on one side and demand on the other. If you’re creating demand from the customer side, the flywheel speeds up and you need to meet the demand, if you introduce AI there too, it’ll speed up even more. We also looked at some supply stuff too, helping to optimize inventory through the last Christmas trading period, which was valuable to the overall ecosystem that led to a great period for us. Lots of people will say “we don’t know where to start,” but you can quickly break it down. At Peak, tangible results is the bit we want to see, and that’s what the customer wants to see too. It’s about starting on the journey to get there – and pace is key.
I work in insurance, where price comparison sites are king. There are some retail equivalents popping up but haven’t truly grained traction yet – do you think these will grow?
Tom: This is a question around delivery, where the battleground around websites is essentially in the checkout. Delivery is getting cheaper, you’ve probably done it as a consumer – I’ve done it myself; you hover over the pay securely button when the delivery isn’t free, depending on what it is. When you’ve been at the checkout of a website, the reason “I couldn’t get something delivered where I wanted, when I wanted,” needs to disappear for consumers as an excuse used by businesses, because it’s just not good enough.
This is where we have found deploying AI for people can help optimize where their inventory is, in order to help the whole fulfillment process, whether it’s free or otherwise as long as it’s available. This deployment of AI is not going away, it will only get more advanced and be more of a help to businesses. It does get more complex if you have multiple distribution centers (DCs), if you’re fulfilling from stores or if you’re dealing internationally. But, we were talking to a retailer this morning that has multiple DCs around the world, and we’re trying to optimize the fulfillment process for them at this very moment. It’s something we are engaging in a lot of conversations about.
Can you see a role for AI in marketing town centers and the diverse offers that exist there?
Tom: For multi-channel retailers specifically, it’s almost like the traditional P&L method is not fit for purpose in terms of keeping stores and websites separate, which still exists in most businesses. I think if you were starting from scratch today, you would not have that. In terms of delivering regional marketing, I have been talking about the idea of trading in zones, where you have channelless zones that allow you to focus on areas of particular interest. Maybe you have specific stores, products or brands that exist within your network and have an appetite towards them.
Something we’ve done at Peak is we ingest multiple data sources into one place, which allows a visibility for localized digital marketing campaigns (although the actions can be physical), which comes back to the Peak AI System where all your data is interconnected, allowing for better visibility, agility, decision making and the actionable outcomes that come from it. I would say, on the town center example, whilst it is a micro-topic, we should be moving towards this more channelless vision and aggregation of data where we run the machine learning models across to help you make those marketing moves appropriately.
Merchandising is a great fit for AI but we are struggling to get it going. Do you have any thoughts on this?
Tom: What we’re seeing in merchandising is that, culturally, it can be a challenge to embed an AI system because merchandisers are probably already hamstrung by poor systems (generally speaking) and siloed data and so on. Whereas, it’s actually one of the most significantly valuable areas of opportunity, certainly in the work we’ve been able to do with people. There are opportunities around allocation of stock, markdown optimization, rebuying, buying, range planning; all disciplines that involve merchandising both multi-channel and pure play retailers. There is so much opportunity. Most of this is the common sense that the merchandisers already want to implement, but they don’t have the technical systems to implement these changes and outcomes. We currently have exciting things happening within merchandising for those traders and planners. It is super, super valuable and probably my favorite area – even though I’ve done a lot with Customer AI, Demand AI is just as interesting and really cool.
How robust have Peak’s AI models been during COVID-19? Did businesses working with Peak deploy their AI model and expect it to be totally hands off and fine without intervention?
Manjit: I think the AI has become more critical, and some of our retail customers have been asking, “what is happening? Can you give us the early signals?” This has been by day and by week, whether that’s in the UK or international markets. What we’ve found is, yes, we realize AI models are built on historical data, but because they are live and working today, from a trading perspective it gives people a view of what happened yesterday and what’s going to happen in the next couple of days. Whether that’s on merchandise, rebuy products or whether it’s just on general trading decisions.
Tom: To build on that, it gives a vital insight into the future, built on data that already exists. But because it can augment that data and help it look into the future for certain business outputs, it’s so crucial. With Customer AI, aspects like customer facing advertising, segmentation within databases, how you maintain customer retention and new customer acquisition, it’s about how you can tweak the models to optimize for certain outputs. Take Ad Optimization for one; companies like Facebook and Google do not optimize ads for profit, whereas we do help customers optimize for profit, so there’s been a pivot from “we just want more exposure” to “how much money can we gain from each pound spent?” Helping to optimise profit has been a key focus in the current market.
We hope you have enjoyed our AI in retail blog and most of your questions have been answered! If you want to learn more about the relationship between retailers and AI, check out a few other pieces we have written below 👇