How do retailers make sure artificial intelligence delivers real worth? By being harder on themselves and their technology suppliers argues Valter Andersson, technical solutions manager at Nosto
Talk of artificial intelligence (AI) and its benefits are everywhere, but it’s important not to be wowed by all the AI driven marketing content - and to be laser-focused on technology that is going to deliver real practical benefits.
Retail brands who are looking to bring in AI must start by questioning themselves hard about why and where they want to introduce it. The use cases need to be firmly tied into concrete business goals, such as helping to generate or protect revenue and profit.
Let’s be clear. What most people mean when they are talking about the practical applications of AI today actually involve machine learning. This is a subset of AI in which algorithms are trained to learn and get better at performing specific tasks based on patterns of data they are exposed to.
In a retail context machine learning extends from areas such as intelligent and automated warehouses using robots, to customer-facing technology such as chatbots and natural language processing, to predicting the most relevant products and content to put in front of ecommerce visitors to personalise their experience and maximise sales or margins.
Ideally you want to start by identifying applications where AI can have the biggest impact, with minimum effort and disruption.
Begin your AI journey by applying tools with a proven ROI and that influence a large share of your revenue. Then, move into use cases that are more marginal. For example if you want to use AI to improve or enhance the online user experience, focus first on the home page or checkout page, where the impact of any changes are potentially greatest – before moving on to roll out improvements on less prominent parts of the site.
In fact it’s important to consider whether the use cases you are considering actually need to use AI or machine learning at all. For example in an ecommerce context, you might still make more money doing something as simple as displaying a 10% discount coupon on a simple trigger – such as every time a shopper visits a product page X times – rather than using machine learning to implement predictive targeting based on customer-lifetime value, for example.
This is why it is essential to make sure your AI supplier has a deep experience and understanding of the retail and ecommerce business.
For example, we had a retailer sign up to use Nosto because it wanted to increase its low online conversion rate. After implementing machine learning aimed at showing visitors relevant products in real-time based on their online behaviour, the customer saw a good uplift but nowhere near what they wanted.
Our investigations revealed that that the customer was actually driving extremely unengaged traffic to the site – people that were just not interested in the overall product offering. So machine learning couldn’t deliver because no matter how relevant the items being shown, the audience didn’t want them.
In layman's terms, they were driving dog owners to a store selling cat supplies - but still showing them a red, discounted, food bowl because the visitors were into red items and had an affinity for discounted products.
After some adjustments to the retailer’s audience targeting and tailoring the machine learning algorithm to also match the exact user base, the results improved dramatically.
This underlines how machine learning AI needs to be guided by a ‘human hand’ that can apply the technology in the most effective way.
Your AI supplier should map out pain points and be able to identify what actions should be taken to address specific problems – whether that is raising average order values or increasing customer retention. Ensure that machine learning is trained on the right data to deliver.
The technology provider should work with you to set out the correct business goals and be flexible enough to help you use the tool in the best possible way to achieve those targets, be it increasing conversions, customer lifetime value or something completely different.
Data is all important for machine learning. The first and most important step is to focus on the quality of the data: collect all available data points, validate them and constantly keep this data up to date.
Sadly, however, for many retail use-cases, nowhere near the full potential of the available data is used. For example one approach to retail personalisation is to base it on transactional data (related to previous products a customer has purchased). But our research suggests that typically this is less than 2% of the data generated by online shoppers. So retailers are ignoring a huge amount of valuable signals captured in the remaining 98% - typically behavioural data - which could be use in machine learning algorithms.
Make sure your AI supplier is able to ingest and use data that comes from a variety of sources to allow your use of artificial intelligence to evolve and grow. Typically in retail, data is locked away in silos – your CRM, email tools, Facebook marketing and search as well as separate data from the web business, instore and warehousing. The result is that you can have really valuable data shut away in separate tools that never gets utilised to its fullest extent - which is a big opportunity cost.
At the same time, it’s important to remember that there may be occasions where you have to make intelligent decisions without a huge pool of data points.
Everyone talks about deep learning – neural networks which attempt to replicate how neurons in the brain work, processing significant amounts of data layered on top of each other.
What is often ignored, however, is that many applications - especially in ecommerce - would do better to make use of shallow learning based on fewer data layers, rather than overcomplicating things.
For example, a product might be in stock for a limited time, or a customer engagement onsite might be restricted to just a few clicks - but as an online retailer you still want to use whatever data is available to make intelligent decisions and deliver a relevant online experience based on those limited signals.
Ensure your supplier is clear about how long it will take before the machine learning technology they are delivering will take to deliver practical benefits.
AI can take time before results start to flow, so you want to have realistic expectations.
The key is to be highly rigorous about the way you select and implement AI. Driving significant early wins can fuel even greater adoption over time.