General AI News Retail & Commerce
Aug 17, 2018 ● Nikki Baird
Retail Has Three Big AI Dilemmas

In retail AI use-cases are centering on four main areas...

If you’ve been paying attention to the application of artificial intelligence in retail, you may feel like the buzz around the topic has gone from zero to “arrived” in less than a year. In retail time, even at the speed of the modern consumer, that is incredibly fast.

Some of the hype has come from activity around specific use-cases for the application of AI in retail. While companies like Baidu profess over 100 AI capabilities, in retail it appears that use-cases are centering on four main areas:

Predictive analytics / forecasting – This is forecasting with an emphasis on either products, or customers. For products, retailers appear to be focusing on three main areas of opportunity. First, they are looking at understanding product attributes in a new, AI-driven light. By looking beyond the obvious attribute connections between products, retailers are looking to machine learning to identify and make connections between products that get lost in the noise. They are then connecting those attributes to drivers of demand, to make finer-grained predictions of how well products will sell and why. And finally, retailers are looking to incorporate non-traditional demand signals to get a better picture of demand – seeing if there are connections to be made about consumer behavior related to products that can be exploited in the future, using new kinds of data. For example, predicting that a restaurant will sell 25% more salads if the lunch-time temperature is above 80 degrees F. Or, conversely, that lettuce contamination in the headlines creates a 10% decline in salad sales.

Predictive analytics is also being applied to customer behavior. Matching product to customer behavior can be used in the product sense above, but it can also be use in a customer sense to predict the next product a specific customer would be interested in buying. It can also be used to predict when, in which channel, and at which price (or with which offer) a customer would be most likely to buy, and which product would most have their attention. This has made its way into retail through personalization solutions, mostly targeted at the digital portion of the online journey.

Voice / Natural Language Processing In – While the retail industry tends to group natural language processing (NLP) together into in AND out, in reality there are applications that focus only on inputs, and applications that more heavily focus on outputs, which are much more difficult, and covered next. On the input side, the applications focus on speech-to-text, and then text recognition, which can then be used to analyze for sentiment or emotion. Examples include call center chats or phone calls that detect when a customer might be getting angry, or traditional social media analysis that is smart enough to self-learn – so, instead of a person having to go through and note when there are exceptions to language that is traditionally considered negative (“This vacuum sucks” is sometimes not a bad thing), the AI will be able to detect and categorize exceptions on its own over time.

Voice / NLP Out – The output side is much harder, because it requires the AI to approximate human behavior enough to sound “natural”. Chatbots are on the learning curve, as are automated copy writers. Chatbots are a little easier to pull off because you can rely on a smaller subset of information to seed the chatbot, and they tend to be focused on specific objectives, like problem solving or sales. Copy is a lot harder because it tends to rely on a broader range of inputs and human expectations might include more difficult language concepts like metaphors or poetic license. But retailers are looking to these capabilities to either offset human communication and customer service costs, as in a call center, or to be able to generate a lot more unique copy about products a lot faster – or both.

Image Analysis – The last area of AI opportunity in retail focused on image analysis. This could either be static images or video, as in for facial recognition or footpath tracking. Image analysis is especially important to the fashion industry, to help distinguish a dress from a tunic, for example, or a plaid from a floral print – which helps make attribute connections or assessments that will ultimately assist in forecasting.

Facial recognition, helped along by general acceptance in China and adoption of Face ID-driven Apple iPhone X, is another area for image analysis use-cases, whether that is in “paying with your face” or in more sophisticated use-cases like skin care analysis or color matching.

With Opportunity Comes Challenges

I’ve written about AI challenges in retail before, but this time around, given the opportunities that retailers seem to be pursuing, the challenges are a bit more specific. For all the talk about AI, the level of investment remains low – according to Constellation Research, while 70% of executives surveyed report that they are making investments in AI, over half of all respondents also reported that they’re planning on spending $1 million or less in 2018. Most of that effort is either on bringing data together into data lakes, or on pilots (note that CR’s research covered all industries, not just retail, but considering that retail is usually lagging in new technology investments, the results are probably highly directional for the state of retail as well).

Pilots are good because they help companies discover the “unknown unknowns” – and hopefully plan for those before beginning official rollouts and spending big money. But there are some known unknowns too – and these questions continue to be sticking points for developing an AI strategy. Until retailers have solid answers to these questions, AI will continue to be either a point-solution with AI delivered as a service, or experiments with no concrete plans beyond a trial. Here are the big three.

Platform vs. Point Solution

Every big enterprise software company worth their salt is developing or has developed an AI platform as part of their solution offering. IBM has Watson, SAP has Leonardo, Salesforce has Einstein, and of course, Google has Deep Mind and Amazon and Microsoft both have solutions as well. Retailers who want AI but also want control over the user experience and workflow associated with how it’s used would look to platforms like these.

But getting them to be embedded in their processes requires a lot more work than just buying something off the shelf and plugging it in.  Sometimes the AI solutions offered by these platforms have preconfigured parameters – for example, that a visual engine could distinguish between happy faces and angry ones – and sometimes they need to be trained for the specific process or workflow, like training a visual engine to recognize the difference between a tunic and a dress.

And even if they come trained, they typically offer an answer against a set tranche of data: you send me a bunch of pictures, and I’ll tag them for you. What you do after that is up to you – and can require straight-up development effort, if not at least some heavy-duty workflow configuration. And that requires skills – AI developers are in short supply, and retail often finds it tough to attract IT talent, especially when the platform companies themselves and other industries (like finance) pay more.

On the other hand, point solutions embed the AI-driven answer into a software application that manages something specific. Personalization is a great example – the AI recognizes a set of consumer behaviors, and makes product recommendations based on those behaviors, which are plugged into the user experience as they shop online.

The problem with point solutions is that they’re very good at what they do, and not so good at doing anything else. A visual engine might be useful far beyond facial recognition or product recognition – once it’s trained. Online personalization is only good for online personalization. You could potentially adapt these solutions to move beyond their original intent, but it again takes development effort, for which you are either dependent on the software provider to work into their roadmap, or you’re on your own.

Some retailers are hedging their bets by investing in both – gaining a familiarity with how AI works by engaging at a component level from a platform, and working with startups (the stereotypical 3 PhDs, an AWS account, and a million dollars in seed funding). The challenge with startups is, one, they get acquired, and for AI startups in particular it is often more a case of acquiring brainpower than it is about acquiring capabilities – which means your solution could get sidelined as that brainpower is deployed against other priorities.

Worse, though, the startup could fizzle out. The idea didn’t work as intended, for example. In which case the retailer is left with either turning it off, or ending up back in square one – looking for the capability via a platform, but having to do all the work to get the capability embedded in workflow.

I don’t envy retailers these choices – neither are really very good ones. Definitely retailers should get their feet wet in order to understand what the fuss is about and what the technology is truly capable of. But for a technology that has as much hype as AI does, this is a very immature state of things. People will be the constraining factor more than technology, as every company, tech or otherwise, scours the planet for data scientists and AI programmers.

Black Box vs. Glass Box

Black box solutions take in data and spit out an answer, without giving much context into how the decision was made or how it was influenced along the way. Most first generation AI was designed that way, in part because developers who were creating these advanced algorithms were trying to protect the average end user from confusion and distraction.

But black box AI leaves a lot of value on the table. The whole idea is that the “machine” is learning something that people would not naturally see, or is making connections by wading through a sea of data at a level of granularity that most humans don’t have the time or inclination to examine.

Glass box, on the other hand, is about exposing the connections that are made – in effect, having the machine teach the human what it learned as it starts making recommendations. The learning could come simply from identifying a problem or an opportunity (depending on how you look at it), or it could come from the calculations that went in to making a recommendation – why was something prioritized more heavily in the answer vs. something else.

Black box solutions are a bit easier to deploy. You send data, it spits back an answer, and maybe even makes the connections to execute the most likely or highest-value answer. Glass box requires exposing the sausage factory a bit more, and that means spending more time educating employees on what kinds of answers it may give, and what kind of reasoning goes into those answers. That means training. Some companies are starting to step into that fray, reasoning that the demand for data skills isn’t going to get any smaller in the future. Marks & Spencer is the most high profile retailer making recent investments here, with a commitment to train 1,000 employees on data skills through a partnership with Decoded.

It’s a great idea in theory. Will it work? Will Marks & Spencer invest in these employees only to lose them in the talent war around AI – even if all the employees can do is interpret the results of AI-driven optimization? That remains to be seen.

The Data

The last challenge comes back to the data. Retailers have a lot of data, but just because they do doesn’t mean that it’s all useful. It could be stored at too high of a level of aggregation to be useful. It could be missing key components that prevent retailers from matching their data to other important attributes – you can’t get an understanding of how the weather impacts buying if you don’t know where the shopper physically was when they made the purchase. Retroactively applying location data won’t help, if you never captured the real-time location in the first place.

But a bigger hangup for data is its cleanliness. Retailers have over-estimated the cleanliness of their customer data for years. It hampered their ability to deploy promotion optimization, because their data about customers or their data about the promotions – or both – just had too many holes, errors, or inconsistencies to be useful.

Retailers also need to be careful about whether they actually have enough of the right kind of data, too. There are some awful examples of selection bias in the data used to train AIs that have led to some disastrous results. If an AI trained to recognize faces is only shown Caucasian faces, when your AI encounters the world in all of its diversity, the results are downright embarrassing. You have to make sure that your data is diverse enough to bring valuable conclusions – especially as AI developers attempt to improve the efficiency of AI learning so that learning can happen on a much smaller set of data than is needed today. But you also have to make sure that your AI is learning what you wanted it to learn in the first place – back to black box vs. glass box. No matter the box, the old adage applies: garbage in, garbage out.

The Bottom Line

With all the buzz out there, it can feel like AI is “mature”. When it comes to enterprise deployments, it’s not. Companies are testing, experimenting, and piloting. In some cases, they are running highly focused deployments. Retail tends to lag in adopting new technologies, though I would argue that personalization has been a high enough priority long enough that the industry probably is not behind there relative to other industries. But outside of that particular use-case, there is a long way to go – not because retailers aren’t willing, or the technology isn’t ready, but simply because the delivery mechanisms for how AI becomes a continuous part of the retail enterprise has simply not been defined yet, let alone implemented.

AI is coming, and it will have an impact – but it would be a mistake to assume it has already arrived.


This article originally appeared in Forbes

Article by:

Nikki Baird