Not all AI is created equal

Not all AI is created equal

Max Smolaks

August 8, 2019

7 Min Read

Three ways mature retailers differ in their AI and data strategy

by Rashed Haq 8 August 2019

A recent Gartner survey revealed that six in ten organizations (59 per cent) have already deployed artificial intelligence (AI) or machine learning (ML) systems. Within this group, companies have an average of four projects in place and investments are expected to double within the next year. The implications are significant, as the survey speaks to the growing popularity of this technology across industries, its widespread application within the business and even its acceptance among consumers.

But this researchfails to account for one very important point: value. For so manyorganizations, the mere presence of AI/ML technology does not equate tomeaningful business impact. Likewise, having more applications does not guaranteebetter results. The value derived from AI depends on what tools are used, how theyare applied and what is done with the insights. Put simply, not all AI isequal.

For manyretailers, their initial foray into AI and ML involved strengthening data capabilitiesas a way to improve key performance indicators like purchase frequency, averageorder value and customer satisfaction. However, this piecemeal approach barely scratchesthe surface of the true value opportunity. To realize the full potential ofthis technology, retailers must develop a holistic AI and data strategy thatnot only enables efficiency gains within each function, but also supportsbroader organizational goals such as frictionless shopping, operationalexcellence and experiential differentiation that spans across functions.

Here we look at the three strategic differentiators of mature retail organizations and the thinking that sets them apart in the AI landscape.

1: Get to know the customer – not the segment

For years,big data and AI has been an important enabler of personalization andcustomization within the retail industry. Or has it?

Whileretailers have used this technology to better understand customers and theirbehaviors, most still fall short of offering an individual, personalizedcustomer experience. That’s because traditional AI models rely on customersegmentation to generate insights about a specific group, as opposed to recognizingthe customer as an individual.

However,with deep learning, a subset of ML, retailers can create a much more preciseunderstanding of each customer. This capability goes beyond simple segmentationand actually formulates a detailed, individual profile with all known customerinformation in a single vector—sometimes referred to as the “customer genome.” Inpractice, this means that the retailer can reach each shopper on a personallevel, as opposed to targeting various segments.   

For example, many retailers’ eCommerce strategy includes AI-enabled product recommendations. Typically, this effort is based on the customer segment and has a decent conversion uplift as compared to doing nothing. However, incorporating deep learning capabilities, as well as integrating multiple AI models into a single activation, can further enhance results. For example, product recommendations based on the customer genome can improve conversion by two percent. Incorporating additional models to determine what channel the shopper is most responsive to and what time of day they are most likely to engage can further increase conversion six-fold, from two to twelve percent.  

2: Build an AI platform – not just a product

At present, AI is being leveraged in retail organizations in a variety of ways, helping them do anything from optimizing pricing to automating payment. Individually, these models help drive growth or enable efficiency. However, the true value of AI comes from compounded efficiencies when these models are linked as part of a platform and deployed at scale.

But this begs the question: How do organizations know whichefforts to link? How can they determine which combination of activity willyield the highest value? The main benefit of an AI platform is speed.Organizing efforts in this way enables up to five times faster, more efficient executionand analysis, which greatly improves the ROI of building and productionizingmodels. Even more importantly, the speed of the platform allows the organizationto perform more experiments, select the best models and get them to productionfaster.

For example, many retailers use AI to inform their e-commerce fulfillment process and determine the most cost-effective way to ship an order, whether that is from the factory, a local store or a regional warehouse. However, by leveraging an AI platform, retailers can experiment with multiple models to get a deeper understanding of total cost to serve. In this case, the retailer can combine the shipping model and global inventory management model to assess both delivery costs, as well as product availability. This process can identify a shipping option that may be slightly more expensive initially, but helps a specific region or location avoid upcoming markdowns or reduce excess stock. In this way, an AI platform helps maximize total profitability as opposed to minimizing the cost of one performance metric.

3: Focus on the reward – while mitigating risk

While many customers have grown accustomed to everyday AI applications like predictive text and Netflix recommendations, some may remain wary of the technology—and the organizations wielding it. To maintain integrity with the customer and other stakeholders, businesses must conduct themselves safely, fairly and transparently.

Matureretailers build model governance into each phase of the AI project lifecycle toensure that every application is researched, developed and deployed in a safe,ethical way. An advanced AI strategy includes multi-checks at each step of themodeling lifecycle to help eliminate bias and ensure the validity of outputs.This may include fairness testing, which ensures underprivileged and protectedgroups are treated fairly, or sensitivity and boundary condition analysis,which studies how each input factor influences outcomes.

Further,businesses must understand the power of this technology and maintainaccountability for where and how the insights are used. While the technologymay process data and develop recommendations, it is still the responsibility ofhumans to explain the process and interpret the results.  

Going back to the Gartner survey, we learn that nearly all organizations expect to accelerate their AI and ML efforts in the coming years, with the most advanced organizations anticipating the deployment of as many as 35 projects by 2022. An exciting prospect, to be sure. But without a concerted effort to integrate those programs into a single strategy, many organizations may fail to generate the full benefits of their investments. For many retailers, there will be truth in the age-old adage: quality over quantity.

Rashed Haq is the global head of Artificial Intelligence and Data Engineering at Publicis Sapient. He is known for helping global companies transform their businesses and create competitive advantage by defining AI strategies, and developing and deploying AI solutions at scale.

Rashed is a frequent speaker at conferences and is a published author of numerous articles, papers and an upcoming book on enterprise AI transformation. Prior to joining Publicis Sapient Rashed conducted research in theoretical physics at the Los Alamos National Lab and the Institute for Theoretical Science.

Get the newsletter
From automation advancements to policy announcements, stay ahead of the curve with the bi-weekly AI Business newsletter.