Can AI Turn Grocery Retailers Into E-Commerce Giants?

According to the 2018 Activate Outlook Report, grocery is the next battleground for e-commerce. Interestingly, the report makes a case that brick-and-mortar stores are an enabler of online growth, beckoning a future less interested in the extinction of physical stores, but rather a harmonization of physical and e-commerce retail experiences. Central to that harmonization is a compelling vision that integrates data, technology and increasingly artificial intelligence (AI). Despite rising success rates of enterprise IT projects, those with AI at their core are highly complex and are arguably less likely to resemble their predecessors. In an era where approximately 14 percent of enterprise IT projects are bound to fail, and 32 percent suffer uncontrolled scope creep, how can traditional enterprise retailers hope to compete against Amazon.com and Wal-Mart?

The unsurprising answer was buried deep in Jeff Bezos’ 2016 letter to shareholders:

“But much of what we do with machine learning happens beneath the surface. Machine learning drives our algorithms for demand forecasting, product search ranking, product and deals recommendations, merchandising placements, fraud detection, translations, and much more. Though less visible, much of the impact of machine learning will be of this type — quietly but meaningfully improving core operations.”

Key to Bezos’ vision for machine learning (the subfield of AI showing the most promise in enterprise problems) is its deceptive modesty. Machine learning is a means to an end, rather than an end in itself. Though Gartner has placed machine learning (as well as deep learning) at the peak of its hype cycle, Amazon has been embedding machine learning into the core of its business for years. In practice, how does this come to life?

While Amazon and Wal-Mart’s organizational models are respectively unique, the harmony they integrate deep into their business models generates a few formative lessons which can be applied to grocery retailers to see more efficient e-commerce adoption. I’ve distilled a few of these “cardinal rules” that will be required for any enterprise applications of AI to succeed.

The System is the Sum of its Parts

Contrary to popular belief, AI is not a panacea for all of the world’s problems. The novelty of the math involved is of much less importance than the robustness of the system within which it's embedded. For example, grocers may input only a subset of promotion and pricing data into a legacy enterprise resource planning (ERP) system, sitting on a legacy database. Any effort to accurately forecast demand in the presence of changing business complexity requires not only complete data, but the ability for the algorithms to automatically scale, their outputs quality controlled, and re-integrated back into equally legacy supply chain systems.

Iterate and Prove Value Early

Grocery retailers are low-margin businesses and need to see investments pay off early. In addition, one should never take for granted that every problem, on first assessment, can be solved using AI. Enabling a culture of quick iteration and experimentation, while using techniques such as cross-validation and various kinds of simulation on historical datasets, allows for a very quick path to understanding what’s possible and the financial value that can be unlocked.

Data Needs to Be Abundant, Clean and Accessible

Whether enabling in-house machine learning talent or partnering with a capable vendor, the common pitfall is reliable data. Large enterprise retailers are typically plagued with siloed and fairly opaque data systems. Any successful application of AI requires the traditional shackles on data to be removed, and for data to be cleaned and deployed to environments more suitable for experimentation.

The Human is Critical to the System

In Amazon’s warehouses, automation has gradually cascaded across more and more functions. Yet new jobs are being created. In fact, human interaction with AI systems generates useful data that can be reincorporated into certain kinds of algorithms through feedback loops. For example, a marketer can provide constraints to a recommendation engine to not only provide oversight, but also empower reinforcement learning algorithms to tune the recommendations over time.

Harmony Between Tools and Talent

Enough has been said about the scarcity of tech talent; yet little has been said about the kinds of organizational models and tools required to make use of it. Traditionally, data scientists have been prized for their math and statistics skills, less so for their ability to write good code. Ultimately, this leads to a culture of spaghetti code being tossed over to production teams to automate. Instead, consider a system whereby data scientists are armed with better tools that empower them to own their code.

So, if grocers and other traditional enterprise retailers follow these cardinal rules, do they have a chance to compete against Amazon and Wal-Mart? Yes. It will be an uphill battle, but the integration of AI is a crucial first step in helping to transform and unify traditional brick-and-mortar and e-commerce operations.

Waleed Ayoub is the chief product officer at Rubikloud, a company changing retail with intelligent decision automation using AI.