The Dos and Don’ts of Deploying AI for Retailers

Artificial intelligence (AI) and related fields — data science, machine learning, predictive analytics — have become mainstream across industries dealing with big data, including retail. One of my favorite ways to explain this influx is a quote (and great article) from Harvard Business Review: “[Data scientists’] sudden appearance on the business scene reflects the fact that companies are now wrestling with information that comes in varieties and volumes never encountered before.”

However, with a lot of talk about big data and data science also comes a lot of “big talk” about AI. There’s an increasing amount of overselling out there of what AI can actually do (I’ll get to what I mean by that in a second), and an even larger increase in spending. Retailers are predicted to invest $7.3 billion in AI by 2022 compared to $2 billion in 2018.

Optimism about AI capabilities continue to this day, but a limitation with AI is that humans are still very much needed to make these capabilities successful. AI, at its core, is complex math, and it’s up to the right people and the right data to solve real problems retailers face in order to grow their businesses and deliver better customer experiences.

And while all AI isn’t hype — I’ve seen many success stories from retailers leveraging models to power personalization based on predictions of future shopper behavior, propensity to convert, likelihood to churn, potential future spend, etc. — there are some key things retailers need to do (and avoid) to keep their investments from falling short:

Do: Make Sure Data Scientists Collaborate With Stakeholders.

Successful data science and AI initiatives cannot be achieved by technology alone or even by data scientists working with technology. In today’s tech landscape, data scientists bemoan that 87 percent of data science projects never make it into production. The primary reason for this is a disconnect between the technology (machine learning, algorithms, and computing infrastructure) and our inability to connect the data and models with business needs.

Both data scientists and business stakeholders need to coordinate with IT to ensure the right data is available to solve business problems with machine learning, and that the data is available at the right time. I often refer to this combination of skills/people as the “three-legged stool of successful analytics."

Don’t: Dive Right Into Large Projects. Start Small.

How do organizations get started and avoid their projects falling into the 87 percent that never launch? First, begin with collaboration in mind at every stage of the process, from defining the problem to solve through final model deployment. Defining what to predict with AI and how to measure “good” from the perspective of the business requires inputs from all three legs of the stool — business stakeholders, data science, and IT.

Next, start small with focused, tactical projects rather than planning for a large “game changer.” Most retail organizations, when beginning to apply AI, fail to understand the vast number of hurdles involved with any solution, including but not limited to obstacles in data collection, data preparation, and model deployment. These often require several iterations with input from all three legs of our stool before getting every part of the solution workable and operational.

Don’t: Be Afraid of Failure and Pivoting En Route.

Don’t be afraid of failure. We often find that our expectations at the beginning of a project are inflated due to a misunderstanding of data quality (we often think our data is better than it actually is) and infrastructure limitations. Accept the misunderstandings or failures and be prepared to alter your thinking, approach, and resources along the way.

Do: Consider Your Options to Build vs. Buy.

Most retailers don’t have an entire team of data scientists at their disposal, and significant amounts of training and experience are needed for any data analyst to properly build models and interpret findings that would lead to effective insights and action.

If you don’t have these capabilities in-house, there are solutions out there that can do the heavy lifting and help solve the predictive marketing challenges retailers are experiencing. Just be sure to do your research to confirm these technologies can actually do what they promise and that the company will collaborate to align their solutions with your direct business needs.

Dean Abbott is co-founder and chief data scientist at SmarterHQ, a behavioral marketing firm and personalization platform.