Artificial intelligence and machine learning (AI/ML) technologies are transforming businesses. In fact, the global annual spending on AI by retailers is estimated to reach $7.3 billion by 2022. However, along with this growth comes hype, uncertainty and questions around what AI/ML can really do for retailers today.
One thing, if anything, is certain: AI/ML in retail is here to stay.
For those of us who have been in retail a long time know this isn’t the first time data science was used to address retail business challenges. In the early 2000s, data science was applied to common problems within retail merchandising and the supply chain with mixed success. Markdowns were optimized, followed by statistical forecast-driven replenishment, then price and transportation optimization.
Despite the application of data science, these new approaches to retail challenges were flawed. First off, adoption of these systems was slow and drove a need for new roles in an organization which were, for many, too expensive. Secondly, the systems were not sophisticated enough to handle all the data nuances (i.e., data sparsity), so confidence levels in the recommendations were low. While these early data science forecasting and optimization systems drove some business improvements, they’ve yet to make a dent on the $1.4 trillion inventory challenge confronting retailers today.
Data Science in Retail Today
As the $1.4 trillion inventory challenge persists, retailers continue to rely on historical analysis, gut instinct, and averages to run their business. But a lot has changed since the early 2000s. AI/ML techniques have advanced to the point where:
- predictions are accurate enough to drive the business
- humans do not need to decide what data sets are relevant or what to model
- relevant context can be considered
- sparse data and anomalies can be dynamically handled
- optimization models can handle uncertainty and multiple objectives
- models continue to improve and learn
To be successful, there is no choice but to embrace change, and use AI/ML to shift to a predictive mindset while running your business.
5 Practical Steps to Using AI Successfully
Embracing change, however, can be difficult. Many retailers believe AI/ML can help improve their business, yet commonly worry about where to start, whether they have the right data, what level of sophistication is needed by the business, and how to drive adoption in the user community. While all of these concerns are valid – there is a right way to go about adopting AI/ML, and, more importantly, ensuring success across your organization.
Here are a five practical steps I’ve seen make AI/ML projects successful:
1. Start With the Data You Have
Start with the data set you have and don’t wait to get a full data strategy implemented. From my experience, retailers will still experience good results and can then continue build from there. You’re not tied into staying with just the initial data sets used. As more data becomes available, our models can dynamically ascertain the statistical relevance of the data set and incorporate as needed. For example, one retailer using Celect Allocation Optimization didn’t have store inventory, so we started them with assumed inventory (using transactions and shipments). From there, we were able to apply AI/ML to the allocation process and still drive significantly improved allocation results.
2. Start Where It's Easy to Measure & Avoid Subjectivity
The value of AI/ML is uncovering opportunities typically not considered. Users often question the validity of recommendations that fail to align with their gut instinct and fail to execute them. Because of this, it is easier to start with a process where the volumes and time requirements force decisions to be automated, much like store fulfillment. For example, replacing rules-based decisions will show results quickly, while augmenting automated processes will make it easier to measure the results before (and after) adding AI/ML to the decision-making process. After you’ve built confidence in the decision-making, it becomes much easier to expand usage to other processes.
3. Augment Existing Systems – Don’t Rip and Replace
Replacing planning and execution systems to implement one with AI/ML embedded is costly and timely. Instead, you can achieve a faster ROI by using a specialized real-time AI/ML system to augment existing systems. Not only is it easier to implement, but real-time AI/ML is important as you are working off the most recent data and can use new or changing constraints.
4. Measure and Compare Before/After
Measure and compare results before and after AI/ML. This builds the confidence in the user base, which will aid in adoption. However, be ready as it is typical for the business to want more AI/ML applied after seeing the results.
5. Embrace AI Continuous Learning
Continuous learning is the essence of AI/ML. Not all results will be perfect the first time through the models. It is best to start with models that have already been trained for retail as the time to value will be much faster.
My advice is simple, don’t boil the ocean. Get started in an area where you can drive business results quickly and successfully, then grow from there.