Implementing AI-Based Price Optimization: 4 Mistakes to Avoid
Since the dawn of retail as a human occupation, it has turned into a complex industry, with millions of players and networks between a manufacturer and an end buyer. So now, years and years after, retailers are amidst a highly saturated competition, diversified landscape of sales channels and store formats, tangled purchase funnel, and sophisticated approaches to customer experience. If earlier, traditional analytics fully satisfied retailers' needs, today's realities and industry challenges raise the bar high and push retailers to call upon technology for help. Be it boosting holiday sales, tracking customers' in-store behavior, assortment planning, and retail price optimization, advanced technologies like artificial intelligence (AI) and machine learning (ML) cover the industry's needs.
Given the new level of data processing and business analytics AI/ML solutions provide, global spending by retailers on AI services tech investment is estimated to reach $12 billion by 2023, which will spark the tech race among industry players.
ML and AI are cited as the top technologies that will have the most significant impact on the industry in the nearest future. A number of players can attest to the positive effect already. However advanced and well-trained intelligent algorithms can be, retailers should always use the technology cautiously and ensure they feed the correct inputs into the system. Sometimes a mix of external conditions and human oversight can offset the achievements of even the most excellent algorithms.
Below you'll find a few real-life cases we faced while running a project for one of our clients. The challenges once again demonstrated the crucial role of every preparative step to adopting AI-based price optimization solutions. Therefore, beware of these mistakes to ensure you successfully land the solution within your organization and reach the desired results.
Failure and Learning 1
As every AI price optimization project starts with a test run that allows trying out the capabilities of the algorithms and engines, retail teams should thoroughly choose the proper test and control groups.
For test and control groups, the retailer chose stores that had different product mixes and sales patterns in our story. This difference played a crucial role in achieving the pilot's goals: part of the stores were predominantly selling the retailer's private-label products, which suffered a long delay in delivery during the solution pilot. Subsequently, it led to a significant revenue decrease.
Key takeaway: To correctly select test and control store groups, one should carefully study the sales structure for each of the stores, identify key product groups in greatest demand, and analyze procurement of goods. Later on, they will significantly influence the profit and revenue growth.
Failure and Learning 2
The next step of the project workflow implied choosing the right pricing strategy for reaching the desired goals. Predictive algorithms were gradually driving the prices down to increase profit while managing promotions in test and control groups. Yet again, the difference in sales patterns across the stores in test and control groups led to a profit increase and an unexpected revenue decrease, followed by less penetration of essential product categories into the check and a significant outflow of traffic.
Key takeaway: While developing a pricing strategy, always consider sales patterns across different store formats, product elasticity, and customer response to price changes.
Failure and Learning 3
During the pilot project, the retailer chose several pricing strategy vectors, including mass promo across the selected stores. While the advanced systems implemented the strategy, the team received strikingly different results across different shop formats. Overwhelming promo campaigns led to significant increases in items sold in both store formats in some of them. However, at the same time, both store clusters were producing somewhat different results in the number of sold items and revenue generated.
Key takeaway: A significantly different reaction to promo in different stores may indicate the need for re-clustering the product range both in the test and control groups and in the retail chain as a whole.
Failure and Learning 4
Finally, the project results were significantly affected by the product scope entrusted to the system by the retailer's team. As a result, it was simply insufficient for the system to fully unroll and apply all the potential to grow the intended financial indicator.
Key takeaway: To achieve all the set goals for the growth of key indicators, it's vital to consider the number of SKUs and revenue share under management. Insufficient sales volume may limit the system's ability to work on boosting financial performance.
AI and ML are gradually taking over retail, and retailers enthusiastically welcome new technologies within their organizations. However, to reach new financial heights, retail teams entirely rely on advanced systems while forgetting that the final project success is due in no small part to themselves. We've come through the path of test and trial to share the mistakes to avoid while starting on the AI-driven price optimization journey.
Vladimir Kuchkanov is a pricing solutions architect at Competera, pricing software for online and omnichannel retailers.
Related story: How to Boost Revenue and Gross Margin of Lifecycle Products
Vladimir Kuchkanov is a pricing solutions architect at Competera, pricing software for online and omnichannel retailers. He is a Data Scientist, a top-rated domain expert in business analytics, pricing and media management with a successful track record in world-class FMCG companies.