AI for Retail Execution. A Best Practice Guide for CPG's. Part One: The Data-Centric AI Approach

Welcome to the first part of our comprehensive 5-part best practice guide, where we will explore the critical role of adopting a Data-Centric approach when leveraging AI and ML (Machine Learning) for Retail Execution. 

Data Quality Banner_1200x800

AI is revolutionizing the landscape of Retail Execution, ushering in a new era of intelligent, hyper-automated, personalized, and highly effective operations. When we commence the AI transformation process with our clients we do not start with chatbots, advanced analytics, playbooks, or performance engines but with data quality 

By focusing on data quality as the foundation of AI transformation, we harness its power to deliver complete, real-time, and 100% accurate insights for all retail execution requirements. The potential of AI/ML to tackle data issues is unlimited from process automation, data augmentation, and synthesis of missed data to sophisticated data analytics. 

Industry leaders aiming to harness their data for advanced AI models are realizing that poor data quality consistently hinders the most valuable AI applications. (McKinsey,2023) Common data quality issues include inconsistency, inaccuracy, incompleteness, and duplication, which can severely impact decision-making processes. (Technologyadvice).  Figure 1 below highlights common data quality issues.   


Figure 1: Data Quality Issues

The Data-Centric AI Approach  

A recent article published by Springer, highlights that in recent decades, the AI community has been dedicated to enhancing Machine Learning (ML) models within AI systems. However, the significance of sourcing and selecting appropriate data cannot be overstated, as it directly impacts the efficacy, performance, and efficiency of ML. Despite this, the importance of data quality and quantity in AI systems is often disregarded in both research and practice. Data-centric AI advocates for the strategic development and implementation of methodologies, tools, and best practices to meticulously craft datasets and enhance data quality and quantity, ultimately boosting the effectiveness of AI systems. It's not about just gathering more data; it's about having the correct data.  

Data-centric AI stands out from model-centric AI through its emphasis on data, the significance of domain expertise, and the comprehension of data quality. 

Emphasis on Data: Data-centric AI focuses on enhancing the quality and quantity of data while keeping the ML model constant. By optimizing the data, performance improvements can be achieved with a fixed model. 

Data Processing and Domain Expertise: Domain-specific data processing plays a crucial role in data-centric AI. This involves utilizing methods and semi-automated tools to expedite the creation of successful AI systems. 

View on Data Quality: Data-centric AI drives enhancements in performance by utilizing more relevant data. Consequently, changes in ML model performance metrics reflect the efficacy of data adjustments, offering a fresh perspective on data quality that aligns with machine learning metrics. (Springer,2024) 

Traditional Coding Tools vs AI and ML  

The primary difference between machine learning and traditional programming lies in their dependence on user input versus algorithms that facilitate independent learning. Traditional programming necessitates specific instructions from users to solve problems, with developers needing to establish boundaries, rules, and logic for each process. This approach has long been the hallmark of traditional computing and is commonly utilized for tasks such as calculations and data sorting. 

In contrast, machine learning systems place a greater emphasis on training data rather than predefined rules. This enables algorithms to adapt, make classifications, and predictions based on new information, equipping them to handle a variety of scenarios. (Institute of data, 2023 

Figure 2 below demonstrates the difference between using traditional coding tools in the data analytics process versus AI and ML solutions.  


 Figure 2: Data-centric AI Solutions vs Traditional Tools

How Data Quality Issues hold back CPG Sales from Excellent Retail Execution. 

According to the Promotion Optimization Institute Annual State of the Industry Report, 2024, “61% of CPGs are held back from exceptional retail execution due to data and insights not being fully leveraged”; “70% of RetX sales teams noted that they don’t have the insights required to take appropriate actions at the store level”; “80% of companies state their headquarter support teams don’t have the necessary capabilities to support pricing, trade allocations, and go-to-market strategies” (POI, 2024) 

These insights paved the way for Spring Global to develop MAESTRO, an AI data integration layer bridging data warehouses and applications to facilitate AI integration within the operational software ecosystem.

Through a meticulous process of data extraction, preparation, processing, and loading AI insights back into relevant applications, MAESTRO effectively addresses crucial architecture gaps for AI implementation. Notably, it excels in data orchestration at a granular level and streamlines data preparation for AI, accommodating the intricate product hierarchies, diverse retail channels, and varying data granularities specific to the CPG industry.

Furthermore, MAESTRO harnesses domain-specific expertise to identify and rectify data quality issues, such as inconsistent product names, missing attributes, and discrepancies in sales and inventory data from multiple sources. Please see Figure 3 below for an overview of the ML-powered insights process enhanced with Data Quality.  


Figure 3: ML-powered insights process with enhanced data quality

Learn more about Maestro AI