Improved data quality and outputs will close the loop on AI-driven supply chain optimization, according to Austin White-Gaynor
The last few years have laid the groundwork for US CPG brands to leverage data at a scale and speed that was previously unimaginable. 2024 was the year many supply chains introduced data harmonization, with streamlined and structured data democratizing access to insights across the organization. With the onset of agentic AI, 2025 focused on building contextual layers on top of that data foundation, enriching them with a semantic layer, so systems can grasp business context and surface meaningful recommendations.
The AI-ready data foundation is enabling real progress in automated supply chain optimization. From here, defining guardrails to govern autonomous actions and prioritizing the highest-ROI use cases that justify continued investment will be key for deployment at scale.
The US retail industry is ripe for more integrated solutions to enhance demand forecasting, identify anomalies, automate replenishment, and keep teams informed end-to-end in real-time. The framework is there, and much of the model development is complete. What is left is maturing the development cycles and closing operational gaps to enable a rapidly self-improving feedback loop that will drive real momentum in 2026.
Grade-A data fuel
For decades, retail supply chains have struggled with fragmented data sources, unlabeled records, and data fields that teams have lacked the time or tools to
manage effectively. Automatic data harmonization technology has led to significant improvements; however, with millions of rows of new data generated daily, there are gaps – which AI is poised to address in 2026. Agent-led data enrichment and quality assurance can take datasets that are ’80 percent clean’ and push them closer to true completeness.
The systems can also identify outliers, revive stale fields, and provide insights into pricing, shelf life, and other complex, highly variable data attributes. For example, clustering becomes a breeze when agent-led classification can quickly label 10,000 stores as ‘ski town locations’, with integration directly into analytics and forecasting models. These data sets are then validated for accuracy, enabling the models to continuously improve on clean-up and data management. Data engineering teams are relieved of manual troubleshooting and maintenance, resulting in cleaner, near-100 percent quality input for more accurate and consistent business decisions.
Better results, exponential improvements
With quality foundations improving, supply chains can achieve AI-powered, closed-loop operations that the US retail industry has been chasing. ML demand forecasting models, anomaly detection, and inventory optimization are strengthened by continuously monitored performance at the item-location level. Suppose a single SKU at a single store begins drifting off pattern. In that case, agentic systems can detect it and adjust the forecast without affecting the other 19 SKUs that are performing as expected.
Another major unlock will see qualitative feedback transformed into structured, operational actions. US retailers, for example, are inundated with hundreds of emails each week containing input from store managers about missed deliveries, mislabeled products, moldy shipments, and more. These real-time insights are rarely incorporated into the forecasting engine. In 2026, AI agents will be able to process and record this feedback, triggering direct system edits such as reconciling inventory, adjusting shelf-life, and prompting change orders.
As improvements unfold, demonstrated success will feed back into data models in real-time, reinforcing the corrective actions taken and training agents to recognize similar patterns more quickly, with greater accuracy. Some agents will become long-term fixtures, continuously improving as they are used, while others can be deployed quickly to manage new or unusual use cases, serving as temporary stopgaps, that yield immediate improvements without overburdening data teams.
Automated replenishment and waste-free supply chains
Over the last few years, Dollar General has scaled automatic ordering for its fresh produce program to over 7000 store locations, delivering healthy, non-processed options to food insecure regions. The program has effectively reduced perishable waste, while fresh category sales increased, and offers a framework for greater adoption of automated replenishment solutions.
Looking ahead, there is an opportunity to move toward Distribution Center (DC)-level forecasting derived from a retailer-grounded, single source of truth. When DCs plan inventory using the same item-location signals that drive store-level replenishment, the retail supply chain becomes even more synchronized. Tighter alignment between brands, distributors, and retailers can even reverse the notorious bullwhip effect, where local miscalculations amplify into greater imbalances upstream.
As these foundations mature, agentic systems can extend further upstream and downstream, to automate replenishment decisions, reduce waste, and tighten demand signals across channels. The main challenges facing teams between now and this reality will involve stakeholder alignment around a clean and accurate data truth, executing local pilots that prove measurable ROI, and scaling with the guardrails and governance needed to maintain quality as automation expands.
Austin White-Gaynor is the Senior Director of Data Science at Crisp, where he leads work in AI-driven retail supply chain optimization. With six years of experience focused on fresh-category replenishment, he has helped build forecasting, inventory, and shelf-life models that reduce waste and strengthen day-to-day operations for CPG brands and retailers. Austin holds a PhD in Geosciences from Penn State University.