AI-driven agents in retail enhance decision-making by automating workflows, analyzing data, and providing dynamic recommendations while involving humans for complex judgments.

Business Situation

In the retail industry, many tasks that were once sequential or conditional were programmed during the Information Age, automating routine processes. However, major decisions—such as supplier selection, inventory management, and pricing strategies—are still largely handled by humans.

But what if all of this could be reimagined in the Intelligent Age? What if AI-driven agents could augment human decision-making, ensuring efficiency while keeping a human in the loop for critical judgments?

Scenarios

  • Scenario 1 - General Merchandiser
    If the cost of a Purchase Order is less than $10,000 from a Tier 1 Supplier, approve it, set the status of the PO to ‘Approved,’ and notify both the Buyer and Supplier.

  • Scenario 2 - High-End Retailer
    Here, quality is the top priority. When awarding a Purchase Order, analyze supplier history, material sourcing, and craftsmanship quality to recommend the best choice.

  • Scenario 3 - Fast Fashion Retailer
    In this case, speed and cost are imperative. Evaluate suppliers based on turnaround time, bulk pricing, and logistical efficiency, ensuring an optimal balance of speed and affordability.

Scenario 1 can easily be implemented in a rules engine or backend code. However, for Scenarios 2 and 3, these decisions were traditionally left to specialists within the Buying Teams.

Now, Agentic Workflows can dynamically evaluate multiple conditions, weigh trade-offs based on broader context and historical data, and provide recommendations while still involving humans for complex decisions.

Technology Evolution

Traditionally, products such as ILOG CPLEX and Groovy were used to streamline business operations and implement deterministic decisions.

However, the new paradigm shifts toward Agentic Workflows, where intelligent systems can access diverse datasets, analyze multiple factors simultaneously, and generate dynamic recommendations—non-deterministic decisions.

While these workflows are currently mostly within the tech realm with tools like Cursor, Claude, and MCP Servers, they will soon proliferate into the business ecosystem as vertical agents, providing a wrapper on concepts like Mixture of Experts and Human-in-the-Loop.

Thank you to Nate Herkelman for explaining deterministic and non-deterministic workflows and Minki Jung for explaining Flow Engineering.

Workflows and Decisions