• Written by Sonali Subhadarsini, Product Manager

A category manager stares at a spreadsheet, trying to make sense of the promotional distortion in the historical sales data, wondering how to account for the shelf life write-offs and the lack of visibility into the distributors' inventory levels, all while trying to optimize the production and distribution of fast-moving consumer goods.

FMCG complexities

The fast-moving consumer goods industry presents unique challenges when implementing agentic AI workflows, primarily due to the high volume and short shelf life of products, as well as the promotional-driven demand that can significantly impact sales forecasts. The promotional distortion in historical sales data makes it difficult to identify baseline sales levels, while the short shelf life of products requires rapid action to avoid write-offs. Additionally, the lack of visibility into the distributors' inventory levels can lead to stockouts or overstocking, further complicating the supply chain.

The shift supervisor, responsible for managing the production line, must also contend with the complexities of FMCG, including the need to quickly respond to changes in demand and the requirement to minimize waste and maximize efficiency. This can involve adjusting production schedules, managing inventory levels, and coordinating with logistics and distribution teams to ensure that products are delivered to stores in a timely manner.

Agentic AI workflows

An agentic AI workflow is distinct from a chatbot or a simple question-answering system, as it is capable of taking multi-step actions to achieve a specific goal. This requires the use of various tools and systems, as well as the ability to maintain state and adapt to changing circumstances. In the context of FMCG, an agentic AI workflow might involve analyzing sales data, adjusting production schedules, and coordinating with logistics teams to ensure that products are delivered to stores in a timely manner.

The key characteristic of an agentic AI workflow is its ability to use tools and systems to achieve a specific goal, rather than simply providing information or answering questions. This requires a deep understanding of the underlying systems and processes, as well as the ability to integrate with various data sources and applications. For example, an agentic AI workflow might use data analytics tools to analyze sales trends and identify opportunities for optimization.

Human approval and oversight

While agentic AI workflows have the potential to automate many tasks and processes, they still require human approval and oversight to ensure that they are operating effectively and efficiently. Removing the human approval step prematurely can lead to errors and inefficiencies, as well as a lack of accountability and transparency. The category manager, for example, must review and approve the production schedules and inventory levels recommended by the agentic AI workflow to ensure that they align with business objectives and constraints.

The role of human approval and oversight is critical in FMCG, where the consequences of errors or inefficiencies can be significant. By maintaining a human approval step, organizations can ensure that their agentic AI workflows are operating effectively and efficiently, while also providing a level of accountability and transparency. This can involve establishing clear guidelines and protocols for the use of agentic AI workflows, as well as providing training and support for employees who will be working with these systems.

Failure modes and logging

An agentic AI workflow can fail in different ways than a traditional model or chatbot, primarily due to its ability to take multi-step actions and interact with various systems and tools. When an agentic AI workflow fails, it is essential to log the failure and analyze the underlying causes to prevent similar failures in the future. This can involve logging errors and exceptions, as well as tracking key performance indicators and metrics to monitor the effectiveness and efficiency of the workflow.

The shift supervisor, for example, might log errors and exceptions that occur during the production process, such as equipment failures or inventory discrepancies. By analyzing these logs, the organization can identify areas for improvement and optimize the agentic AI workflow to minimize errors and inefficiencies. This can involve working with the support team to resolve issues and improve the overall performance of the workflow.

Limitations and constraints

While agentic AI workflows have the potential to transform the FMCG industry, they are not a panacea for all challenges and complexities. There are certain situations where agentic AI workflows may not be the best choice, such as in situations where the rules and processes are highly complex or nuanced, or where the consequences of errors or inefficiencies are extremely high. In these situations, alternative approaches, such as traditional modeling or manual processes, may be more effective.

Additionally, agentic AI workflows require significant investment in data quality, system integration, and employee training, which can be a barrier to adoption for some organizations. By understanding the limitations and constraints of agentic AI workflows, organizations can make informed decisions about where and how to deploy these systems, and ensure that they are using the most effective and efficient approaches to achieve their business objectives.