• Written by Azhar Aliyar, Director

A maintenance log on the operations dashboard shows a series of overdue tasks, each one waiting for a specific team to take action, and a shift supervisor checks the log every morning to see what needs to be done. This is a common scenario in many organisations, where a single question about the status of a task can lead to a series of follow-up actions. However, most people get this wrong by thinking that a chatbot can handle these complex workflows, which is simply not the case.

What separates an agent from a chatbot

An agent is separated from a chatbot by its ability to use tools, maintain state, and work towards a goal. In the context of operations, this means that an agentic AI workflow can take multi-step action, rather than just answering a single question. For example, a shift supervisor checks the maintenance log and sees that a task is overdue, so they use the system to assign the task to a team, track the progress, and receive notifications when the task is completed.

This is in contrast to a chatbot, which can only provide a response to a single question, without taking any further action. Most people underestimate the complexity of operations workflows and think that a chatbot can handle them, but this is a mistake.

Where a human approval step belongs

A human approval step is often necessary in operations workflows, especially when it comes to tasks that require a high level of expertise or have significant consequences. However, removing this step is usually premature, as it can lead to errors or unintended consequences. For instance, a category manager compares two suppliers and decides to switch to a new one, but this decision requires approval from a senior manager to ensure that it aligns with the organisation's overall strategy.

In this case, the human approval step is necessary to ensure that the decision is sound and that the organisation is not taking on unnecessary risk. Removing this step would be premature, as it would rely solely on the AI system to make the decision, without any human oversight.

Operations data dependence

Agentic AI workflows for operations depend on a range of data, including process and exception records, project status, budget consumption, and supplier performance. This data is necessary to inform the workflow and ensure that it is working towards the correct goal. For example, a project manager uses the system to track the progress of a project and receive notifications when the budget is being consumed faster than expected.

This data can be found in various systems, such as operations management software, and can be used to inform the workflow and ensure that it is working correctly. Without this data, the workflow would not be able to function effectively, and would likely lead to errors or unintended consequences.

What it changes about operations

Agentic AI workflows for operations can change the way that anomalies are detected and handled, as well as the way that projects are managed. For instance, the system can detect anomalies that are only visible when looking at multiple data points in combination, such as a project that is consuming budget faster than progress. This can help to identify potential issues early on, and prevent them from becoming major problems.

In addition, the system can help to manage projects more effectively, by providing real-time updates on progress and budget consumption. This can help to ensure that projects are delivered on time and within budget, and can also help to identify areas where improvements can be made.

Measuring success

The success of agentic AI workflows for operations can be measured using a range of metrics, including on-time delivery, exception rate, and expediting cost. These metrics can help to provide insight into the effectiveness of the workflow, and can also help to identify areas where improvements can be made. For example, a shift supervisor can use the system to track the on-time delivery of tasks and identify areas where the workflow can be improved.

By using these metrics, organisations can get a clear picture of the effectiveness of their operations workflows, and can make data-driven decisions to improve them. This can help to drive business success, and can also help to improve customer satisfaction.

When it is not the right choice

Agentic AI workflows for operations are not the right choice for every organisation, and are not suitable for every type of workflow. For example, workflows that require a high level of creativity or human judgement may not be well-suited to agentic AI workflows. In these cases, it may be better to use a different approach, such as a customised solution that is tailored to the specific needs of the organisation.

In addition, organisations that do not have the necessary data or infrastructure to support agentic AI workflows may find that they are not the right choice. In these cases, it may be better to focus on building the necessary foundations, such as implementing operations management software, before attempting to implement agentic AI workflows.