Agentic workflows for procurement
Agentic AI workflows for procurement streamline purchasing by integrating data from various systems, enabling agents to manage spend and contract terms from end to end, and providing recommendations to inform human approval decisions, thus reducing fragmented spend and ensuring contract compliance.
A procurement manager stares at a dashboard that nobody opens, a relic of a previous attempt to bring order to the chaos of purchasing and invoicing across the organisation. The problem is not just one of data visibility, but of taking action on that data to actually manage spend and contract terms.
What separates an agent from a chatbot
An agent is not just a chatbot that answers questions about procurement data. What separates an agent from a chatbot is its ability to use tools, maintain state, and work towards a goal. This means an agent can take multi-step action, using the procurement system, the contract database, and the invoice processing tool to manage the procurement process from end to end.
A chatbot, on the other hand, can only answer a single question at a time, and does not have the ability to take action or maintain state. This limits its usefulness in managing complex procurement processes. For example, a category manager can ask an agent to compare two suppliers, and the agent can use the procurement system to gather data on the suppliers, the contract database to check the terms of the contracts, and the invoice processing tool to check the payment history.
Where a human approval step belongs
Many procurement processes have a human approval step, where a manager must approve a purchase order or contract before it can be executed. While it may be tempting to remove this step and automate the entire process, this is usually premature. Human approval is often necessary to ensure that the procurement process is aligned with the organisation's goals and policies.
However, an agent can still be used to manage the procurement process up to the point of human approval, and can even provide recommendations and data to inform the approval decision. For example, an agent can gather data on the suppliers, check the contract terms, and provide a recommendation on which supplier to choose, but the final approval decision is still made by a human.
Procurement data dependencies
An agentic workflow for procurement depends on having access to a range of procurement data, including purchase orders, supplier contracts and terms, invoice history, and category structure. This data must be accurate, up-to-date, and accessible to the agent in order for it to make informed decisions and take effective action.
For example, a shift supervisor can check the maintenance log to see which equipment is due for maintenance, and an agent can use this data to generate a purchase order for the necessary parts. The agent can then use the procurement system to send the purchase order to the supplier, and the invoice processing tool to track the payment.
This integration with operations systems is critical to the success of an agentic workflow for procurement.
Impact on spend and contract terms
An agentic workflow for procurement can have a significant impact on spend and contract terms. By managing the procurement process from end to end, an agent can help to reduce spend fragmented across suppliers, and ensure that contract terms are aligned with what is actually paid.
For example, an agent can analyze the invoice history to identify areas where the organisation is being overcharged, and use the contract database to check the terms of the contract and negotiate a better price. The agent can then use the procurement system to generate a new purchase order with the updated price.
Measuring success
The success of an agentic workflow for procurement can be measured using a range of metrics, including spend under management, price variance to contract, and supplier concentration. These metrics provide insight into the effectiveness of the agent in managing the procurement process and reducing spend.
For example, an organisation can track the spend under management to see how much of its procurement spend is being managed by the agent, and use this data to identify areas for improvement. The organisation can also track the price variance to contract to see how well the agent is doing at negotiating prices that are aligned with the contract terms.
Limitations and wrong choices
An agentic workflow for procurement is not a solution for every procurement problem. It is not suitable for organisations with very simple procurement processes, or those with highly variable or unpredictable procurement needs. In these cases, a more traditional approach to procurement may be more effective.
Additionally, an agentic workflow for procurement requires a significant investment in data quality and systems integration, and may not be suitable for organisations that do not have the resources or expertise to support this. If you are considering implementing an agentic workflow for procurement, you can contact us to discuss your options and determine whether this is the right solution for your organisation.
Frequently asked questions
What is the difference between an agent and a chatbot in procurement?
An agent uses tools, maintains state, and works towards a goal, whereas a chatbot only answers questions and does not take action.
Where does human approval fit into an agentic workflow for procurement?
Human approval is necessary for ensuring alignment with organisational goals and policies, and an agent can provide recommendations and data to inform this decision.
What data is required for an agentic workflow for procurement to be effective?
Accurate and up-to-date data on purchase orders, supplier contracts, invoice history, and category structure is necessary for informed decision-making.
How can the success of an agentic workflow for procurement be measured?
Metrics such as spend under management, price variance to contract, and supplier concentration provide insight into the effectiveness of the agent in managing procurement processes.