AI in Procurement: Automating RFQs and Quote Comparisons

Requesting quotes via email and comparing them in Excel costs buyers hours every day—and still doesn’t provide a reliable basis for decision-making in the end. AI can streamline this process in clearly defined areas, but there are others where it has no place. This article explains which steps can be automated today, what limitations exist, and how to make the transition without changing systems.

Yes, AI helps with quote requests—especially in two areas: creating a well-structured RFQ from unstructured requirements and parsing incoming quotes in various formats. Your purchasing rules and approval processes remain unchanged. In customer projects, the time required to process a request for quotation drops from about two hours to roughly fifteen minutes. Your procurement rules and approval processes remain unchanged. You decide for yourself how much automation to allow.

The most important points at a glance:

• Two hours down to fifteen minutes: That’s how the time required to send a request for quotation changes in customer projects when the need to process the responses is eliminated.
  
• The time-consuming part of the RFQ process isn’t sending the request—it’s processing the responses.
• AI extracts quotes from PDFs, Excel files, and email text and makes line items comparable—without manual data entry.
• Supplier recommendations become more accurate when they’re based on actual order history rather than a general language model.
• Product range limits, pricing rules, and approval processes still apply even when an agent places an order—you determine the level of autonomy.
• A dedicated RFQ tool isn’t necessary: Through the simple system MCP, your own AI assistant can also connect directly to the platform.

Anyone who still sends RFQs via email isn’t doing it wrong—just expensively. The process works, but it doesn’t scale: the more requests, the more Excel spreadsheets, the less clarity. We’ve been building procurement processes for medium-sized industrial companies for 26 years and see this bottleneck in almost every project. Below, we outline where AI has a measurable impact in procurement today, where the limits lie, and what a realistic transition looks like.

The bottleneck isn’t in the request itself, but in the comparison

Sending a request takes minutes. Making the responses analyzable takes hours. This is exactly where the procurement team loses the time it needs for negotiations and supplier development.

The reason is the format. One supplier responds with a PDF, the next with an Excel file, and the third writes the prices directly in the email body. Item names vary, units of measure differ, and tiered pricing is listed in footnotes. Before anything can be compared, someone has to retype everything—and that’s where the errors creep in that become apparent later in the award process.

Definition: An RFQ (Request for Quotation) is a request for specific prices and terms for a predefined service—as opposed to an RFI, which gathers information, and an RFP, which solicits proposed solutions.

We explain in detail which type of request is appropriate in which situation in our article on the differences between RFQ, RFP, and RFI. For the purposes of this article, we’re focusing on the RFQ—that is, the scenario involving a clear need and multiple bids on the table.

Five Stages in the Request Process Where AI Is Making an Impact Today

AI does not play a role throughout the entire process, but rather in five clearly defined steps. It is crucial that a human continues to provide final approval at each of these steps.

The biggest lever is in line four. Everything before that is preparation, everything after is rework—but comparing offers is the step that no one automates today and that everyone finds tedious.

Tip: If you want to measure the benefits of AI in procurement, measure the time from receipt of the last quote to a comparison overview ready for a decision—not the total turnaround time of the request. As a guideline: In customer projects, this figure is about fifteen minutes after the transition, compared to around two hours before.

simpleAI: Agents for the Request Process Instead of a Chatbot

With the simpleAI Agent Suite, we’ve broken down the steps described above into individual agents, each of which handles exactly one task. No assistant that can do a little bit of everything—instead, clearly defined functions within the ongoing process.

Agents work within the platform, not alongside it. An assigned quote becomes an order, and the order flows through the SAP integration via OCI Punchout into the ERP system. No additional tool is created that someone has to maintain.

Why Purchasing Data Makes the Difference

A supplier recommendation is only as good as the data it’s based on. This is exactly where a procurement platform differs from a pure AI provider.

simpleAI draws on the order history of over 28,000 active buyers and more than 1,000 connected suppliers, accumulated over 26 years. Which suppliers reliably deliver which product categories, where delivery times are realistic, and which suppliers actually respond to which types of inquiries—this information is contained in this data, not in the training data of a general-purpose language model.

A general language model knows half the internet, but not your purchasing process. It doesn’t know what a normal delivery time is for your business or which rule should stop an order. We explain in more detail why, in purchasing, it’s not the model that decides but the data foundation in our article on the simpleAI Engine.

Tip: Ask every AI provider in procurement what data foundation their supplier recommendations are based on. A model without its own transaction data can only estimate relevance.

For the request process, this means that buyers can query recurring orders and product lines and view existing RFQs without opening the interface. Approvers are shown a list of open shopping carts and can identify where an order deviates from historical patterns. Login is handled using existing credentials; no IT project is required.

What’s crucial is what doesn’t happen: The agent doesn’t receive a master key, but rather a plant ID card. Product range restrictions, price limits, and approval processes apply just as they would if a human had clicked.

Which indirect procurement processes AI handles today

The quotation process is the most visible use case, but not the only one. In indirect procurement, the bulk of the effort is spent on tasks that occur daily and are unremarkable on their own: recording requirements, triggering reorders, posting to accounts, and verifying invoices.

Not all of these steps can be automated to the same extent. The key factor is determining where rules apply and where human judgment remains necessary.

Automation in indirect procurement rarely fails due to technical issues; rather, it often fails because no one can trace who approved what. That is why approval remains a system rule and not a model decision.

Definition: In procurement, the degree of autonomy refers to the extent to which a system is permitted to complete transactions without human confirmation. It is defined for each process step and product group and is independent of how powerful the model used is. We discuss which approval levels can be implemented without slowing down the process in our article on approval workflows in procurement.

Why AI in Procurement Is Limited Without an ERP Connection

The existing text explains how a purchase order moves from the quotation process into the ERP system. For automation, however, the reverse direction is more important: what AI retrieves from the ERP system.

An account assignment can only be suggested if cost centers, general ledger accounts, and PSP elements are known from the ERP system. A purchasing cycle can only be identified if the complete purchase order history is available. A variance can only be reported if the purchase order, goods receipt, and invoice are based on the same data. Without an integration, AI lacks the foundation in all three cases and can only make estimates.

This is precisely why AI tools that aren’t integrated into the process have little impact on indirect procurement. They sit alongside the process, not within it. An assistant who drafts a purchase requisition—which someone then enters into the ERP system—has not eliminated the media discontinuity, but merely shifted it.

simple system is integrated with SAP ERP, SAP S/4HANA, and Microsoft Dynamics 365 via the OCI interface and EDI. The simpleAI agents work with the same data used to process the purchase order. The SAP Integration page explains how the integration works technically and what it entails for IT .

Tip: For any AI tool used in procurement, check whether it has read-write access to the ERP system or only read access. A tool that makes suggestions which someone then has to manually enter shifts the workload rather than reducing it.

Human in the Loop: You determine how much autonomy the AI is given

AI in the engine room, not in the showcase. This principle doesn’t mean that AI in procurement is never allowed to make decisions. It means that you determine where the line is drawn.

The degree of autonomy ranges from the approval of each individual transaction to agents that you connect via MCP. What remains the same at every level are your rules: budget limits, approval hierarchies, and product-group-specific ordering rights are defined as system rules—not as the judgment of a language model. An order that violates the purchasing policy is blocked—regardless of whether it was initiated by a human or an agent.

Definition: In procurement, “Human in the Loop” refers to an automation principle in which AI handles routine steps, while rules and approval levels are defined by the company and remain unchanged at every level of autonomy.

This is the difference between automation and loss of control. We’ll address separately how approval workflows in procurement can be designed so that they don’t slow down the process.

From Excel to AI-powered quote comparison in four steps

The transition doesn’t have to be a year-long project. It requires an honest assessment of the current situation and a clear plan of action.

1. Measure your request volume: For one month, track how many RFQs you send out and how many quotes you receive in return. If you have fewer than 20 requests per month, automation isn’t worth it.
2. Identify time-wasters: Track how long it takes to process the returned quotes. This figure is your business case—not the total turnaround time.
3. Start with one agent: Begin with quote comparisons, not demand. That’s where the benefits are most quickly apparent and the transition is the smallest.
4. Define rules before rollout: Determine what the AI prepares and who approves it—before the first requester uses the system.

Conclusion: AI handles the comparison, not the decision

AI in procurement is useful when it transforms quote comparison from a data-entry task into an evaluation task. It replaces neither the negotiation nor the award—it frees up time for them. Anyone still comparing quotes in Excel today is wasting time on a step that has been automatable for two years.

→ Book an AI Check: 15 minutes, honest assessment
Using your actual requests, we’ll show you which simpleAI agents have a measurable impact for you—and which ones you can do without.

FAQ – Frequently Asked Questions About AI in Procurement