Procurement with AI can automate price comparisons, quotation requests and decision summaries without allowing the software to place orders itself. The AI agent collects and checks information, but a named person approves every binding order. This requires clear access rights, approval limits and complete documentation.

How does procurement with AI work without automatic ordering?

An AI agent handles preparatory tasks but is not authorised to place an order. The final decision remains with management, the head of procurement or another person who has been expressly designated.

Bitkom (Germany’s digital industry association) describes AI agents as systems that complete tasks independently, such as researching price comparisons and obtaining quotations. Unlike a simple chat programme, an agent can carry out several defined steps in sequence.

A possible process for purchasing workwear could look like this:

  1. A responsible person provides a list of requirements with sizes, quantities and the requested delivery date.
  2. The agent contacts approved suppliers or evaluates their current quotations.
  3. It standardises prices, shipping costs, minimum quantities, payment terms and delivery times.
  4. It prepares a decision summary with two or three suitable quotations.
  5. A person reviews the summary and approves the order in the existing system.

This does not automate the purchasing decision, but its preparation. It reduces the risk of an incorrect price, the wrong quantity or an unverified supplier leading directly to expenditure.

Which procurement tasks are suitable for AI agents?

Recurring, comparable purchases with clear requirements are suitable. These include consumables, workwear, standard spare parts, packaging and regularly required materials.

This is particularly useful when employees currently have to gather the same information from emails, supplier portals, PDF quotations and Excel spreadsheets. AI automation for recurring office work can bring these sources together and evaluate them using defined criteria.

The decision summary should include at least the following information:

Not every purchase is suitable. Custom products, safety-critical spare parts or services whose scope is difficult to compare usually require specialist review. The agent can prepare documents, but it should not assume technical suitability if this is not clear from the data.

Which approval rules prevent uncontrolled expenditure?

The AI agent should only receive the rights it needs for research, communication and evaluation. Ordering rights, access to payments and changes to bank details should remain separate and under human control.

For controlled use, Bitkom recommends reliable data, connected systems and clear rules defining what an agent may do independently and when human approval is required. For example, a company rule could specify that the agent may send requests and compare quotations, but may not confirm orders.

A simple approval matrix could look like this:

These amounts are examples, not general recommendations. They must reflect the company’s size, liquidity and allocation of responsibilities. The system should also record which sources were used, which quotations were rejected and who approved the order.

The agent should also communicate only with approved suppliers or clearly flag new providers. Login details belong in a protected access management system, not in freely accessible work instructions. If requests contain personal data, data protection requirements and retention periods must also be reviewed.

What does the technical infrastructure for procurement with AI cost?

The ongoing software licence is often not the largest cost. Most of the work relates to data quality, computing capacity and connections to inventory management, email and document storage systems.

According to the Bitkom presentation on artificial intelligence in business 2026, 51 per cent of companies cite cloud services and computing capacity as a major cost, 50 per cent cite data preparation and 41 per cent cite connections to existing systems. By contrast, AI subscriptions and licences are a major cost for only 21 per cent.

Before implementation, the company should therefore establish where item numbers, supplier master data and previous quotations are stored. If the same parts are stored under different names, even a technically capable agent cannot produce reliable comparisons.

Missing interfaces do not need to be replaced immediately through a large software project. For a limited trial, an approved email inbox, a well-organised data folder and a fixed output template may be sufficient. Only after the trial should the company decide whether a direct connection to its inventory management system is cost-effective.

How can SMEs start with a controlled pilot project?

The first step should focus on a single, frequently purchased product category. During the pilot, the company measures the time required, data quality, price differences and any necessary rework.

For example, a suitable pilot could cover the procurement of gloves, cleaning products or packaging for eight weeks. Before it begins, suppliers, comparison criteria, approval amounts and responsible people should be defined in writing.

The Bitkom study on the use of AI in companies, published on 14 September 2026, reports that 11 per cent of the companies surveyed already use AI agents, a further 29 per cent plan to use them and 31 per cent are discussing their use. For 24 per cent, AI agents are currently not under consideration.

According to Bitkom, the findings are based on a representative survey of 603 German companies with at least 20 employees, conducted between calendar weeks 28 and 33 of 2026. Very small businesses with fewer than 20 employees are therefore not directly represented.

Among the companies using AI, Bitkom reports that 48 per cent have significantly accelerated internal processes and 45 per cent have seen a measurable contribution to business performance. However, these figures do not prove that every procurement agent will achieve the same results. The specific process, the available data and the amount of manual rework are what matter.

ibx company advises small and medium-sized businesses in Hannover and the surrounding region on the selection and introduction of suitable digital processes. If you would like to assess whether a procurement agent is suitable for your business, you can arrange a no-obligation initial consultation about your procurement process.