Offers compared and confirmations checked before your buyers decide

Your buyers decide from a prepared comparison instead of retyping offers. Procurement software vendors describe the problem plainly: “Einkäufer sind operativ oft überlastet”, buyers are often overloaded with operational work, as the procurement software company AskLio put it in Logistik Heute in 2025. Tacto, another procurement software vendor, lists the causes on one of its webinar pages: “Manuelles Aufsetzen, verstreute Daten, uneinheitliche Angebotsformate und aufwendige Vergleiche”, manual setup, scattered data, inconsistent offer formats and laborious comparisons.

How widely AI is used in purchasing depends on the survey. In the BME and Apsolut impulse survey reported by Logistik Heute in January 2024, almost 20% of procurement respondents used AI, and the areas named included invoice control, offer comparability and contract drafting. Fewer than 1% saw headcount savings as the lever. In a 2025 vendor study by Ivalua with Sapio Research among 600 decision-makers in Germany, France, the US and the UK, 65% of German companies already used AI tools in procurement and supplier management. The two surveys differ in date, sample and sponsor, so together they do not show a trend.

• One comparison table for every request for quotation, instead of offers retyped from PDF and Excel.
• Every order confirmation checked against the order, not only the ones someone has time for.
• Supplier certificates and declarations tracked, with a reminder before they expire.
• Answers to spend questions from your ERP data without waiting for a report.
• Effort per offer comparison and per confirmation measured on your own documents, before and after.

The agent reads, compares and drafts, and brings together information that is spread out today. Tacto describes where it sits: “Oft liegen Teile der Informationen in Dateien, andere in Postfächern und wieder andere in persönlichen Notizen”, often part of it in files, part in mailboxes and part in personal notes. These are the tasks we build.

• Comparing supplier offers: prices, price scales, units, delivery times, payment terms and Incoterms from PDF, Excel or the mail body, in one table against your request for quotation, with missing items and deviations from your specification marked.
• Checking order confirmations against orders: each line of the supplier's confirmation against the purchase order in SAP MM or your ERP, covering material, quantity, price, delivery date and terms. Mismatches go to the buyer, and matching confirmations are prepared for entry.
• Supplier documents: certificates, declarations and insurance proofs read for issuer, scope and expiry date, filed in your document management system such as d.velop, ELO or SharePoint, with a reminder to the buyer before a document expires.
• Contract terms: supplier terms and contract drafts compared with your purchasing terms and playbook, with deviating clauses on liability, payment, retention of title or price adjustment quoted. Your legal team decides.
• Spend questions, such as how much you bought from one supplier across all plants: the agent writes a query on your ERP data, the database computes the figure, and the answer shows the query so a controller can check it.
• Requisitions, requests for quotation and supplier mail: drafts built from the request and your article data, and reminders to suppliers about missing confirmations, sent after approval. Stadtwerke Pforzheim, for example, uses a bot called “Banfred” that helps staff create purchase requisitions in SAP, as reported by the trade paper ZfK in 2026.

Every decision that commits money or changes a supplier relationship stays with people. The agent prepares the basis and records the decision.

• Awarding: which supplier gets the order, and on what terms, is the buyer's decision. The comparison table prepares that decision.
• Changes in the ERP: confirmed dates, prices and quantities are written only after the buyer accepts them, and release strategies in SAP stay as they are.
• Messages to suppliers: nothing leaves the company without approval, unless you decide that a named type, such as a reminder about a missing confirmation, may go out automatically.
• Contract deviations go to legal, and new suppliers are approved by the people your process names.

The agent connects to the systems your buyers already use and replaces none of them.

• SAP MM or S/4HANA, or another ERP: read access to purchase orders, requisitions, info records and supplier master data. Write access only through approved actions such as recording a confirmed delivery date.
• The purchasing mailbox in Outlook or Exchange, and supplier portals where they exist, as the entry point for offers and confirmations.
• d.velop, ELO or SharePoint for contracts and supplier documents, with their permissions applied, so a buyer sees only what they may already open today.
• Hosting in your own cloud account or on your own servers with an open-weight model, a model whose weights you can run yourself. Supplier prices and conditions are trade secrets, so they are processed only in your environment, and one LLM gateway, a controlled access point to all large language models with keys, budgets and logs, keeps every call visible.
• Build or buy: procurement suites such as SAP Ariba, Coupa or JAGGAER and specialist tools such as Tacto or AskLio cover parts of this job, and SAP itself offers an app that predicts late deliveries for purchase order items, as PIKON described in 2025. If you run one of these and it covers your documents, extend it. A custom build pays off where offers and confirmations arrive outside the suite by mail, PDF or Excel, where purchasing data must stay on your own servers, or where one check spans ERP, document management and mailbox.

Before go-live, the system runs an acceptance test on a set of your past offers and order confirmations where the correct values and the real deviations are known. Results are reported per field and per supplier format, because one supplier's confirmation can be easy to read and another's hard.

After go-live we measure the effort per offer comparison and per confirmation before and after, on your own documents, and the same tests run after every change of model, prompt or supplier format. Ingestion from the mailbox and the ERP interface are monitored, with runbooks, written operating procedures, for when a supplier changes its layout.

• Field accuracy for price, quantity, unit, delivery date and terms, measured on your documents.
• Deviations found and deviations missed. A missed price deviation costs more than a false alarm, so the threshold is set together with your buyers.
• Answers to spend questions checked against reports your controlling team already trusts.
• Shadow mode on live documents: the agent's results are recorded but not used, and compared with what your buyers did.

Supplier offers contain prices and conditions that suppliers share in confidence, often under a confidentiality clause, plus names and contact details of their staff. They are processed only in your environment, and if a model provider is involved in your cloud account, its terms must exclude use of your data for training. The purchasing tasks on this page are not among the AI Act's high-risk uses.

Every decision is recorded with the document, the deviation found, the buyer's choice and the time, which gives internal audit a clear trail. Such records could also show how fast each buyer works, and a system that is objectively capable of monitoring performance falls under § 87 (1) no. 6 BetrVG, as Bitkom's 2026 guide on AI and co-determination notes. So we keep the approval record your process already requires, add no reports on individual buyers, and store the AI logs with pseudonymised IDs and agreed retention. You decide the details with your data protection officer, works council and lawyers. This describes the technical side and is not legal advice.

Order confirmations are often the best first step: high volume, clear rules, a measurable result and low risk, because the buyer approves every deviation. Offer comparison for one material group is a good alternative if your bottleneck is requests for quotation.

• Collect a sample of past confirmations or offers from your main suppliers, with the orders they belong to.
• Measure the baseline: effort per document and the deviations found today.
• Agree which ERP fields the agent may prepare and who approves them.
• Run the acceptance test, then shadow mode, then live use with approval. Each step has a fixed price, agreed in writing after the first call.