At one company, creating a purchase order took about 15 minutes. A buyer copied data from a supplier quote, checked the price, formatted a PDF, sent it for approval, uploaded it, and logged the order in a spreadsheet. Every step was manual.
The same task now takes about 3 minutes.
That improvement did not come from a large procurement platform. We built a small tool that reads the supplier offer, extracts the relevant fields, prepares the order, stores the document, and logs the transaction. A person reviews the result and approves it.
Invoice logging was similar. Someone opened each PDF, typed the data into a spreadsheet, and filed the document. Our extraction tool now handles more than 95 percent of those fields correctly in the measured workflow. Finance, procurement, and logistics use it because it saves time, not because anyone ran an adoption campaign.
We also processed about 50,000 archived documents with inconsistent names and folders. A tool classified each file, created a standard name, and moved it to the relevant supplier folder. The job ran for weeks. Doing it manually would have taken months.
These are good uses of AI because the inputs are repetitive and the output can be checked. Extraction, classification, logging, and naming consumed time without needing much judgment. Automating them gave people more time for work that did.
Negotiation is different. AI can summarize spend, compare quotations, and prepare market information. It cannot read the supplier's position with the same context as a buyer who knows the relationship. It does not know when a delay is a tactic, when a margin is genuinely exhausted, or when closing now is better than asking for one more concession.
Supplier evaluation has the same limit. A scoring model can compare responses, but important information often appears when something goes wrong: a missed deadline, an unclear drawing, or a quality dispute. How a supplier handles that conversation matters more than a polished questionnaire.
Contract review sits between the two. AI is useful for finding clauses and comparing language. The difficult part is deciding which risk to accept and what to trade for it. That decision depends on the product, the supplier, and the cost of failure.
AI has made procurement faster where the work was tedious and structured. It has not removed the parts that depend on judgment. The best results I have seen came from fixing one recurring task, measuring the change, and keeping a person responsible for the output.