Procurement / supply chain specialist

A national agency's firm survey, and the reason it is recorded under worker adoption rather than deployment is written into the same report. Asked how generative AI is used inside the organisation, 63.3% of companies answered that individuals use it for their work and 44.8% that individuals or departments are trialling it, while only 11.9% said it is built into a department's work process and 18.6% into a company-wide service. That is the distinction this stage exists for: people are using the tool, and employers have largely not rebuilt anything around it. IPA is an incorporated administrative agency under Japan's METI and sells none of the tools it counts, which is what makes this a better instrument than vendor telemetry. Two limits matter for this page. The figure resolves to a department, not a task — it establishes that this function is among the least-served, not which of its duties are touched; the report's separate question on purposes puts summarising, translating and proofreading documents at 82.5% and information search, gathering, analysis and reporting at 77.0%, among companies that have adopted, are trialling or are considering AI. And the survey ran from mid-April to mid-June 2026 in Japan, among companies that answered; only 41.6% of firms with 100 or fewer employees report any DX activity, against 98.7% of firms with more than 1,000, so an average across companies hides a sharp split by size.

Measured, large-scale use of a tool for real work, where the decision to use it was the worker's rather than an employer's. It is more than a capability record — the work is real, not a demo — and less than a deployment record, because no employer put it into production, required it, or built a process around it. Weighted `cautious`: `automating` means the machine can do the task AND there are adoption signs, and this is an adoption sign — but usage can be experimental, and much of the measurement comes from a party with a stake, so one record is never enough and two independent ones are. Note who is counting. Vendor telemetry sees this directly and sells the tool, so such a record names that stake in its scope; a statistics agency asking firms whether their workers use AI in tasks sees the same channel with no stake at all, and that is the better source where it exists.

Conversations on one vendor's consumer and professional product, classified by that vendor to occupational task statements, worldwide, for the month beginning 2026-05-01. The figures below are not printed in the report; they were computed here from the release's own dataset file at release_2026_06_26, filtered to the global rows, and anyone can recompute them. Automation share means the share of conversations whose shape was directive or a feedback loop rather than learning or iteration; it describes how a person used the tool, not whether anyone lost work. For finding and comparing suppliers, the task statement about researching and evaluating suppliers on price, quality, service and reliability reads 57.71% automation on traffic that is 51.22% work. For the order and the paperwork, preparing and processing requisitions and purchase orders reads 67.42% automation on traffic that is 88.30% work, which is the cleanest row in this occupation. The reason to trust the second more than the first is the contamination, and it is the most important thing in this record: the highest automation share here belongs to reviewing catalogues and directories to locate goods, at 80.71%, on traffic that is only 20.52% work, because people researching their own purchases land on the same task statement as buyers. The pattern runs the other way at the bottom. The three statements that are almost entirely work all sit below the median: writing product specifications 45.90% on 90.16% work, negotiating and administering supplier contracts 38.30% on 89.36% work, and analysing price proposals 31.73% on 93.33% work, against a median of 48.55% across the 2,691 task statements with a published automation share (50.00% across 3,650 nodes if detailed, intermediate and generalized work activities are counted too). Read honestly, this says that among the purchasing tasks people actually bring to this tool as work, delegation sits at or below the middle of all occupations, and the tasks that look most automated are the ones consumers are asking about. Stability: the April window preserves the same ordering for seven of these eight statements, with negotiation and price analysis swapping the bottom two places. What it does not establish: nobody in this data said they were a buyer, since the task assignment is made by a classifier; the shares of total usage for these nodes are 0.00% to 0.05%, so the underlying counts are small and are not published; and a task statement absent from the file was suppressed rather than measured at zero. The publisher operates the tool and sells it, so only the party with a commercial interest can see this data at all, which is why a single record of this kind is never enough on this site.

Measured, large-scale use of a tool for real work, where the decision to use it was the worker's rather than an employer's. It is more than a capability record — the work is real, not a demo — and less than a deployment record, because no employer put it into production, required it, or built a process around it. Weighted `cautious`: `automating` means the machine can do the task AND there are adoption signs, and this is an adoption sign — but usage can be experimental, and much of the measurement comes from a party with a stake, so one record is never enough and two independent ones are. Note who is counting. Vendor telemetry sees this directly and sells the tool, so such a record names that stake in its scope; a statistics agency asking firms whether their workers use AI in tasks sees the same channel with no stake at all, and that is the better source where it exists.

One of the world's largest buyers describing its own procurement. Procurement employees specialised in restructuring and supply reliability constantly monitor the financial situation of suppliers worldwide, and supply risks should be identified by early warning systems. For its human-rights and environmental due diligence, Group Procurement Sustainability runs ongoing, risk-based media monitoring of relevant suppliers with a specialist IT tool from an external service provider; in 2025 it covered more than 40,000 direct suppliers (more than 39,500 in 2024), accounting for 96% of purchasing volume (89%), and abnormalities are examined further by people. The report calls these systems and IT tools, not AI, and gives no figures on how often a flag was right or what it changed.

An employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.

Buyers and purchasing agents are one of 44 US occupations in the benchmark, across the nine sectors contributing most to US GDP; 1,320 tasks in the full set with at least 30 per occupation, each built from actual work product by a professional averaging 14 years of experience and covering the majority of that occupation's O*NET work activities. What is graded is a deliverable, one shot, by a blind pairwise expert comparison - not a job. There is no client, no revision round, no negotiation and no consequence for being wrong. The headline figure (47.6% of one model's gold-subset deliverables rated better than or as good as the expert's) is for the model generation of September 2025 and has been superseded; more usefully, the paper reports that win rates are highest on tasks taking 0-2 hours and decline steadily as the task gets longer. Per-occupation rates are published only as a figure, so no purchasing-specific number is quoted here. OpenAI designed the benchmark, ran it, and sells models measured by it.

A demo, benchmark or paper shows the task can be done. Updates what the technology can do — not what employers will do.

A peer-reviewed experimental study in which people playing suppliers negotiated with generative-AI chatbots prompted either competitively or collaboratively. It reports that a competitively prompted chatbot achieves a higher price discount, better payment terms and a quicker negotiation, while suppliers trust a collaboratively prompted chatbot more and show higher outcome satisfaction and a stronger desire for future interaction. It is a controlled experiment, not a deployment, and participants were not in real long-term supplier relationships.

A demo, benchmark or paper shows the task can be done. Updates what the technology can do — not what employers will do.

An EU regulation binding operators that place seven commodities — cattle, cocoa, coffee, oil palm, rubber, soya and wood — and products made from them on the EU market or export them. Article 4(1): operators shall exercise due diligence before placing products on the market in order to prove that they comply, and may not place them without a due diligence statement; Article 9(1)(d) requires the geolocation of all plots of land where the commodities were produced. As amended in December 2025, these obligations apply from 30 December 2026, later for micro and small operators, and downstream traders mainly keep supplier details and statement references. It creates the obligation to prove provenance for those commodities; it does not say who inside a company does the work.

Regulation, subsidy or public procurement is requiring or funding adoption — the mirror of a constraint. It shows adoption is being required, not that it has happened, so one mandate is never enough on its own; two independent ones are.

An article co-written by two Walmart International procurement executives and two academics. It says Walmart deployed AI-powered negotiation software with a text-based interface to connect with suppliers, and that so far the chatbot is negotiating and closing agreements with 68% of suppliers approached; it frames the target as the roughly 20% of suppliers that had signed agreements with cookie-cutter terms that are often not negotiated. It covers the long tail of small suppliers nobody used to negotiate with, not strategic relationships, and it is the buyer's own account.

An employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.

A US federal statute. Section 3 directs Customs and Border Protection to presume that goods mined, produced or manufactured wholly or in part in the Xinjiang Uyghur Autonomous Region, or by listed entities, are made with forced labour and barred from import; the exception requires the importer to have complied with the government's guidance, to have answered all of Customs' inquiries, and to show by clear and convincing evidence that the goods were not made wholly or in part with forced labour. The Act also calls for guidance to importers on due diligence and effective supply chain tracing. It puts the burden of proving where things came from on the importer when goods are held; it does not require importers to file an annual report (the annual reports it requires are the government task force's), and it says nothing about who inside a company does the tracing.

Regulation, subsidy or public procurement is requiring or funding adoption — the mirror of a constraint. It shows adoption is being required, not that it has happened, so one mandate is never enough on its own; two independent ones are.

United States, federal procurement. The regulation says contracts may be entered into and signed on behalf of the government only by contracting officers, and its definitions count orders such as purchase orders as contracts. It reserves the binding signature, not the matching of invoices or the paperwork around an order; agencies' 2025-26 overhaul deviations were not checked.

Failure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.

The half of this job that looks like expertise from outside — finding suppliers, comparing quotes, raising orders — is the exposed half, and it is what you will be given first. The half that holds is about people: what a supplier is not telling you, and who your company protects when there is not enough. Neither is learned from a system; both are learned on calls you are not yet invited to. Ask to listen in early.

Your leverage moved from knowing the market to knowing the counterparties. A quote comparison is now something anyone can produce in minutes; what nobody can produce is the judgement that this supplier will hold through a bad quarter and that one will not. Notice that this is also the knowledge your company has no record of — which is a risk for them and a position for you.

Stay and strengthen

Move to the categories with few suppliers

Where there are three possible suppliers instead of three hundred, comparison is worthless and the relationship is everything — which is the half that holds.

Reshape the role

Own what the company can prove about its chain

The US already puts the burden of proving where goods came from on the importer, and the EU will require the geolocation of every plot from 30 December 2026 — neither says who inside the company does it. In most companies nobody has been given it.

Adjacent move

Cross to the side that plans demand

Most supply problems are demand problems arriving late, and a buyer who has watched shortages arrive knows things the forecast does not.

Ask it per task, because the split here is unusually clean. Finding suppliers, comparing quotes and matching paperwork are structured work, and they are the two tasks on this page judged to be automating. Judging whether a supplier will hold is now helped by monitoring tools whose flags people examine, and deciding who goes short has not moved. A signal you can check yourself: of your last ten decisions, how many needed information a search could have produced. If most did, the exposure is real and the move is toward the ones that did not.

It can, and doing so is the most expensive mistake in this job. The cheapest quote is cheapest for a reason, and the reasons — thin margins, an overloaded plant, a customer they are about to lose — are exactly what does not appear in the quote. Procurement is not a price-selection problem; it is a counterparty-risk problem wearing a price tag. A system that ranks by price optimises the thing that is easy to measure, which is how a supply chain ends up with a single point of failure nobody chose.

Depends which half you would be doing, and it is worth settling before you move. A role that is orders, matching and quote comparison sits in the exposed half and is what most junior procurement jobs are. A role with category ownership and supplier relationships sits in the other half, and it usually requires the first kind of role first — which is the uncomfortable shape this site keeps finding. In the interview, ask who decides when there is not enough to go round.

No, and the difference matters more than it sounds. In public procurement the rules are law rather than company policy, the process is designed to be challengeable, and an automated decision inherits that challengeability. That changes both what can be automated and who has to be able to explain it — the same pattern as the government service counter, and a different page from this one.