Executive summary
Price increases no longer cover rising input costs, so the cost base has to come down, and 50–80% of it is external spend. AI agents let Procurement work all of it: rebuilding should-cost, checking every invoice, and negotiating tail spend within buyer-set mandates. The key is an agentic Procurement workforce: governed agents on one platform. Start with one category and prove value in about three months.
What an agentic Procurement workforce looks like in practice
A supplier writes in asking for +9% on packaging film. In most companies, the buyer spends the next few weeks piecing together the evidence, and often accepts part of the increase to protect supply. With an agentic Procurement workforce, five AI agents build the case in parallel, the buyer makes the call, and the evidence pack is ready the same day. That shift, from weeks of data gathering to same-day decisions, is where the next wave of cost reduction comes from.
An agentic Procurement workforce is a governed set of AI agents that work alongside buyers from demand to payment. Each agent has a job, a mandate, and a human owner. They all run on one platform, built from the same skills and connectors. Here's what that looks like in three everyday situations.
A supplier asks for +9% on packaging film
Here's how the five agents split the work on that +9% request. An intake agent reads the letter, a should-cost agent rebuilds film cost from the polyethylene index, a contract agent checks the indexation clause, and a commodity agent adds the resin outlook. The negotiation agent drafts the counter-offer. The buyer decides.
The annual packaging tender
Take an annual packaging tender and break it into its subtasks: spend data, should-cost, strategy, RFQ and specs, bid comparison, award scenarios, negotiation, contract, approvals, and invoice checks. Today buyers touch every one of them, and much of that time goes on gathering and preparing data.
With an agentic workforce, the picture changes. AI takes over the data-heavy steps, such as pulling spend data, comparing bids, and checking invoices. It also helps with other tasks, such as building the should-cost model, drafting the RFQ, modeling award scenarios, and reviewing the contract. Buyers keep the strategy, the negotiation, and the approvals, and the data work around them shrinks. The time saved goes back into what actually moves cost: negotiation and supplier relationships.
The thousands of purchases no one has time for
Every company has a long tail of small purchases that no buyer has time to tender, and invoices that only get spot-checked after payment. A conversational front door routes each request, and agents run three-bid quotes within buyer-set mandates. Before payment, another agent checks every invoice line against the contract. This is already happening: Honeywell has negotiated over $500m of spend with AI agents,[1] and EG Group prevented £20m of overpayments.[2]
The platform matters as much as the agents: it enforces mandates, price floors, and walk-away points in code. A negotiation agent that reads supplier email, sees your price data, and writes to suppliers always needs human approval before sending.
Why now: the cost base has to come down
Why does this matter now? Because the old way of covering cost increases has run out.
The cost floor has moved, and the swings are bigger
Almost every major input costs more than it did before 2020, and 2026 added a record oil supply shock on top. The World Bank’s April 2026 Commodity Markets Outlook calls it the largest oil supply shock on record and forecasts energy prices up 24% this year.[3] European gas has spiraled between February and September and sits at multiples of the US price.[4] Container freight sits at multiples of its 2019 average after rerouting around the Red Sea and the Gulf.[5]
Then there are the layers that don’t reverse when commodities fall. Euro area hourly labor costs rose 3.1% year on year in Q2 2026.[6] The average US statutory tariff rate is 11%.[7] New EU regulations, including the Carbon Border Adjustment Mechanism (CBAM), packaging rules, and deforestation rules, came into force throughout 2026.
Energy and freight increases hit almost every category: ingredients, packaging, components, and logistics. A higher floor would be manageable if it were stable. It isn’t. Pricepedia’s volatility index shows that price volatility in 2020–24 exceeded 2015–19 in every commodity category with gas volatility significantly higher.[8] Swings of this size move budgets, margins, and cash, not only unit prices.
Price increases can’t cover rising costs anymore
For five years, consumer goods companies grew mainly through price. That route is closing. Western European consumer goods prices are 31% above 2020 while volumes are up just 2%.[9] Private label reached a record 38.8% share of European grocery in 2025.[10] Retailers can’t absorb more either and the squeeze is visible in the P&L statement. The conclusion is simple. If price can’t cover cost, the cost base itself has to come down.
Leaders are acting, but often on the smaller pool
Executives know this. 72% of Chief Procurement Officers (CPOs) in Deloitte’s 2025 Global CPO Survey name margin optimization through cost reduction as a top priority.[11] 56% of CFOs put enterprise-wide cost optimization in their 2026 top five (Gartner).[12] Meanwhile, CEO confidence in revenue growth has dropped to 30%, from 56% in 2022 (PwC).[13]
And there’s money behind it. Nestlé’s Fuel for Growth targets CHF 3.0bn by end-2027. Heineken’s EverGreen 2030 aims for €400–500m in gross savings a year. Ahold Delhaize’s Save for Our Customers targets €5bn over 2025–28. Philips and Bosch Mobility each run €2.5bn programs.[14]
Here’s the thing: many of these programs lead with headcount and overhead. But external spend (what a company buys) is typically 50–80% of the cost base.[15] Headcount reductions are largely one-off. External spend is re-priced with every contract, index move, and tender. That valued pool is where AI in Procurement earns its keep.
Six places AI takes cost out
Procurement still uses the same strategies: manage spend, change specifications, leverage supplier competition, and seek joint advantage with suppliers.[16] What has changed is the ability to apply them.Procurement workload was expected to grow 10% in 2025 while budgets were expected to increase by just 1% according to The Hackett Group.[17] Only a portion of identified savings typically reach the P&L.[18]
This is where AI-driven Procurement changes the equation. The goal stays the same: lower cost at acceptable risk. But AI changes the coverage, the speed, and the depth at which Procurement can impact cost downwards.
There is, of course, no single AI fix for Procurement to drive out cost. What works is a set of targeted use cases, each aimed at a different step from demand to payment, all fed by the same data layer. The six below are examples, not a checklist. Most companies start with the one or two that fit their spend profile.
- Seeing the spend. AI classifies every purchase order and invoice line into a single, continuously refreshed taxonomy and flags price gaps on identical items. BCG reports about 90% less manual analysis.[19]
- Knowing what things should cost. Models rebuild a product's cost from its recipe or bill of materials (BOM), commodity indices, energy, and labor. The gap with the supplier's quote becomes the negotiation target. Carrier, for example, reported $21m in annual savings potential from should-cost models.[20]
- Timing commodity buys. Machine learning forecasts prices, maps them to the company's exposure, and recommends coverage within the risk policy.
- Running more tenders and negotiations. Generative AI drafts tenders and normalizes bids; agents negotiate one-to-one with suppliers within buyer-set mandates. BMW reports significantly less tender preparation time.[21]
- Stopping leakage. AI checks every invoice line against the contract before payment, rather than sampling a few afterward. EG Group prevented £20m of overpayments.[2]
- Competing the tail. A conversational front door routes requests, and agents run three-bid quotes on tail spend. Honeywell has negotiated over $500m of spend with AI agents.[1]
Where the value sits depends on the sector. For FMCG, the biggest prize often lies in should-cost on recipes and packaging, and commodity intelligence for dairy, grains, energy, and resins.Retailers are likely to see more value in better supplier negotiations, contract compliance, and tail spend in-store and logistics operations. In contrast, industrial companies often see the greatest benefits in should-costing and specifications, as well as sourcing and negotiation.
Why does AI in Procurement stall?
If the technology works, why isn't everyone using it? To some extent, they are: 49% of Procurement teams have piloted Generative AI… but only 4% run it at scale (The Hackett Group)[22], while Gartner expects over 40% of agentic AI projects to be cancelled by end-2027, mostly on cost and unclear value.[23]
Essentially, there's a wiring problem. The first Procurement agent is often easy…the tenth means point-to-point connections to your ERP, your Procurement suite, your contract system, accounts payable, and market data, each secured for, and monitored separately. An agentic Procurement workforce solves this by running every agent on one platform, built from the same skills and connectors.
From AI-augmented to AI-led: five levels of control
The biggest gains come when buyers direct agents rather than doing all the work themselves. It helps to think of it as a five-step scale:
- Human-led. Buyers do and decide everything, and AI helps with search. They rebuild the spend cube in Excel and negotiate only with the largest suppliers. This is where many teams are today.
- AI-augmented. Buyers drive the process and use AI tools to draft, summarise, and suggest. For example, a buyer asks a chat assistant to summarise a contract or draft a request for quotation (RFQ), then reviews and approves it. This is where pilots often land.
- AI-native. Buyers and AI share the work. Agents classify spend and check every invoice, and buyers review, decide, and handle the exceptions. The most advanced teams operate here.
- AI-orchestrated. AI runs processes and coordinates agents, and people set goals and handle exceptions. Agents run tail-spend negotiations within mandates while category managers set the targets. This is the leading edge today.
- AI-led. AI runs the process end to end, and people focus on governance and high-stakes calls. The CPO governs policy, while the system buys and re-prices catalog and tail items autonomously. That's the next S-curve.
The organization sets the level of control, for example based on the value at stake. Spend classification and invoice matching might run human-on-the-loop: AI runs, people monitor. On the other hand, award recommendations and counter-offers to strategic suppliers might be human-in-the-loop: the buyer approves every output. While selecting a supplier for a critical category or signing a contract stays human-in-command.
How to get started?
We'd start with one category and prove value in about three months. A rapid spend and cost diagnostic builds a classified spend view and sizes the potential AI use cases. Then a sprint to put the first agents into production on one lever and one category, such as should-cost for packaging, with a savings baseline agreed with Finance. After a few months, you add agents lever by lever on the same platform and turn your best buyers' playbooks into shared skills. You don't need perfect data to begin: the diagnostic starts from the purchase-order and invoice lines already in your ERP.
Remember: the real asset isn't the software. Your competitors run broadly the same Procurement suites as you do. Your edge is what only you have: spend data, contracts, supplier history, and the know-how of your best buyers. AI connects that scattered knowledge at a scale no team can match, and turns it into should-cost models, exposure maps, and negotiation playbooks. Your competitors can buy the same software. They can’t buy your intelligence.
