RESHUFFLE An interactive companion to the book

Procurement Observatory The workflow

How are agents reorganizing the buying process?

Most companies use AI to improve individual procurement tasks. The larger change begins when agents coordinate work across the buying process, reducing the manual handoffs between requirements, supplier selection, execution and payment.

  1. Need
  2. Define
  3. Find
  4. Compare
  5. Check
  6. Negotiate
  7. Approve
  8. Order
  9. Pay

The important shift is from automating individual tasks to reorganizing how a business need becomes an executed transaction.

The familiar view

AI improves individual procurement tasks.

A request-writing assistant here, a supplier-search tool there, an invoice matcher in finance. Each is bought separately, justified separately and measured on the time it saves in its own step.

What the breadth analysis shows

AI is active across the procurement journey, but unevenly.

We examined 54 offerings across nine parts of procurement. Of 486 offering-stage observations, 184 showed a demonstrated effect.

  1. 54

    offerings

  2. 9

    procurement stages

  3. 486

    offering-stage observations

  4. 184

    demonstrated effects

Where AI is active across the journey

  8. The instruction leaves the buyer

  • Higher activity3 of 9 stages
  • Middle activity5 of 9 stages
  • Lower activity1 of 9 stages

Share of 54 examined offerings demonstrating an effect at each stage. A documented capability, not a deployment.

What the task view misses

Separate task improvements can leave the procurement process disconnected.

A faster request, supplier search or invoice check does not automatically improve the handoffs between them. Information may still have to be transferred, checked and interpreted manually before the next step can begin.

The larger opportunity is to connect the buying process, not only improve its individual tasks.

Where this is already visible

Some systems are beginning to connect previously separate parts of procurement.

A purchase begins with a business need. Company rules determine what is allowed. Suppliers are compared. The selected option becomes an order, contract or payment. The examples below show AI linking parts of that journey which are normally handled through separate systems and manual handoffs.

No single offering in this analysis performs the entire journey. The same pattern appears at several of the points where one part of the journey meets the next.

  1. Where this is already visible

    From a business requirement to supplier comparison.

    Globality turns a requester's business need into a written brief, identifies supplier options and compares their proposals. The buyer makes the final decision.

    What the record shows

    Globality / Glo

    What goes in
    requester project intent
    What the system does
    Proposes a brief a supplier field and a proposal comparison; drafts the SOW after the decision
    What remains manual
    The buyer decides to award; Glo acts only after that decision

    What changedActivities that would normally be completed separately are brought together in one assisted buying process.

    The same link can apply the company's own rules.

  2. Where this is already visible

    From company rules to a purchase.

    Oracle applies configured sourcing rules to eligible requests. It can launch a negotiation, apply the configured award and create purchasing documents. Approval follows the company's configuration.

    What the record shows

    Oracle Fusion Agentic Applications for Procurement

    What goes in
    eligible requisition lines
    What the system does
    Creates and publishes negotiations; applies sole-source or line-level awards; generates purchasing documents
    What remains manual
    Approval is submitted only when approvals are enabled; Oracle ADVISES setting the negotiation owner as first-level approver but does not require it

    What changedThe rules do not have to be checked and reapplied manually at every step.

    Further along the journey, the same pattern reaches invoices.

  3. Where this is already visible

    From invoice evidence to payment processing.

    Vic.ai compares an invoice with the purchase order and receipt. Matching invoices within the company's tolerance settings can be approved and posted for payment. Exceptions are reviewed manually.

    What the record shows

    Vic.ai AP autonomy

    What goes in
    invoice, purchase order and receipt
    What the system does
    Approves and posts an invoice for payment; withholds approval outside tolerance
    What remains manual
    AP review occurs only for out-of-tolerance or low-confidence items - none on the autonomous path

    What changedFinance can focus its attention on exceptions instead of reviewing every invoice in the same way.

    Three examples are three examples. The question is how common this is.

  4. The wider evidence

    Seventeen of the 21 examined examples link more than one part of the buying process.

    Fourteen require the system to interpret the available information and select the next action within defined limits. Four follow fixed instructions, and three could not be classified conclusively.

    14 agentic · 4 conventional automation · 3 unresolved

    These 21 examples examine how procurement activities are being connected. They are separate from the 184 effects used to show where AI is active.

    What that changes operationally is a question about handoffs.

  5. What has changed

    The key difference is whether the handoffs still have to be managed manually.

    The same systems can be set up three ways. What separates them is how much manual work is needed between one step and the next.

    A diagnostic framework informed by the evidence, not a measured distribution of companies

    The opportunity is not to remove every human decision. It is to reduce the manual work required to connect those decisions.

  6. Why it matters competitively

    AI can affect which suppliers are found, qualify and are recommended before the buyer makes the final decision.

    The buyer may still make the final decision. But AI can increasingly influence which suppliers reach that decision.

    What these systems can do

    • LevelpathCan identify qualified suppliers and recommend an option.Record: historical data, contracts and risk profile → generated RFP, qualified suppliers and a recommended option
    • 3D Spark procurement hubCan apply technical and commercial criteria before sending an RFQ.Record: part geometry, cost, lead-time and sustainability data → instant cost analysis and an automatically despatched RFQ
    • LightSourceCan compare supplier information and model an award.Record: unformatted BOMs, drawings and supplier quotes → item master and a modelled optimal award

    A manual role is recorded in the evidence for 16 of the 21 examples. For 4, including these three, the public evidence does not establish where approval sits.

    Strategic implication

    Suppliers may need to compete not only for the buyer's attention, but also to be visible, understandable and eligible to the systems preparing the decision.

    Strategic implication to investigate, not an established market outcome

    The next question is how close AI gets to an order, contract, instruction or payment that becomes official.

Inspect all 21 examined examples

The eight groups the 21 examples fall into, and the records inside each. Three of them are shown above.

  1. DEFINE to ACT6

    • Globality / Glo Agentic coordinationSPECIFY to AUTHORIZE · NR-PRS-001
    • Oracle Fusion Agentic Applications for Procurement Agentic coordinationTRIGGER to EXECUTE · NR-PRS-003
    • Stampli / Billy the Bot Agentic coordinationTRIGGER to AUTHORIZE · NR-PRS-005
    • Precoro UnresolvedTRIGGER to AUTHORIZE · NR-PRS-009
    • Vroozi Conventional automationTRIGGER to EXECUTE · NR-PRS-014
    • Moss Conventional automationTRIGGER to SETTLE · NR-PRS-015

  2. DEFINE to EVALUATE6

    • Icertis Contract Intelligence Agentic coordinationSPECIFY to NEGOTIATE · NR-PRS-006
    • JAGGAER AI (JAI) Agentic coordinationSPECIFY to EVALUATE · NR-PRS-011
    • Levelpath Agentic coordinationSPECIFY to EVALUATE · NR-PRS-013
    • Promena PROMi UnresolvedSPECIFY to DISCOVER · NR-PRS-019
    • 3D Spark procurement hub Agentic coordinationSPECIFY to DISCOVER · NR-PRS-020
    • LightSource Agentic coordinationSPECIFY to EVALUATE · NR-PRS-021

  3. CONSTRAIN to ACT2

    • Vic.ai AP autonomy Agentic coordinationVALIDATE to SETTLE · NR-PRS-002
    • Corcentric AP automation Conventional automationVALIDATE to AUTHORIZE · NR-PRS-018

  4. EVALUATE to ACT2

    • Workday Contract Negotiation Agent Agentic coordinationNEGOTIATE to AUTHORIZE · NR-PRS-004
    • Procbay Bid Analysis Agent Conventional automationEVALUATE to AUTHORIZE · NR-PRS-016

  5. EVALUATE within EVALUATE2

    • Vertice (Ana) Agentic coordinationNEGOTIATE to NEGOTIATE · NR-PRS-007
    • Arkestro predictive procurement Agentic coordinationNEGOTIATE to NEGOTIATE · NR-PRS-012

  6. ACT within ACT1

    • Duvo procurement and category-management agent Agentic coordinationEXECUTE to AUTHORIZE · NR-PRS-010

  7. CONSTRAIN within CONSTRAIN1

    • VERSO Supply Chain Hub UnresolvedVALIDATE to VALIDATE · NR-PRS-017

  8. CONSTRAIN to EVALUATE1

    • Resilinc supply chain risk Agentic coordinationVALIDATE to DISCOVER · NR-PRS-008

How far it reaches

How close does AI get to an order, contract or payment that becomes official?

Some systems only prepare a recommendation. Others complete an approved action. The difference matters, because it decides how much a company is relying on the rules it configured rather than on a check before the fact.

How far the recorded effects reach

  1. Advises

    104 effects

    Shapes or recommends a decision.

  2. Executes

    78 effects

    Acts after or within an approved decision.

  3. Binds the outcome

    2 effects

    Creates an order, contract, instruction, payment or other official consequence.

  8. The instruction leaves the buyer

  • Advises
  • Executes
  • Binds the outcome

Each stage carries its deepest demonstrated effect, not its typical one.

Very few examples reach an outcome that becomes official. Most inform a decision or complete one that a buyer has already approved.

A short check

Where does your procurement process still depend on a manual handoff?

Choose one important process and identify the point where information has to be transferred, checked or interpreted manually before work can continue.

How far does the process get on its own?

Choose an option to see what the configuration is called.

What this means for the operating model

Two automation requirements may reveal one larger coordination problem.

Client reframe · Construction field services

  1. The client requirement

    Automate two existing processes

    A large construction-services company wanted to use AI to respond to field-maintenance enquiries faster and improve invoice reconciliation.

  2. Our reframe

    Treat them as one coordination problem

    Both processes depended on the same information about the requested work, field response, completion evidence, applicable terms and approvals.

    Automating them separately would preserve the gaps between customer service, field operations and finance.

  3. The real opportunity

    Design an AI-native operating model

    Connect the maintenance request, work order, field response, completion evidence and invoice decision.

    This could reduce the need for manually prepared status responses and allow more matching invoices to be processed automatically, while safety decisions and exceptions remain under human control.

The reframe moved the conversation from making two tasks faster to redesigning how service, operations and finance work together.

Connecting the buying process changes more than the handoffs. It can also change who shapes the decision before a person approves it.

Apply the reframe

Design your AI-native operating model

Move beyond automating individual tasks. Redesign how work, information, decisions and human oversight come together across the operating model.

Explore the AI-Native Operating Model

Common Methodology