Main site

Construction AI agent · Procurement

Procurement & Supply-Chain AI Agent

Substituting a product is a design act, not a purchasing one. The agent matches deliveries, screens registers and assembles conformity evidence; the consultant evaluates and the owner consents.

Typical delivery
4–6 weeks
Deployment
Your stack
Buyer approves
Evidence only
After launch
Agent Care

What this agent does

Assembles the evidence; it never approves the swap

In

  1. 01

    Match each requisition line to the specified product, its declared use and the approved list.

  2. 02

    Match every product to its conformity artefact — declaration of performance, listing or test report.

Reason

  1. 03

    Screen nominated products against the product-use bans in force in that state, and date the check.

  2. 04

    Screen suppliers against sanctions, licence and modern-slavery registers, and report the exceptions.

  3. 05

    Monitor supplier lead times and raise an exception well before a date threatens the programme.

Decide

  1. 06

    Monitor delivery dates against the programme, and hand the window itself to the person holding the duty.

  2. 07

    Match received deliveries against PO lines and flag quantity and documentation discrepancies.

Out

  1. 08

    Retain the evidence pack, the batch records and the buyer's decision against the line item.

  2. 09

    Execute write actions only inside the approval boundaries agreed during implementation.

Product statement

The agent assembles evidence; the design consultant evaluates a substitution and the owner consents — and once the product is installed, the remedy is a rectification order, not a return.

Example workflow

One line item, requisition to release

  1. 1Requisition receivedSpecification, submittal register, approved product list or site request
  2. 2Evidence assembledThe specified product, its declared characteristics, the artefacts on file and the date each was issued
  3. 3Exceptions listedMissing artefacts, lead-time risk, register hits and confidence
  4. 4Controls appliedApproved-list and delegation checks, ban-register screening, substitution tripwires and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing is bought or approved at any of them — the agent is assembling evidence, and the buyer's lane opens at the confidence gate.

5DecisionBranches at the confidence threshold
High confidence

Goes to the buyer holding the delegation.

Low confidence

Adds a design-consultant read first.

Buyer approval

The line is held with its evidence pack, its exceptions and the confidence.

Release · Amend · Send to the consultant

Approved — released to purchasing

  1. 6Procurement systems updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedEvidence completeness, register hits, substitution referrals and non-conformities found later

Amendments

Every amendment made in review is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Approving a product substitution or an “or equal”.
    • Determining that a product conforms or is fit for its use.
    • Releasing a PO outside delegated authority or the approved list.
    • Setting or compressing a delivery window or driver schedule.

  • Automation boundary

    Agent acts unaided

    • ✓Match received deliveries against PO lines and flag discrepancies.
    • ✓Monitor lead times and raise an exception.
    • ✓Assemble the conformity evidence pack.
    • ✓Screen suppliers and products against sanctions, licence and ban registers.

    Any write happens inside the boundaries agreed at implementation, and never onto a submittal or a schedule.

    • Approving or closing out a submittal.
    • Certifying domestic content or applying for a waiver.
    • Onboarding a supplier or waiving a screening failure.
  • Drafting any part of a modern slavery statement.

Example output

One line item, annotated

Everything the agent assembles is attached to the line item it was gathered for.

Procurement output · single line item

Illustrative example
Package
Façade cladding
Specified product
Insulated panel to the specified core and fire classification
Lead time
14 weeks
Conformity artefact
Declaration of performance on file
Confidence
91%
Register check
Ban registers clear at the check date

As received

Taken from the specification and the artefacts on file — nothing on this side is written by the agent.

What was read

  • Specification clause
  • Declaration of performance
  • Batch and consignment record

Why this is an exception

Only a consultant may answer a lead-time problem with a different product.

What the score decides

Below the threshold the line picks up a consultant read before the buyer.

Action

ReleaseAmendSend to the consultant

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value lands

Value 01 – 04

Every line item

From the specification and the PO

03Evidence

Check against this project's own list

Approved products, delegation limits and use bans come from the project and the jurisdiction it sits in, not a global default.

01Approved path

Buy what was specified

Routine matching, screening and evidence assembly happen before the buyer opens the line.

02Human review

Send review to the substitution pressure

Lead-time exceptions and missing artefacts are marked, so the buyer's read starts where substitution starts.

04

Build an evidence trail

The evidence pack travels with the line item, not in a side folder.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

  • Construction ERP and procurement

    • Trimble Viewpoint Vista
    • RIB Jobpac Connect

  • Project management and POs

    • Procore
    • Autodesk Construction Cloud

  • Submittals and document control

    • Oracle Aconex
    • Newforma
    • Submittal Exchange

Agent

Procurement & supply chain

  • Reads the specification
  • Assembles the evidence
  • Holds for the buyer

  • Sourcing and screening

    • Coupa
    • SAP Ariba
    • Certification and ban registers

  • Observability & evaluation

    • OpenTelemetry
    • Langfuse
    • Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six layers between the model and the order

Each control wraps the one inside it. What a layer does not catch is named in the failure-mode map below it.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modePull the agent back to evidence assembly only when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, catalogue and approved-list changes.Track
L4TraceabilityRecord the artefacts read, their issue dates, the exceptions raised and the buyer's decision.Record
L3Buyer approvalHold lines for the buyer holding the delegation; it governs release, not whether the product conforms.Gate
L2Policy guardrailsTest lines against the approved list, delegation limits and ban registers; a hit returns the line.Restrict
L1Confidence thresholdsRoute low-confidence lines to a design-consultant read before the buyer sees them.Require review
Model coreLine prepared — specified product, evidence pack, exceptions and confidence
L1 – L2
Test whether a line may stand
L3
Puts the release in the buyer's hands
L4 – L5
Keep the evidence pack and its provenance intact
L6
Reverts to exception reporting when signals slip

How Nestack evaluates it

Evaluate the procurement workflow — not only the delivered product.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the product that reaches the wall

  1. E1Final-output evaluationDid the evidence pack cover the declared characteristic that was specified?
  2. E2Step-level evaluationDid the agent read the current datasheet, declaration and use-ban register?
  3. E3Tool evaluationDid it read and write the correct PO and the correct line?
  4. E4Confidence calibrationDo low-confidence lines actually attract more consultant referrals?
  5. E5Slice evaluationHow does performance change across specific product classes?
  6. E6Business outcomeHow many lines needed rework, or failed verification once installed?

Floor — the product the building is left with

Failure modes

Where each failure originates in the agent

Seven failure modes, set against the stage that produces each one.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • PU-01

      Expired evidence read as current

      A superseded datasheet or a lapsed declaration stands in for live evidence.

    Stage gathers

    Specifications, declarations and use-ban registers

  2. 02 · Reasoning

    2 modes

    • PU-02

      Equivalence inferred

      Headline specs match while the declared characteristic that mattered does not.

    • PU-03

      Lead time solved by swapping

      A logistics constraint is answered with an alternative product, silently.

    Stage proposes

    Specified product, evidence pack and confidence

  3. 03 · Tool / write

    2 modes

    • PU-04

      PO released before evaluation

      The substituted product is ordered before any consultant has evaluated it.

    • PU-05

      Delivery window compressed

      In Australia that is a scheduler act, taken with no human in the loop.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • PU-06

      Self-declaration read as tested

      A supplier's own claim is presented in the format of a verified test report.

    Stage returns

    The evidence pack the buyer relies on

  5. 05 · Change / Version

    1 mode

    • PU-07

      Catalogue remap on open orders

      A successor code with different declared performance is inherited by open POs.

    Stage tracks

    Model, prompt, catalogue and approved-list config

  • Sev-1 · a held act performed by the agent
  • Sev-2 · a non-conforming product reaches the site
  • Sev-3 · evidence degrades, line routes to review
See the complete Construction failure-mode catalogue

Affected slices

One package type can own the whole rate

Post-installation non-conformity is split by package, and the exposed cohorts announce themselves — façade and passive fire first, because rectification there dwarfs the product.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Façade and passive fire6.5%3.6×Review
US federally assisted work4.5%2.5×Review
Long-lead structural and MEP2.7%1.5×Watch
Commodity repeat-buy items0.9%0.5×Normal

Bar: non-conformity-rate lift vs. an all-line-item baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

Closing a cycle means adding a case

Nothing is closed by an account of what went wrong. The cycle ends when the failure is a regression case.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Non-conformity rate rises in a product class.

  2. 02

    Diagnose

    The buyer and the design consultant read the evidence pack and the order history until the cause is one of them.

  3. 03

    Improve

    Buyer and consultant agree the change, and it ships against a version.

  4. 04

    Verify

    A failing case stops the release, not a reviewer's judgement.

  5. 05

    Learn

    The case sticks, and the approved-product list moves with it.

  6. Learn → Detect

    The return edge. The suite the next detection runs against is longer by one.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, procurement workflow, evaluation, integration, then production validation and handover.

Workstream

Week 1Week 2Week 3Week 4Week 5Week 6

  1. 01Procurement workflow discovery and boundary work.Week 1
  2. 02ERP, PO and submittal source assessment.Week 1, Week 2
  3. 03Approved-list, delegation and register mapping.Week 1, Week 2
  4. 04Requisition ingestion and normalisation.Week 2
  5. 05Evidence-pack logic and exception rules.Week 2, Week 3
  6. 06Confidence scoring and substitution routing.Week 3
  7. 07Buyer approval workflow.Week 3, Week 4
  8. 08ERP and procurement-system integration.Week 5
  9. 09Conformity regression cases.Week 4, Week 5
  10. 10Guardrails and delegation controls.Week 4
  11. 11Evidence-pack instrumentation.Week 5, Week 6
  12. 12Deployment, documentation and Agent Care handover.Week 6

12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel

Final scope and sequence confirmed in discovery

Engagement tiers

What each tier includes

Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.

Capability✓ in scope · — not at this tierPilotOne project, one approved listProductionProduction procurement systemsAdvancedMultiple projects / regions
Introduced at Pilot
Evidence packs against your spec✓✓✓
Buyer approval✓✓✓
Conformity baseline✓✓✓
Introduced at Production
Reporting by product class—✓✓
Review workflow in your systems—✓✓
Approved write-back—✓✓
ERP and finance integration—✓✓
Introduced at Advanced
Multi-jurisdiction product rules——✓
Multi-stage buyer approvals——✓
High order volume——✓
Enterprise supplier controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on integrations, workflow complexity, order volume, approval controls and deployment requirements.
Separate from build
Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.

What we need from you

What you bring, and what we build with it

Each input maps to a piece of build scope and a week in the delivery timeline.

  1. 01Your approved product and supplier lists

    Approved-list and delegation mapping

    Week 1

  2. 02Representative specifications and POs

    Evidence-pack baseline, matching and screening rules

    Week 2

  3. 03Your delegation limits and substitution protocol

    Substitution-protocol mapping and automation-boundary definition

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    ERP, PO and submittal system assessment, then integration setup

    Week 2

  5. 05Substitutions that should not have been approved

    Conformity cases and failure-mode testing

    Week 4

  6. 06Your delegation limits and approved-product list

    Confidence scoring, exception routing, guardrails and approval controls

    Week 3

  7. 07Named buyers and the consultants they refer to

    Buyer approval workflow, then pilot and production validation

    Weeks 5–6

Nothing else is required

Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

Every phase sits across the weeks it truly occupies. Week 5 carries evaluation and pilot at once.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Procurement workflow discovery, list mapping and the automation boundary

W2Source integration and the evidence-pack baseline

W3Exception rules, confidence logic and approval controls

W4Evaluation suite, guardrails and failure-mode testing

W5ERP integration, pilot packages and targeted corrections

W6A live package procured under the buyer's authority, then handover

Reading the band
Each bar covers only the weeks its work is named in. The week 5 overlap is real, not padding.
At the end of W6
Validation complete; Agent Care holds the monitoring from week seven.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Construction AI agent

Build a procurement agent around your substitution protocol.

Show us your specifications, your approved lists and who consents to a substitution. Once a non-conforming product is installed the remedy is rectification, and on US federally assisted work the domestic-content position cannot be cured at all.

Typical delivery
4–6 weeks
Deployment
Your procurement stack
Buyer approves
Evidence only
After launch
Agent Care available

Nestack Agents · Procurement and supply chain

AGT-CN-06 · Agent Care available after launch

Build a procurement agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your projects

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.