Procurement AI agent · Negotiation
Autonomous Negotiation AI Agent
Exchange offers with a supplier over price, terms and volumes inside a mandate a named holder wrote, stop at its edge, and leave the move that closes the deal to that holder.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Named holder
- Mandate first
- After launch
- Agent Care
What this agent does
Trades offers, never makes the last move
In
- 01
An offer leaves, and it is the company acting: a tool has no authority of its own to exceed.
- 02
An offer goes out unread, and UETA § 14(1) forms the contract anyway, aware or not.
Reason
- 03
A counter arrives, and it is read against the live mandate rather than against the counter before it.
- 04
An offer would cross a limit, and it is stopped before it leaves rather than unwound afterwards.
- 05
An offer for goods is taken up, and UCC § 2-204(1) forms it from conduct alone.
Decide
- 06
An offer sits inside an internal limit, and the supplier who never saw that limit is not bound by it.
- 07
An offer closes on goods, and a merchant confirmation binds unless a buyer objects within ten days.
Out
- 08
An offer is released, and the mandate version, the counter it answered and the binding moment stay on it.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
Offers, counters and the mandate check belong to the agent. The move that closes belongs to a named mandate holder, who releases it and owns it after.
Example workflow
One negotiation, mandate to release
- 1Mandate receivedDelegation of authority, category strategy, prior agreed terms or the approved position
- 2Position assembledThe price band, the term limits, the volume floor and the walk-away, each with its holder
- 3Offer composedThe price, the terms, the volumes offered and the confidence
- 4Controls appliedMandate checks, limit checks, counterparty checks and offer confidence
No human action required
Stages 1 to 4 run unaided, and nothing reaches the supplier at any of them — the agent is composing, and the mandate holder lane opens at the release gate.
Goes to the mandate holder to release.
Adds a category lead read first.
Mandate holder review
The offer is held with the mandate, the limit it sits under and the counter it answers.
Release · Amend · Send to category review
Released — by the named holder
- 6Negotiation and contract records updatedOnly where write access and records policy allow it
- 7Outcome evaluatedMandate adherence, counterparty behaviour, holder amendments and what review found
Amendments
Each holder amendment is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Sending a binding offer to the counterparty.
- Setting the mandate an offer is made under.
- Deciding what the company will concede.
- Signing the agreement an exchange reached.
Automation boundary
Agent acts unaided
- ✓Compose the offer inside the written mandate.
- ✓Score each counter against the live mandate, not the counter before it.
- ✓Stop an offer that would cross a limit before it leaves.
- ✓Stamp the moment an exchange became binding onto the deal record.
Nothing reaches the supplier except by a named mandate holder, inside the agreed boundaries.
- Judging whether a counterparty may be dealt with.
- Telling a supplier where the walk-away sits.
- Choosing which supplier an offer goes to.
- Changes to the mandate, the limits or the rules.
Example output
One offer, annotated
Cal. Civ. Code § 1714.46, in force since 1 January 2026, bars a defendant from asserting that the artificial intelligence autonomously caused the harm; below is one offer exactly as the agent leaves it.
Negotiation output · single offer
Illustrative example- Offer
- Second-source contract manufacturer
- Recorded as
- Counter answered inside the mandate
- Counterparty
- Offer, unreleased
- Mandate of record
- Mandate schedule, 3 August 2026
- Confidence
- Held unreleased
- Held for
- The named mandate holder
As received
Composed from the mandate on file and the counter received, and it asserts nothing beyond them.
What the record holds
- Mandate schedule
- Counter received
- Prior agreed terms
Why no release here
Sending an offer that binds is a move the mandate holder makes.
What the score decides
At the edge of the mandate an offer gets a category read before the holder sees it.
Action
ReleaseAmendSend to category reviewValue
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 offer
From the mandate that permits it
Where the mandate sits
Carrier rate negotiation runs under a DOJ decree and supplier chargebacks are deduction claims; this is a general commercial negotiation, and its subject is the moment a machine bound the company.
No one has to read it
UETA § 14(1) forms a contract by the interaction of electronic agents even if no individual was aware of or reviewed the terms, and 15 U.S.C. 7001(h) backs that federally.
What was checked, and not found
UETA § 10(2) says individual, never person, the word the Act uses for an entity: a human who mis-keys against your agent may unwind it, while a company whose own agent errs has no statutory unwind and falls to the law of mistake, which § 10(4) bars varying by agreement. Checked and not found: no signature, certification, filing, registration or human approval before such a contract binds, no cap on what an electronic agent may agree, nothing in Annex III making this high-risk, no US duty to tell a business it is dealing with a machine, and no US case on it yet.
Build an evidence trail
The offer, the mandate it was made under and the moment it bound stay together.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Sourcing and negotiation
- SAP Ariba
- Coupa Sourcing
- Jaggaer
- Ivalua
- GEP SMART Source
Contract records
- Icertis
- Agiloft
- Sirion
- Signed terms, price schedules
Negotiation channels
- Supplier portals
- email threads
- Where the offers and counters travel
Agent
Autonomous negotiation
- Reads the mandate
- Composes the offer
- Holds for the holder
Spend and price history
- ERP spend history
- price files
- What was last paid and last agreed
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six catches between the model and the supplier
Six catches on one door, the last the stiffest. What opens it is set out in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether an offer may stand
- L3
- Leaves the release to a named holder
- L4 – L5
- Keep the offer and the mandate behind it
- L6
- Stops before any binding move when signals degrade
How Nestack evaluates it
Evaluate the whole exchange — not only the offer that goes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the offer a supplier receives
- E1Final-output evaluationDid the offer stay inside the mandate it was composed under?
- E2Step-level evaluationDid the agent read the live mandate, the right counter and the right prior terms?
- E3Tool evaluationDid it read and write the correct negotiation and the correct term?
- E4Confidence calibrationDo low-confidence offers actually attract more holder amendments?
- E5Slice evaluationHow does performance change across specific counterparties?
- E6Business outcomeHow many offers needed an amendment before the holder released?
Floor — the mandate an offer rests on
Failure modes
Where each failure originates in the agent
Seven failure modes, set at the stage where each one first shows.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- OL-03
Stale mandate read
The mandate read is not the one now in force.
Stage gathers
The mandate, the limits, the counter and the terms
02 · Reasoning
2 modes
- OL-04
Offer composed past the limit
A term is set beyond what the mandate allows.
- OL-06
Retired mandate worked as live
A lapsed limit is treated as the current one.
Stage proposes
The price, the terms, the volumes and the limit
03 · Tool / write
2 modes
- OL-02
Edge offer passed forward
An offer at the edge moves on without the category read.
- OL-05
Offer bound to wrong counterparty
The offer is filed against another supplier.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- OL-01
Bound with no reader
The record shows a binding term and no human read.
Stage returns
The offer the supplier reads and relies on
05 · Change / Version
1 mode
- OL-07
Silent mandate drift
A limit widens while the stored offer keeps the old one.
Stage tracks
Model, prompt, mandate rules and limit dates
- Sev-1 · an offer bound outside the mandate
- Sev-2 · a stale limit reaches the supplier
- Sev-3 · mandate degrades, offer held back
Affected slices
First-time counterparties absorb the amendments
A counterparty-level mandate-adherence figure can read clean while first-time counterparties carry most of the amendments. Nestack reports the amendment rate by counterparty, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| First-time counterparties | 12.5% | 3.7× | Review | |
| Sole-source counterparties | 8.9% | 2.6× | Review | |
| Cross-border counterparties | 5.6% | 1.7× | Watch | |
| Established counterparties | 2.7% | 0.8× | Normal |
Bar: amendment-rate lift vs. established-counterparty baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What an unread offer costs
A loop shuts when the offer that bound without a reader is a regression case. That suite is what the next mandate run is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Amendment rate rises on first-time counterparties.
- 02
Diagnose
The counter that went out at three in the morning and bound the company before anyone read it is worked backwards until one cause is left standing.
- 03
Improve
The change ships numbered, with the offers that caused it attached.
- 04
Verify
Each touched offer case is run again, and one red holds it back.
- 05
Learn
It stays as a standing test, and the mandate rules travel with it.
Learn → Detect
The return edge. The next offer is measured against a suite one case longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, mandate, negotiation workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Mandate capture and delegation-of-authority mapping.Week 1
- 02Negotiation-channel and system assessment.Week 1, Week 2
- 03Counterparty-to-mandate and limit-coverage mapping.Week 1, Week 2
- 04Mandate ingestion and limit normalisation.Week 2
- 05Offer composition and mandate binding.Week 2, Week 3
- 06Edge scoring and review routing.Week 3
- 07Mandate holder release workflow.Week 3, Week 4
- 08Negotiation-channel integration.Week 5
- 09Mandate and formation cases.Week 4, Week 5
- 10Guardrails and binding controls.Week 4
- 11Offer-trail instrumentation.Week 5, Week 6
- 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 tier | PilotOne category, one counterparty | ProductionProduction negotiation workflow | AdvancedMultiple categories / regions |
|---|---|---|---|
| Introduced at Pilot | |||
| Offer composition to your mandate | ✓ | ✓ | ✓ |
| Named holder release | ✓ | ✓ | ✓ |
| Mandate-and-limit baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by counterparty | — | ✓ | ✓ |
| Holder review workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Negotiation-channel integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-currency term structures | — | — | ✓ |
| Cross-category negotiation packs | — | — | ✓ |
| Large counterparty portfolios | — | — | ✓ |
| Multi-state formation controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, negotiation 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.
You bring
We build with it
01Your live mandates and the limits each one carries
→Mandate capture and limit versioning
Week 102Representative past negotiations and the terms they reached
→Mandate binding, offer logic and the negotiation baseline
Week 203Your delegation of authority and the holders it names
→Mandate mapping, limit binding and the automation boundary
Week 104Access to relevant APIs, feeds or exports
→Negotiation-channel and system assessment, then integration setup
Week 205Offers you would not want enforced
→Mandate cases and the evaluation round
Week 406What no mandate may excuse
→Edge scoring, review routing, guardrails and release controls
Week 307A named mandate holder who releases the offer
→Release to the named holder, 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
Six weeks of real work rather than six equal columns; the fifth carries a pair because it must.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Mandate discovery, limit versioning and the automation boundary
W2Channel integration and the mandate-and-limit baseline
W3Offer composition, concession logic and release controls
W4Evaluation suite, mandate cases and failure-mode testing
W5Negotiation-channel integration, pilot offers and targeted corrections
W6One negotiation cycle run under the mandate holder, then Agent Care handover
- Reading the band
- A bar sits on the weeks its own work is named for, and the fifth is genuinely shared.
- At the end of W6
- Once the mandate record validates, Agent Care picks the agent up.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Procurement AI agent
Build a negotiation agent around the mandate your last counter never checked.
Show us one category you negotiate and the mandate behind the last counter that left. UETA § 14(1) binds a company on an exchange no individual read. FAR 1.602-1 will not let an unwarranted human bind the government; nothing stops a company being bound by an unread machine.
- Typical delivery
- 4–6 weeks
- Deployment
- Your live mandates
- Named holder
- Mandate first
- After launch
- Agent Care available
Nestack Agents · Autonomous negotiation
AGT-PR-06 · Agent Care available after launch