Main site

Mining AI agent · Commercial & ESG

Commercial, ESG & Trading AI Agent (Cargo to Disclosure)

Assemble the settlement picture behind a cargo, prepare ESG figures on a stated boundary and method, summarise the desk's position — while a commercial manager, a mandated trader and a named officer decide.

Typical delivery
4–6 weeks
Deployment
Your stack
Published figures
Named officer
After launch
Agent Care

What this agent does

Prepares the pack, never commits the company

In

  1. 01

    Take contract terms, cargo records, assay certificates and shipping documents as the commercial file holds them.

  2. 02

    Read emissions, energy, water and payments data from the systems that own it, carrying the boundary and the period.

Reason

  1. 03

    Work each cargo against its own specification — payable elements, penalty and bonus scales, moisture and weighing.

  2. 04

    Compare provisional against final assays, apply the contract's splitting limit and lay out the umpire path it defines.

  3. 05

    Build every ESG figure on a stated boundary, method and period, and keep assured and unassured figures apart.

Decide

  1. 06

    Flag a figure with no basis, a claim with nothing to substantiate it, and exposure sitting outside the desk's mandate.

  2. 07

    Route settlement differences, demurrage evidence gaps and disclosure questions to the people who own them.

Out

  1. 08

    Present the settlement pack, the reporting pack and the position summary, each figure beside the document behind it.

  2. 09

    Retain every source read, the basis stated, the check applied and every amendment a reporting owner made.

Product statement

The agent prepares and flags. Pricing, hedging, buying, selling, nominating and settling stay with the commercial desk; publishing stays with a named officer and an assurance provider.

Example workflow

One cargo, end to end

  1. 1Cargo or period opensA shipment declared, a settlement window reached or a reporting period closing
  2. 2Terms resolvedSpecification, penalty and bonus scales, invoicing terms, and the boundary and method a figure is prepared on
  3. 3Pack assembledAssays, weights, moisture, shipping documents and consumption data, each carried with the source it came from
  4. 4Checks appliedSpecification limits, the splitting limit, provisional-to-final movement, evidence completeness and mandate limits

No human action required

Stages 1 to 4 run without a person in the loop — reading the contract, building the pack and testing the basis finish before anyone is asked to look at anything. No cargo settles and no figure is published in that stretch.

5DecisionSplits on check confidence and whether a figure can state its basis
Inside spec, basis complete

Reaches its owner as a checked pack.

No basis, or outside mandate

Held for the person who owns that call.

Commercial manager and reporting officer

Reads the pack, the documents behind it and the check that stopped it, then decides what is agreed, claimed or published.

Accept findings · Amend · Refer up

Findings accepted — handed back

  1. 6Pack routed to its ownerHanded to the commercial manager, the trader inside mandate or the reporting officer; nothing is settled, signed or published by the agent
  2. 7Outcome evaluatedWhat moved between provisional and final, what a reviewer amended, and what assurance queried, by cohort

Amendments

Figures amended before sign-off are counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Pricing, hedging, buying or selling a cargo.
    • Nominating, declaring or settling a shipment.
    • Agreeing a final assay or a settlement figure.
    • Approving an invoice, a credit or a claim.

  • Automation boundary

    Agent acts unaided

    • ✓Assemble the cargo pack against its own contract terms.
    • ✓Compare provisional against final assays and lay out the umpire path.
    • ✓Prepare each ESG figure with its boundary, method and period.
    • ✓Summarise position and exposure against stated limits.

    Write actions run only inside the approval boundaries agreed during implementation. A settled cargo is never one of them.

    • Signing or publishing a sustainability disclosure.
    • Lodging a payments-to-governments report.
    • Deciding a standard or framework has been met.
  • Changing a boundary, method, mandate or contract term.

Example output

One demurrage claim, annotated

Everything the agent raises is attached to the charter clause and the document it came from.

Evidence pack · one laytime claim

Illustrative example
Claim as raised
Demurrage at the discharge port
Charter terms
Laytime, exceptions, time bar
Time bar
Runs from discharge
Raised
Evidence gap in the pack
Confidence
89%
Claim
Not presented by the agent

As received

The claim as raised, the charter's laytime terms and the clock the time bar runs on — nothing here is re-keyed.

Evidence used

  • Charterparty laytime clause
  • Statement of facts, signed
  • Notice of readiness tendered

Why the pack is short

The clause names a pumping log, and the file does not hold one.

What the score decides

The score sets how closely the owner checks this, never whether the claim is presented.

Action

Accept findingsAmendRefer up

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 cargo and every figure

From contracts, systems and shipping documents

03Preparation

Work to the contract's own words

Use the specification, the penalty and bonus scales, the invoicing terms and the boundary each figure is prepared on.

01Approved path

Be ready before the clock runs out

Assay exchange, final invoicing and demurrage claims all run to contractual clocks, so the documents are gathered while the window is open rather than after it shuts.

02Human review

Show the basis, or show the gap

A figure whose boundary, method or period cannot be stated is raised as a gap, instead of travelling quietly into a pack that someone is asked to sign.

04

Build an evidence trail

Retain each document read, the basis stated, the check applied, the confidence and every amendment made before anything was agreed or published — on both paths.

Integrations

Typical integrations

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

  • Marketing & contracts

    • Sales contracts
    • pricing terms
    • CTRM/ETRM systems
    • settlement files

  • Assay & cargo evidence

    • Assay certificates
    • umpire assays
    • Draft surveys
    • moisture reports

  • ESG & emissions data

    • Fuel and energy data
    • GHG inventory
    • Site EHS systems
    • water and tailings

Agent

Commercial, ESG & trading

  • Reads the contract
  • Prepares the pack
  • Routes what has no basis

  • Finance & disclosure

    • SAP
    • Pronto
    • invoicing
    • Reporting platforms
    • assurance files

  • 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 market

Each control wraps the one inside it. A pack clears every layer before an owner reads it, and settling, trading and publishing sit outside all six.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeReturn preparation to the commercial and reporting teams if evaluations or signals degrade.Roll back
L5Change controlRecord contract amendments, boundary and method changes, the approver and every correction.Record
L4Mandate and sign-offSettling, invoicing, trading and publishing stay with the named person who owns them.Gate
L3Substantiation testClaims are tested against the evidence held; what fails is held, not carried.Hold
L2Basis stampingEach figure is stamped with the boundary, method, period and assurance status it was built on; one that cannot be stamped is raised, not totalled.Stamp
L1Contract as sourceEach cargo term is scored against the executed contract, not against the last shipment.Cite
Model corePack prepared — cargo checks, ESG figures with their basis, position summary and confidence
L1 – L2
Decide whether a figure may stand
L3
Decides what is held for a person
L4 – L5
Keep the sign-off and the record intact
L6
Pulls automation back when signals degrade

How Nestack evaluates it

Evaluate the whole pack — not only the figure at the top.

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

Surface — the pack a commercial owner opens

  1. E1Final-output evaluationDid the pack carry what the contract actually requires?
  2. E2Step-level evaluationDid it read the right clause, boundary, method and period?
  3. E3Tool evaluationDid it resolve the correct cargo, entity and reporting period?
  4. E4Basis and substantiationCan every figure and claim state what it rests on?
  5. E5Slice evaluationHow does pack quality change across specific cohorts?
  6. E6Business outcomeWhat was amended before sign-off, and what did assurance query?

Floor — the settlement and the disclosure it feeds

Failure modes

Where each failure originates in the agent

Seven failure modes plotted against the five stages of the agent lifecycle.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • CE-01

      Provisional read as final

      A provisional assay is carried as though it had settled.

    Stage gathers

    Contract terms, assays, shipping papers and source data

  2. 02 · Preparation

    2 modes

    • CE-02

      Boundary moves under the figure

      A divested site leaves the inventory with no restatement.

    • CE-03

      Assured and unassured mixed

      An unassured figure is added into an assured total.

    Stage proposes

    Cargo checks, ESG figures and the basis of each

  3. 03 · Verification

    2 modes

    • CE-04

      Claim with nothing behind it

      A low-carbon line rests on evidence nobody actually holds.

    • CE-05

      Exposure read short of the mandate

      Offsetting positions net away a limit already passed.

    Stage checks

    Spec limits, substantiation and the desk's own limits

  4. 04 · Handoff / write

    1 mode

    • CE-06

      Time-barred claim evidence

      The demurrage pack misses a document the clause names.

    Stage presents

    The pack a manager, trader or officer reads and acts on

  5. 05 · Change / Version

    1 mode

    • CE-07

      Restatement arrives unmarked

      A recalculated prior period lands with no version note.

    Stage tracks

    Contract, boundary, method, model and prompt changes

  • Sev-1 · a figure could be published or traded on
  • Sev-2 · a wrong figure reaches the pack
  • Sev-3 · preparation degrades, more is amended
See the complete Mining failure-mode catalogue

Affected slices

The first of anything is where this breaks

Nothing here fails on the tenth cargo under a term contract. It fails on the first with a new counterparty, on figures from somebody else's system, and on a claim with a time bar running. Nestack reports performance by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
First cargo, new counterparty3.9%2.6×Review
Scope 3 and value-chain figures3.3%2.2×Review
Demurrage and laytime claims2.6%1.7×Watch
Repeat term-contract shipments1.2%0.8×Normal

Bar: amended-before-sign-off item rate, lift vs. term-shipment baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

An assurance query is a basis the agent never had

Amending a figure before it is published leaves the reason it was wrong in place. What changes is the source, the basis, or the test that passed it.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Amendments before sign-off or assurance queries rise on one cohort.

  2. 02

    Diagnose

    Work back from the assurance query to the clause read, and then to the basis nobody could state.

  3. 03

    Improve

    The source, the basis or the test is changed under change control, with an approver.

  4. 04

    Verify

    Re-prepared over cargoes already settled and periods already assured.

  5. 05

    Learn

    The corrected basis is written into what every later figure must state.

  6. Learn → Detect

    The return edge. Nothing shipped here changes who agrees a settlement or who signs a disclosure.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, contracts and source data, pack preparation, evaluation, delivery, then production validation and handover.

Workstream

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

  1. 01Workflow discovery and automation-boundary definition.Week 1
  2. 02Contract, spec and penalty-scale mapping.Week 1, Week 2
  3. 03Reporting boundary and method mapping.Week 1, Week 2
  4. 04Cargo, assay and shipping-document access.Week 2
  5. 05Provisional-to-final and umpire-path checks.Week 2, Week 3
  6. 06Emissions, energy and payments ingestion.Week 3
  7. 07Substantiation and assurance-status tests.Week 3, Week 4
  8. 08Position and mandate-limit summarisation.Week 4
  9. 09Evaluation suite, basis recall and regression cases.Week 4, Week 5
  10. 10Laytime and demurrage evidence assembly.Week 5
  11. 11Settlement and reporting pack delivery.Week 5, Week 6
  12. 12Observability, deployment 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. No tier lets the agent settle a cargo or publish a figure.

Capability✓ in scope · — not at this tierPilotOne contract, one reporting setProductionProduction system integrationAdvancedMulti-commodity / multi-entity
Introduced at Pilot
Cargo pack against contract terms✓✓✓
Provisional, final and umpire-path checks✓✓✓
ESG figures with boundary and method✓✓✓
Laytime and demurrage evidence packs✓✓✓
Position summary against stated limits✓✓✓
Settlement and sign-off stay with a person✓✓✓
Baseline evaluation✓✓✓
Introduced at Production
Reviewer workflow and pack delivery—✓✓
Observability and evaluation—✓✓
Introduced at Advanced
Multi-commodity and multi-entity reporting——✓
Group disclosure sets and enterprise controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on the contracts and commodities in scope, the reporting frameworks and boundaries involved, source-system integrations, review 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. 01The contracts in scope, with their specification and penalty terms

    Contract, spec and penalty-scale mapping

    Week 1

  2. 02Who agrees a settlement, who trades in mandate and who signs a disclosure

    Workflow discovery and automation-boundary definition

    Week 1

  3. 03Your reporting boundary, the methods used and the periods reported

    Reporting boundary and method mapping

    Week 2

  4. 04Access to cargo records, assay certificates and shipping documents

    Cargo, assay and shipping-document access

    Week 2

  5. 05Your emissions, energy and payments-to-governments source data

    Emissions, energy and payments ingestion

    Week 3

  6. 06Cargoes that moved a long way between provisional and final

    Evaluation suite, basis recall and regression cases

    Week 4

  7. 07Named commercial and reporting owners to read the packs

    Reviewer workflow, then pilot packs 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

Phases are drawn over the weeks they actually occupy. Nothing is prepared for a live cargo until the basis tests have run in week 4.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Contracts in scope, reporting boundary and the automation boundary

W2Cargo records, assay certificates and shipping documents wired in

W3Emissions and payments data, and the provisional-to-final checks

W4Substantiation and basis tests, mandate limits and the evaluation suite

W5Pack delivery, supervised cargoes and targeted corrections

W6Owners sign from packs prepared beside the existing process, then handover

Reading the band
Week 4 is where a figure has to say what it rests on; week 5 is where an owner first sees one. The bars overlap because the second does not wait for the first to close.
At the end of W6
Cargoes already settled and periods already assured have been rebuilt and compared with what was signed, then Agent Care takes over monitoring.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Mining AI agent

Build a commercial and ESG agent around your contracts.

Show us one sales contract, a cargo you have already settled and the last ESG figure someone asked you to explain. We'll rebuild that cargo's pack from your own documents, and the figures that cannot say what they rest on come back marked as such.

Typical delivery
4–6 weeks
Deployment
Your commercial and reporting stack
Published figures
Named officer, with assurance
After launch
Agent Care available

Nestack Agents · Commercial, ESG & trading

AGT-MN-08 · Agent Care available after launch

Build a commercial and ESG agent

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

About you
Your setup

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