Manufacturing AI agent · CBAM data
CBAM & Value-Chain Emissions-Data AI Agent
Collect the installation and embedded-emissions data your EU importer's declaration rests on, chase the returns, check each method against the last cycle, and hold the pack for your sustainability lead.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Lead approval
- Pre-declaration
- After launch
- Agent Care
What this agent does
Assembles the pack, not the declaration
In
- 01
A supplier is asked for installation-level data on the template set by Implementing Regulation (EU) 2025/2547.
- 02
A boundary is set against the monitoring plan that regulation requires, and carried into the next cycle.
Reason
- 03
A return arrives, and the method, the vintage and the verification status are recorded beside the figure.
- 04
A plant-wide average comes back, and it is held as unusable for actual values rather than passed on.
- 05
A small supplier declines, and the chase stops: Directive (EU) 2026/470 gives it that right.
Decide
- 06
A reporting year opens, and the 30 September 2027 declaration covering calendar year 2026 is worked back.
- 07
A tonne ships, and the importer's cumulative net mass across all CN codes is tracked, not the shipment.
Out
- 08
A pack is assembled, and the returns, the gaps and the method notes are held for the sustainability lead.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent assembles the pack; your importer's authorised CBAM declarant files it, and an accredited verifier verifies actual emissions.
Example workflow
One reporting year, return to pack
- 1Installation record receivedMonitoring plan, metered energy data, process records or a supplier return
- 2Returns assembledActivity data, precursor quantities, factors and method notes, each with its named source
- 3Figures compiledEmbedded emissions per good, the boundary, the method and confidence
- 4Controls appliedBoundary checks, method-consistency checks, vintage and verification status and confidence threshold
No human action required
Stages 1 to 4 run unaided, and nothing is declared at any of them — the agent is assembling, and the lead's lane opens at the confidence gate.
Goes to the sustainability lead.
Adds a carbon-accounting read first.
Lead approval
The pack is held with its returns, its open gaps and the confidence.
Approve · Amend · Send to technical review
Approved — released to the declarant
- 6Reporting systems updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedAmendment rate, verifier findings, return completeness and corrections after the pack has gone
Amendments
Amendments made after the pack went out are counted too.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Submitting the CBAM declaration or surrendering certificates.
- Verifying embedded emissions as an accredited verifier.
- Signing a climate disclosure for the reporting entity.
- Deciding the monitoring method the installation will run.
Automation boundary
Agent acts unaided
- ✓Collect installation and embedded-emissions returns.
- ✓Carry each figure with its method, vintage.
- ✓Check boundaries and methods against the monitoring plan on file.
- ✓Flag the gaps the lead should weigh, and hold the pack for them.
Any write happens inside the boundaries agreed at implementation, never ahead of the declarant.
- Certifying a carbon price paid in a third country.
- Agreeing what a supply contract obliges a supplier to send.
- Registering an installation in the CBAM registry.
- Changes to boundary, method or approval rules.
Example output
One supplier return, annotated
Everything the agent assembles stays attached to the return it came from, correction or not.
Emissions-data output · single good
Illustrative example- Installation
- Hot-rolled coil mill
- Good returned
- Hot-rolled coil, one production route, one reporting year
- Embedded emissions
- 1.83 tCO2e/t
- Basis of figure
- Installation-specific
- Confidence
- 89%
- Verification status
- Unverified — awaits a verifier
As received
Taken from the supplier's return and the monitoring plan — nothing on this side is inferred.
Returns used
- Monitoring plan on file
- Metered energy data
- Precursor declaration
Why this status
It is installation-specific but not yet verified — the line the lead weighs.
What the score decides
Below the configured threshold the pack picks up a technical read before it reaches the lead.
Action
ApproveAmendSend to technical 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 supplier
From the monitoring plan
Ask against your boundary
Draw on the returns on file and the method the monitoring plan fixes.
Your importer files, not you
Routine returns arrive matched to an installation, a good and the importer that will declare it.
Send review to the doubtful figures
Boundary mismatches and low-confidence returns are marked, so the lead's read starts where risk concentrates.
Build an evidence trail
The figure, the boundary it was drawn on and the supplier who returned it stay on the pack.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
ERP and production data
- SAP
- Oracle
- Dynamics
- Infor
- QAD
Energy and metering
- Utility bills
- submeters
- Historian
- SCADA exports
Carbon accounting
- Sphera
- Watershed
- Persefoni
- SAP SCT
Agent
CBAM and value-chain data
- Reads the returns
- Checks the boundary
- Holds for approval
Supplier documents
- Emissions returns
- templates
- Precursor declarations
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 return
Six wraps between the model and the pack. What none of them catches is named in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a figure may stand
- L3
- Puts the release in the lead's hands
- L4 – L5
- Keep the figure and the boundary behind it
- L6
- Holds the pack at incomplete when signals degrade
How Nestack evaluates it
Evaluate the collection workflow — not only the figure at the end.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the figure the importer declares
- E1Final-output evaluationDid the figure match the installation and the good it was returned for?
- E2Step-level evaluationDid the agent use the right monitoring plan, template and method?
- E3Tool evaluationDid it read and write the correct installation and the correct field?
- E4Confidence calibrationDo low-confidence returns actually attract more amendments?
- E5Slice evaluationHow does performance change across specific supplier groups?
- E6Business outcomeHow many packs were short of a return at the 10 November 2026 CARB deadline?
Floor — the exposure the importer answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, placed at the stage each one originates.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- XB-03
Corporate average returned
A plant-wide intensity read as an installation figure.
Stage gathers
Monitoring plan, energy data, returns and templates
02 · Reasoning
2 modes
- XB-04
Boundary mismatch
A return is matched to a good defined at another CN code.
- XB-06
Method drift unread
A changed factor source reads as a reduction.
Stage proposes
Figure, boundary, method, vintage and confidence
03 · Tool / write
2 modes
- XB-02
Unverified used as actual
Pack released while the figure is unverified.
- XB-05
Threshold read per shipment
De minimis tracked per consignment, not per year.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- XB-01
Vintage and status missing
A figure carries no method, vintage or verification status.
Stage returns
The pack the lead approves and the importer declares
05 · Change / Version
1 mode
- XB-07
Silent method regression
A rule change widens what the agent will accept.
Stage tracks
Model, prompt, boundary rules and method config
- Sev-1 · acts outside the reporting boundary
- Sev-2 · a wrong figure reaches the pack
- Sev-3 · return degrades, pack routes to review
Affected slices
One installation can carry most of the gap
A group-level return rate can look complete while a few installations account for most of the amendments. Nestack reports the amendment rate by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| First-cycle steel suppliers | 9.1% | 3.7× | Review | |
| Scrap-bearing aluminium routes | 7.3% | 3.0× | Review | |
| Tier 2 below the value-chain cap | 4.1% | 1.7× | Watch | |
| Established installations | 2.2% | 0.9× | Normal |
Bar: amendment-rate lift vs. established-installation baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
The loop closes on the boundary case
A cycle closes when the inconsistent boundary is a case the next release has to catch. That suite is what the next pack returned is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Amendment rate rises in a supplier slice.
- 02
Diagnose
The mill that sent a number with no boundary attached is read back until one cause remains.
- 03
Improve
The correction ships numbered, with the supplier returns that prompted it attached.
- 04
Verify
Nothing ships while a touched boundary case is failing; it runs until it clears.
- 05
Learn
It is kept as a standing case, and the method rules are revised alongside.
Learn → Detect
The return edge. The next pack is assembled 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, sources, collection workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Reporting discovery and boundary set and boundary definition.Week 1
- 02Installation and supplier assessment.Week 1, Week 2
- 03Boundary, method and template mapping and rule mapping.Week 1, Week 2
- 04Return ingestion and normalisation.Week 2
- 05Collection logic and evidence binding.Week 2, Week 3
- 06Confidence scoring and gap routing.Week 3
- 07Sustainability lead approval.Week 3, Week 4
- 08Reporting-system integration.Week 5
- 09Boundary and method cases.Week 4, Week 5
- 10Guardrails and return controls.Week 4
- 11Pack-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 installation, one good | ProductionProduction reporting systems | AdvancedMultiple sites / schemes |
|---|---|---|---|
| Introduced at Pilot | |||
| Collection to your boundary and rules | ✓ | ✓ | ✓ |
| Lead approval | ✓ | ✓ | ✓ |
| Supplier-return baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by installation | — | ✓ | ✓ |
| Approval workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Energy-and-process data integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multiple schemes and boundaries | — | — | ✓ |
| Multi-stage reporting approvals | — | — | ✓ |
| High supplier volume | — | — | ✓ |
| Multi-scheme reporting controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, transaction 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 installation list and product routes
→Return ingestion and boundary mapping
Week 102Representative supplier returns
→Collection baseline and evidence binding
Week 203Your monitoring plan and method rules
→Boundary, method and template mapping
Week 104Access to relevant APIs, feeds or exports
→Installation and supplier assessment, then integration setup
Week 205Figures you would not want verified
→Verification cases and the evaluation suite
Week 406What no emissions figure may hide
→Confidence scoring, gap routing, guardrails and return controls
Week 307A named sustainability lead to approve
→Lead 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
The bands follow the data-collection season, which is why the fifth week carries both.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Reporting discovery, method mapping and the automation boundary
W2Installation integration and the collection baseline
W3Collection workflow, confidence logic and approval controls
W4Evaluation suite, boundary and method checks and failure-mode testing
W5Supplier integration, pilot returns and targeted corrections
W6One reporting year assembled under the sustainability lead, then handover
- Reading the band
- Each band covers only the weeks its work is named in. The week 5 overlap is real, not padding.
- At the end of W6
- Validation closes on live packs and Agent Care picks up monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Manufacturing AI agent
Build an emissions-data agent around the pack your importer declares from.
Show us your installations, your EU customers and who signs the pack. The declarant who files is your importer, established in the Union; what you owe is contractual, owed to them.
- Typical delivery
- 4–6 weeks
- Deployment
- Your ERP and meters
- Lead approval
- Pre-declaration
- After launch
- Agent Care available
Nestack Agents · CBAM & value-chain data
AGT-MFG-22 · Agent Care available after launch