Main site

Retail AI agent · Shipping exceptions

Shipping & Route-Exception AI Agent

Detect and classify carrier exceptions, draft the delay notice with the consent basis the rule actually requires, and leave the reroute, the claim and the send to the people who own them.

Typical delivery
4–6 weeks
Deployment
Your stack
Manager release
Draft-only
After launch
Agent Care

What this agent does

Works the exception, does not decide it

In

  1. 01

    Ingest tracking events, tender records, freight invoices and lane schedules from supported carrier.

  2. 02

    Normalise carrier milestone codes into one event vocabulary, and keep the raw code beside each one.

Reason

  1. 03

    Detect and classify exceptions against the thresholds configured for that lane and that service.

  2. 04

    Read the order's notice history, and treat a renewed delay differently from the first one.

  3. 05

    Draft the delay notice and the claim packet, each bound to the shipment record behind it.

Decide

  1. 06

    Mark shipments whose claim deadline, hazmat status or temperature setpoint puts the decision beyond the agent.

  2. 07

    Route every notice to the transport manager and every claim position to the claims manager.

Out

  1. 08

    Retain the milestone data, the notice sent, its consent basis and the person who released it.

  2. 09

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

Product statement

The agent works the exception; the transport manager authorises the reroute, the claims manager takes the claim position, and the retailer answers for the notice.

Example workflow

One shipment, tender to resolution

  1. 1Milestone event receivedCarrier tracking event, tender record, freight invoice or a lane-schedule change
  2. 2Milestones assembledOrder and notice history, service commitment, tariff deadlines and the commodity's handling profile
  3. 3Exception classifiedException type, drafted notice, claim exposure and confidence
  4. 4Thresholds appliedNotice-sequence checks, hazmat and setpoint checks, deadline checks and the confidence threshold

No human action required

Stages one to four run unaided, and nothing reaches a buyer at any of them — the agent is drafting, and the manager's lane opens at the confidence gate.

5DecisionSplits at the notice gate
On the expected path

Goes to the transport manager to release.

Off the expected path

Adds a claims-manager read first.

Manager release

The notice is held with its milestone data, its consent basis and the confidence.

Release · Amend · Send to claims review

Released — notice sent

  1. 6Carrier and order systems updatedOnly where write access and approval policy allow it
  2. 7Exception evaluatedNotice timing, consent basis, claim outcomes and refunds owed after the fact

Reversals

Every manager amendment is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Authorising a reroute, re-consignment or mode change.
    • Any hazmat routing, repacking or re-labelling decision.
    • Filing, conceding, denying or settling a freight claim.
    • Changing classification, valuation or country of origin.

  • Automation boundary

    Agent acts unaided

    • ✓Poll carrier tracking and normalise milestone events across lanes.
    • ✓Detect and classify exceptions against configured.
    • ✓Draft delay notices and claim packets for human.
    • ✓Reconcile freight invoices and flag every accessorial variance.

    Any write happens inside the boundaries agreed at implementation, never ahead of the manager's release.

    • Dispositioning a temperature-excursed food shipment.
    • Approving or waiving a demurrage or detention charge.
    • Approving the delay-notice template and its consent logic.
  • Directing a gig delivery driver's method or sequence.

Example output

One delay notice, annotated

Everything the agent drafts is attached to the shipment it was drawn from.

Exception output · single shipment

Illustrative example
Shipment
Cross-border parcel
Drafted notice
Revised delivery date, with the cancel-and-refund option stated in the same message
Notice state
Renewed notice
Source of record
Carrier milestone and order record
Confidence
81%
Where it went
Transport manager

As received

Taken from the carrier's milestone feed and the order's own notice history.

Evidence used

  • Carrier milestone
  • Prior notice sent
  • Service commitment

Why this notice

It is the second notice, so silence is not consent and a person still decides.

What the score decides

Below the configured threshold the notice picks up a claims read before the manager sees it.

Action

ReleaseAmendSend to claims review

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 shipment

From the carrier feed

03Exception handling

Classify the exception

Draw on the milestone feed, the order's notice history and the service commitment configured for that lane.

01Approved path

Catch the exception early

Routine tracking, exception classification and invoice reconciliation arrive already done.

02Human review

Send review to the notices that bind

Shipment means goods physically in the carrier's possession, not a label bought or an order picked — so the notices resting on that distinction are marked and the manager reads those first.

04

Build an evidence trail

The milestone data, the notice sent and its consent basis stay with the shipment.

Integrations

Typical integrations

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

  • Carrier and tracking

    • FedEx
    • UPS
    • USPS APIs
    • DHL
    • regional carrier feeds

  • Visibility platforms

    • project44
    • FourKites
    • Tive
    • Controlant

  • Transport management

    • Oracle OTM
    • Blue Yonder TMS
    • MercuryGate
    • e2open

Agent

Shipping and route exceptions

  • Reads the milestone
  • Classifies the exception
  • Holds for release

  • Orders and customs

    • Shopify
    • order platforms
    • Customs-broker filings

  • 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 notice

Each wraps the last. The map below names what each one lets through.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeStop notices and hold them for a person when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, carrier-API and notice-template changes.Track
L4TraceabilityRecord the milestone data, the notice, its consent basis and the release time.Record
L3Manager releaseHold notices for the transport manager; it governs the send, not whether the consent basis underneath it is right.Gate
L2Policy guardrailsTest notices against the notice sequence, hazmat rules, setpoints and tariff deadlines; a failure returns the notice.Restrict
L1Confidence thresholdsRoute low-confidence exceptions to a claims read before the manager sees them.Require review
Model coreException produced — exception type, drafted notice, claim exposure and confidence
L1 – L2
Test whether a notice may stand
L3
Puts the send in a manager's hands
L4 – L5
Preserve the notice sequence and its basis
L6
Stops notices and holds them for a person

How Nestack evaluates it

Evaluate the whole exception path — not only the notice at the end of it.

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

Surface — the notice the buyer receives

  1. E1Final-output evaluationDid the notice carry the revised date and the cancel-and-refund option?
  2. E2Step-level evaluationDid the agent read the right milestone, notice history and service commitment?
  3. E3Tool evaluationDid it read and write the correct shipment, the correct order and the correct field?
  4. E4Confidence calibrationDo low-confidence exceptions actually attract more manager amendments?
  5. E5Slice evaluationHow does notice timing change across specific lanes and services?
  6. E6Business outcomeHow many notices were amended, and how many refunds were owed afterwards?

Floor — the delivery the customer gets

Failure modes

Where each failure originates in the agent

Seven modes, each at its point of origin.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • SX-01

      Pre-transit scan read as delivery

      An information-received scan closes an exception on an undelivered parcel.

    Stage gathers

    Milestone events, notice history, with the source each came from

  2. 02 · Reasoning

    2 modes

    • SX-02

      Renewed notice treated as a first

      Silence is taken as consent on orders the rule has already deemed cancelled.

    • SX-03

      Placarded load routed on transit time

      A reroute optimised for speed runs a hazmat load through a restricted route.

    Stage proposes

    Exception type, drafted notice and confidence

  3. 03 · Tool / write

    2 modes

    • SX-04

      Re-consignment without a setpoint

      A temperature-controlled load moves to a lane nobody confirmed holds it.

    • SX-05

      Demurrage approved unread

      An invoice clears inside the dispute window and the mitigation right is gone.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • SX-06

      Notice omits the refund option

      A new estimated date goes out without the cancel-and-refund choice.

    Stage returns

    The notice the manager releases and the buyer

  5. 05 · Change / Version

    1 mode

    • SX-07

      Carrier code changes meaning

      An API bump reclassifies delivery refused as in transit, and nothing announces it.

    Stage tracks

    Model, prompt, carrier-API versions and notice templates

  • Sev-1 · a held act performed by the agent
  • Sev-2 · a non-compliant notice reaches
  • Sev-3 · feed degrades, exception routes to review
See the complete Retail failure-mode catalogue

Affected slices

The worst cohort names itself

Cross-border parcels on a renewed notice are their own answer: mishandled exceptions run several times the all-shipment baseline there Nestack reports it by slice rather than in aggregate..

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Cross-border, renewed notice9.1%3.5×Review
Hazmat and cold chain, disrupted7.3%2.8×Review
Intermodal, claim deadline running3.9%1.5×Watch
Domestic parcel, first notice1.3%0.5×Normal

Bar: mishandled-exception lift vs. the all-shipment baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

The regression suite is what closes a cycle

A cycle is finished when the failure sits in the suite as a case the next release must pass.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Mishandled exceptions rise on one lane.

  2. 02

    Diagnose

    The manager reads the notices and the milestones behind them until one cause is left.

  3. 03

    Improve

    The correction carries its version and the exceptions behind it.

  4. 04

    Verify

    The affected cases are re-run before anything ships.

  5. 05

    Learn

    It lands in the permanent suite and in the exception playbook.

  6. Learn → Detect

    The return edge. The next detection runs against a suite this cycle lengthened.

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, exception workflow, evaluation, integration, then production validation and handover.

Workstream

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

  1. 01Lane and exception-type discovery and boundary definition.Week 1
  2. 02Carrier, visibility and TMS source assessment.Week 1, Week 2
  3. 03Notice-sequence, deadline and handling-rule mapping.Week 1, Week 2
  4. 04Milestone ingestion and event normalisation.Week 2
  5. 05Exception classification and notice.Week 2, Week 3
  6. 06Confidence scoring and release routing.Week 3
  7. 07Transport-manager release.Week 3, Week 4
  8. 08Carrier, TMS and order-system integration.Week 5
  9. 09Notice regression cases.Week 4, Week 5
  10. 10Guardrails and notice controls.Week 4
  11. 11Notice-trail 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 lane, one carrierProductionProduction carrier estateAdvancedMultiple modes / regions
Introduced at Pilot
Exception handling on your lanes✓✓✓
Manager release✓✓✓
Notice-timing baseline✓✓✓
Introduced at Production
Reporting by lane—✓✓
Release workflow in your systems—✓✓
Approved write-back—✓✓
Visibility-platform integration—✓✓
Introduced at Advanced
Multi-jurisdiction notice rules——✓
Multi-stage transport approvals——✓
High shipment volume——✓
Enterprise carrier controls——✓
Build priceFrom $5,000From $8,000Custom 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.

  1. 01Your lanes, carriers and service commitments

    Milestone ingestion and event normalisation

    Week 1

  2. 02Representative exceptions already worked

    Exception-classification baseline and notice drafting

    Week 2

  3. 03Your notice templates and consent logic

    Notice-sequence and handling-rule mapping, and automation-boundary definition

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Carrier, visibility and TMS assessment, then integration setup

    Week 2

  5. 05Exceptions you found out about late

    Notice-timing cases and the evaluation suite

    Week 4

  6. 06What must reach a person before a notice goes out

    Exception thresholds, notice routing and controls

    Week 3

  7. 07A named transport manager to authorise reroutes

    Transport-manager release 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

Phases sit where the work is, which is why week 5 holds evaluation and pilot at the same time.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Lane and exception discovery, notice mapping and the automation boundary

W2Carrier feeds connected, then a baseline

W3Exception workflow, confidence logic and release controls

W4Notice-timing cases and carrier guardrails

W5Carrier and TMS integration, pilot lanes and targeted corrections

W6A week of live exceptions under the transport manager, then handover

Reading the band
Week 5 shows two bars because two phases actually run in it.
At the end of W6
Everything verified on live lanes, with Agent Care carrying it forward.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Retail AI agent

Build an exception agent around the manager who releases the notice.

Show us your lanes, your carriers and who signs a claim position. If your notice logic treats a renewed delay the way it treats the first one, the orders it leaves live have already been deemed cancelled unless the buyer affirmatively agreed — and a credit, a voucher or a gift card is not the refund that owes. We map the exception workflow, set the automation boundary, and name what stays with the transport manager, the claims manager, the hazmat employee and your broker.

Typical delivery
4–6 weeks
Deployment
Your carriers and TMS
Manager release
Draft-only
After launch
Agent Care available

Nestack Agents · Shipping and route exceptions

AGT-RT-07 · Agent Care available after launch

Build an exception agent

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

About you
Your shipments

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