Retail AI agent · Customer service
Customer-Service & Order-Status AI Agent
Answer order-status and policy questions from the live order record, hand anything legally operative to a named person, and never re-promise a date the merchant's own systems have not confirmed.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Escalation lead
- Answers only
- After launch
- Agent Care
What this agent does
Answers the question, never the promise
In
- 01
Pull the order, its per-line fulfilment state and the live carrier scan from supported commerce.
- 02
Timestamp every field the agent reads, and carry that as-of time forward into the answer.
Reason
- 03
Compose the reply from those fields, with the AI disclosure stated before anything else.
- 04
Apply the merchant's published shipping, return, warranty and privacy text as written.
- 05
Tie each stated fact to the field it was read from, and mark what the record does not say.
Decide
- 06
Detect a shipping promise, a warranty claim, a cancellation or a dispute inside the contact.
- 07
Route those contacts to the Customer Experience Escalation Lead instead of answering them.
Out
- 08
Retain the fields read, the reply, the disclosure and the routing against the ticket.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent answers from the record; the escalation lead decides anything discretionary, and a delay notice or a refund stays with the people the merchant has named to author them.
Example workflow
One contact, arrival to answer
- 1Contact receivedChat, email, voice or the helpdesk queue
- 2Context assembledFulfilment state per line, carrier scan, tender, policy version and subscription status
- 3Reply composedThe answer, its as-of time, any escalation trigger hit and confidence
- 4Controls appliedAI disclosure, source-field checks, escalation-trigger checks and confidence threshold
No human action required
Stages 1 to 4 run unaided and nothing has been sent — the lead's lane stays empty until a trigger fires or the confidence gate opens it.
Sends on the approved path.
Goes to the escalation lead first.
Escalation lead
The reply is held with the fields it read, its triggers and the confidence.
Send · Rewrite · Escalate to compliance
Released — answer sent
- 6Helpdesk record updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedStatement accuracy against the record, escalation outcomes, reopens and post-answer corrections
Amendments
Every rewrite 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
- Issuing a delay notice, or the renewed notice after it.
- Authorising a refund, or stating when one will land.
- Deciding a warranty claim or what its coverage reaches.
- Denying, delaying or conditioning a subscription cancellation.
Automation boundary
Agent acts unaided
- ✓Report tracking from the live carrier field with its as-of time for the named owner.
- ✓Restate the published return.
- ✓Resend an order confirmation or invoice to the email.
- ✓Answer product-attribute questions from the live catalogue record.
Any write happens inside the boundaries agreed at implementation, never ahead of the lead's release.
- Changing a shipping address, recipient or method after labelling.
- Any statement about a bank or card dispute's timing or outcome.
- Age-restricted, prescription, firearms or recalled-product contacts.
- Granting a policy exception, goodwill credit or price adjustment.
Example output
One answer, annotated
Everything the agent states is attached to the field it was read from.
Order-status output · single contact
Illustrative example- Order
- Two-line order
- Stated answer
- One line is scanned and moving; the second is still unfulfilled
- Carrier status
- In transit
- Source of record
- Live carrier feed
- Confidence
- 93%
- Disclosure
- Stated as AI at the start
As received
Taken from the order record and the carrier feed at the time of reading.
Fields read
- Per-line fulfilment state
- Latest carrier scan
- Published shipping text
Why this answer
It reports what the feed said as of the read, and stops short of a delivery date.
What the score decides
Below the configured threshold the reply routes to the escalation lead instead of.
Action
SendRewriteEscalate to complianceValue
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 contact
From the order record
Answer from the record
Draw on the live order, carrier and published-policy fields the merchant already maintains.
Answer without the queue
Routine status and policy questions come back without a wait.
Send the desk what carries risk
Promise, warranty, cancellation and dispute contacts are flagged, so the lead's time lands where exposure sits.
Build an evidence trail
The question, the record it drew on and the routing stay with the ticket.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Commerce platform and OMS
- Shopify
- Salesforce Commerce Cloud
Post-purchase tracking
- Narvar
- AfterShip
- carrier APIs
Voice and CCaaS
- Amazon Connect
- Genesys Cloud CX
Agent
Customer service & order status
- Reads the order
- Composes the answer
- Routes escalations
Desk and subscriptions
- Zendesk
- Gorgias
- Recharge
- Chargebee
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 customer
Each control wraps the one inside it. What a layer does not catch is named in the failure-mode map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a reply may be sent
- L3
- Puts the release in the lead's hands
- L4 – L5
- Keep the answer and its source reconstructable
- L6
- Pulls the agent back to routing when signals degrade
How Nestack evaluates it
Evaluate the whole contact — not only the sentence that was sent.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the answer the customer reads
- E1Final-output evaluationDid every stated fact match the field it was read from?
- E2Step-level evaluationDid the agent read the right order, the current scan and the live policy version?
- E3Tool evaluationDid it read and write the correct order and the correct ticket?
- E4Confidence calibrationDo low-confidence replies actually attract more corrections?
- E5Slice evaluationHow does performance change across specific contact reasons?
- E6Business outcomeHow many contacts reopened, escalated late or needed a correction after sending?
Floor — the outcome the retailer answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, each pinned to where it starts.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- OS-03
Stale carrier scan
The last known scan is reported as the current one.
Stage gathers
Order state, carrier scans and the published policy set
02 · Reasoning
2 modes
- OS-04
Invented delivery date
Transit days are added to an order date on a line that never shipped.
- OS-06
Warranty read as a return
A defect claim is answered with the closed return window.
Stage proposes
The answer, its as-of time and confidence
03 · Tool / write
2 modes
- OS-02
Address changed post-label
A labelled parcel is re-routed and the carrier record breaks.
- OS-05
Cancellation noted, not called
A note lands in the desk while billing carries on unchanged.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- OS-01
Disclosure deflected
Asked whether it is a person, the agent changes the subject.
Stage returns
The answer the customer reads and relies on
05 · Change / Version
1 mode
- OS-07
One channel regresses
An update drops the disclosure from voice and leaves chat intact.
Stage tracks
Model, prompt, policy text and escalation config
- Sev-1 · a held act performed by the agent
- Sev-2 · a wrong statement reaches the customer
- Sev-3 · source degrades, reply routes to review
Affected slices
A healthy average can rest on one contact reason
The metric is the materially-incorrect-statement rate, read as a lift against the all-contact baseline. A handful of reason codes carry most of the wrong statements Nestack reports it by slice rather than in aggregate..
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Split orders, ship date missed | 4.5% | 3.2× | Review | |
| Cancellation asked mid-contact | 3.5% | 2.5× | Review | |
| Warranty outside the window | 2.1% | 1.5× | Watch | |
| Single line, shipped, scanned | 0.7% | 0.5× | Normal |
Bar: materially-incorrect-statement lift vs. the all-contact baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What a cycle has to leave behind
Nothing closes on an explanation. The cycle ends when the failure exists as a regression case the next release has to pass.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Incorrect-statement rate climbs in one contact reason.
- 02
Diagnose
The escalation lead reads the contacts and the fields behind them until one cause is left.
- 03
Improve
The fix is written against a version, with the orders that motivated it attached.
- 04
Verify
Release is blocked until the affected regression cases pass again.
- 05
Learn
The case joins the permanent suite and the response playbook.
Learn → Detect
The return edge. Each pass begins with more cases to clear than the one before it.
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, answering workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Contact workflow discovery and boundary definition.Week 1
- 02Commerce and order-system assessment.Week 1, Week 2
- 03Policy-text, disclosure and escalation-rule mapping.Week 1, Week 2
- 04Order and carrier ingestion with field.Week 2
- 05Answering logic and source-field binding.Week 2, Week 3
- 06Confidence scoring and escalation routing.Week 3
- 07Escalation-lead review.Week 3, Week 4
- 08Helpdesk and telephony integration.Week 5
- 09Response regression cases.Week 4, Week 5
- 10Guardrails and escalation controls.Week 4
- 11Conversation-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 channel, one brand | ProductionProduction contact volume | AdvancedMultiple brands / regions |
|---|---|---|---|
| Introduced at Pilot | |||
| Answers from your systems of record | ✓ | ✓ | ✓ |
| Escalation-lead release | ✓ | ✓ | ✓ |
| Response-quality baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by contact reason | — | ✓ | ✓ |
| Escalation workflow in your desk | — | ✓ | ✓ |
| Approved write-back to the ticket | — | ✓ | ✓ |
| Helpdesk integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-region policy sets | — | — | ✓ |
| Multi-stage escalation paths | — | — | ✓ |
| High contact volume | — | — | ✓ |
| Enterprise contact 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 contact reasons and the systems behind them
→Contact workflow discovery and boundary definition
Week 102Representative resolved contacts
→Answering baseline, field binding and disclosure wording
Week 203Your published shipping, return and privacy text
→Policy-text, disclosure and escalation-rule mapping
Week 104Access to relevant APIs, feeds or exports
→Commerce, desk and carrier assessment, then integration setup
Week 205Contacts you would not want repeated
→Response regression cases and failure-mode testing
Week 406Where an answer must stop and wait for a person
→Confidence scoring, escalation routing and approval controls
Week 307A named escalation lead to review answers
→Escalation review 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
Each phase sits on the weeks it actually occupies, and week 5 carries both evaluation and launch work.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Contact workflow discovery, policy mapping and the automation boundary
W2Contact sources connected, then a baseline
W3Answering workflow, confidence logic and escalation controls
W4Evaluation suite, guardrails and failure-mode testing
W5Helpdesk integration, pilot queues and targeted corrections
W6A live contact queue answered under review, then Agent Care 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 closes on live contacts, and Agent Care picks up monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Retail AI agent
Build an order-status agent around your escalation path.
Show us your contact reasons, your order systems and who holds the exceptions. Where a contact touches a shipping promise, a warranty, a cancellation or a dispute, we will show you where the agent stops and who picks it up.
- Typical delivery
- 4–6 weeks
- Deployment
- Your commerce stack
- Escalation lead
- Answers only
- After launch
- Agent Care available
Nestack Agents · Customer service and order status
AGT-RT-01 · Agent Care available after launch