Main site

Retail AI agent · Agentic checkout

Agentic-Checkout & Buy-for-Me AI Agent

Assemble the cart, present a per-order authorisation carrying the all-in total and the limits the shopper set, and place nothing until the shopper has authorised that exact order.

Typical delivery
4–6 weeks
Deployment
Your stack
Shopper decides
Per-order
After launch
Agent Care

What this agent does

Assembles the order, never authorises it

In

  1. 01

    Gather candidate products, live prices and merchant terms from supported commerce and agent-protocol sources.

  2. 02

    Rank the candidates, and mark which merchants permit automated purchasing at all.

Reason

  1. 03

    Price the all-in total — item, shipping, tax and every mandatory fee — before anything is shown.

  2. 04

    Draw the mandate the shopper set: scope, price ceiling, category limits, ship-to and expiry.

  3. 05

    Bind the cart to that mandate, and mark where the cart falls outside it.

Decide

  1. 06

    Name any commercial relationship that influenced the selection.

  2. 07

    Present the order to the shopper for a per-order authorisation, and wait there.

Out

  1. 08

    Keep the mandate, the cart, the totals and the authorisation against the order.

  2. 09

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

Product statement

The agent proposes the order; the shopper authorises it — and whether a wrong purchase counts as unauthorised is unsettled, which is why the authorisation record matters.

Example workflow

One order, intent to authorisation

  1. 1Shopping intent receivedA stated request, a reorder, or a standing shopping list
  2. 2Mandate assembledProducts, live prices, merchant terms, agent permissions and an all-in total for each
  3. 3Cart built and heldThe exact item, quantity, merchant, all-in total, ship-to, tender and confidence
  4. 4Mandate checks runMandate checks, ceiling and expiry checks, merchant-permission checks and confidence threshold

No human action required

Stages 1 to 4 run unaided and no order exists yet — the cart sits un-submitted, and the shopper's lane opens at the confidence gate.

5DecisionSplits on the mandate limit
Inside the mandate

Goes to the shopper to authorise.

Outside the mandate

Adds a mandate re-check first.

Shopper authorisation

The cart is held with its mandate, its all-in total and the protections disclosure.

Authorise · Amend · Cancel

Authorised — order placed

  1. 6Merchant accepts or declinesThe merchant stays the seller of record throughout
  2. 7Order evaluatedMandate conformance, total drift, cancellation-window use and orders the shopper later disowned

Cancellations

Every amendment the shopper makes is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Buying age-restricted, prescription or otherwise regulated goods.
    • Accepting merchant terms, arbitration clauses or warranty disclaimers.
    • Enrolling in a subscription, auto-renewal or free-to-pay conversion.
    • Buying where the merchant's terms bar agents, or access was revoked.

  • Automation boundary

    Agent acts unaided

    • ✓Search, compare and rank candidates with each all-in total itemised.
    • ✓Build a cart and hold it un-submitted at a merchant.
    • ✓Present the authorisation record.
    • ✓Cancel an in-flight order and revoke the delegated token behind it.

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

    • Disguising the agent, spoofing a session or passing a bot control.
    • Shipping to an unverified address, or using a new payment instrument.
    • Exceeding the per-order maximum charge, or the authorisation's expiry.
  • Filing or arguing a bank or card dispute on the shopper's behalf.

Example output

One authorisation record, annotated

What the shopper authorises is the exact order the agent is permitted to place.

Authorisation record · single order

Illustrative example
Order
Reorder, same item
What will be bought
One unit of the exact item bought before, from the same merchant
Total shown
Item, shipping, tax
Mandate
Per-order, with expiry
Confidence
88%
Protections
Caps may not apply

As received

Taken from the merchant's live listing and the shopper's own mandate — nothing on this side is assumed.

Mandate terms used

  • Maximum charge amount
  • Merchant and category limits
  • Verified ship-to address

Why this disclosure

The shopper authorised the agent and a person still decides.

What the score decides

Below the configured threshold the cart is re-checked against the mandate.

Action

AuthoriseAmendCancel

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 order

From the shopper's mandate

03Assembly

Build against the mandate

Work inside the scope, the ceiling, the category limits and the expiry the shopper set.

01Approved path

Buy only what was authorised

Reorders and tightly stated intents arrive as a cart ready to authorise.

02Human review

Stop where the mandate runs out

Anything past the scope, the ceiling or the merchant's own agent terms stops and waits for the shopper.

04

Build an evidence trail

The mandate, the cart and the authorisation stay attached to the order.

Integrations

Typical integrations

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

  • Agent-commerce protocols

    • Agentic Commerce Protocol
    • Visa Trusted Agent Protocol

  • Network agentic credentials

    • Visa Intelligent Commerce
    • Mastercard Agent Pay

  • Payment execution

    • Stripe
    • Adyen

Agent

Agentic checkout & buy-for-me

  • Builds the cart
  • Holds it for the shopper
  • Places on authorisation

  • Merchant surfaces

    • Shopify
    • BigCommerce
    • Signifyd
    • Forter

  • 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 checkout

Layers wrap inward, and the map below names what each one misses.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeFall back to cart-building only when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, mandate-rule and merchant-terms changes.Track
L4TraceabilityRecord the mandate, the cart, the totals, the authorisation and the agent signature.Record
L3Shopper authorisationHold each cart for a per-order authorisation; it records consent, it does not make a wrong order reversible.Gate
L2Policy guardrailsTest carts against the mandate, the ceiling and the merchant's agent terms; a failure returns the cart.Restrict
L1Confidence thresholdsRoute low-confidence carts to a mandate re-check before the shopper sees them.Require review
Model coreCart held — item, quantity, merchant, all-in total, ship-to and confidence
L1 – L2
Test whether a cart may be offered
L3
Puts the decision in the shopper's hands
L4 – L5
Keep the mandate and what was bought against it
L6
Suspends buying and holds the cart when signals slip

How Nestack evaluates it

Evaluate the whole purchase — not only whether checkout succeeded.

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

Surface — the order the shopper is charged for

  1. E1Final-output evaluationDid the order match the item, quantity and total the shopper authorised?
  2. E2Step-level evaluationDid the agent use the live listing, the current terms and the shopper's mandate?
  3. E3Tool evaluationDid it transact at the intended merchant, on the intended token?
  4. E4Confidence calibrationDo low-confidence carts actually attract more amendments?
  5. E5Slice evaluationHow does performance change across specific merchants and intent types?
  6. E6Business outcomeHow many orders were cancelled, returned, or disowned by the shopper afterwards?

Floor — the order the shopper is charged for

Failure modes

Where each failure originates in the agent

Seven modes, from mandate to settlement.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • AK-03

      Stale product page

      A delisted variant is bought from a cached listing at an old price.

    Stage gathers

    Listings, live prices, with the source each came from

  2. 02 · Reasoning

    2 modes

    • AK-04

      Trial SKU chosen

      The cheapest option resolves to a free-trial or auto-renewing plan.

    • AK-06

      Session read as per-order

      One stated intent becomes three separate orders inside the budget.

    Stage proposes

    The item, the all-in total and confidence

  3. 03 · Tool / write

    2 modes

    • AK-02

      Double charge on retry

      Checkout times out after authorisation and is retried without a key.

    • AK-05

      Bought where agents are barred

      Checkout completes at a merchant that had revoked agent access.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • AK-01

      Pre-tax total shown

      The shopper authorises a number that is not the amount charged.

    Stage returns

    The order the shopper is charged for

  5. 05 · Change / Version

    1 mode

    • AK-07

      Checkout page changes

      A newly inserted disclaimer or arbitration box is skipped and accepted.

    Stage tracks

    Model, prompt, mandate rules and merchant terms

  • Sev-1 · a held act performed by the agent
  • Sev-2 · an order outside the mandate is placed
  • Sev-3 · source degrades, cart routes to re-check
See the complete Retail failure-mode catalogue

Affected slices

One number cannot carry this

Unintended purchases are counted per order and read as a lift on the all-order baseline. Funding type Nestack reports it by slice rather than in aggregate The cohorts that carry it are named, not averaged away..

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Debit, anti-agent terms, final sale9.0%2.9×Review
Open-ended comparison intents6.8%2.2×Review
Credit, protocol-integrated5.6%1.8×Watch
Reorder under a per-order mandate2.8%0.9×Normal

Bar: unintended-purchase lift vs. the all-order baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

No cycle here closes on agreement

The cycle closes on a test, not a write-up — one the next release has to pass That suite is what the following detection is measured against..

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Disowned orders cluster in one intent type.

  2. 02

    Diagnose

    The authorisation records are replayed against the carts until one cause is left.

  3. 03

    Improve

    The change is stamped to a version, with the orders that produced it attached.

  4. 04

    Verify

    Re-run the affected cases; a fail holds the release.

  5. 05

    Learn

    A standing test, plus a change to the authorisation rules.

  6. Learn → Detect

    The return edge. Each detection meets more standing cases than the one before.

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

Workstream

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

  1. 01Purchasing workflow discovery and boundary definition.Week 1
  2. 02Merchant and payment-source assessment.Week 1, Week 2
  3. 03Mandate, ceiling and merchant-terms mapping.Week 1, Week 2
  4. 04Catalogue and price ingestion with all-in totals.Week 2
  5. 05Cart assembly and mandate binding.Week 2, Week 3
  6. 06Confidence scoring and authorisation routing.Week 3
  7. 07Shopper authorisation workflow.Week 3, Week 4
  8. 08Wallet, token and merchant integration.Week 5
  9. 09Mandate regression cases.Week 4, Week 5
  10. 10Guardrails and spend controls.Week 4
  11. 11Mandate-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 merchant, one mandateProductionProduction order volumeAdvancedMultiple merchants / tenders
Introduced at Pilot
Carts built to your mandate✓✓✓
Shopper authorisation✓✓✓
Mandate-conformance baseline✓✓✓
Introduced at Production
Reporting by merchant—✓✓
Authorisation workflow in your app—✓✓
Approved order placement—✓✓
Wallet and network integration—✓✓
Introduced at Advanced
Multi-merchant mandate sets——✓
Multi-party authorisation——✓
High order volume——✓
Enterprise mandate 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. 01The merchants you buy from and their agent terms

    Merchant-terms and agent-permission mapping

    Week 1

  2. 02Representative past orders

    Cart-assembly baseline, ranking and all-in totals

    Week 2

  3. 03Your ceilings, category limits and ship-to list

    Mandate and ceiling mapping, and the automation boundary

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Merchant, protocol and payment assessment, then integration setup

    Week 2

  5. 05Orders the shopper did not want

    Mandate cases and the evaluation suite

    Week 4

  6. 06The mandate limits the shopper actually set

    Mandate limits, spend routing and authorisation controls

    Week 3

  7. 07The shopper's own authorisation record

    Shopper authorisation 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 band spans the weeks its work is named in, and week 5 carries two because both genuinely run.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Purchasing workflow discovery, mandate mapping and the automation boundary

W2Merchant and wallet connections established

W3Mandate binding, confidence logic and authorisation controls

W4Mandate cases and spend guardrails

W5Token and wallet integration, pilot orders and targeted corrections

W6One mandate run end to end under the shopper's authorisation

Reading the band
No week is padded. The fifth carries evaluation and pilot together.
At the end of W6
Hand-over happens against a verified mandate run, not a demo.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Retail AI agent

Build a buy-for-me agent around a mandate you can produce later.

Show us what your shoppers ask for, which merchants you buy from and the limits they set. Nobody has yet decided whether a wrong purchase is unauthorised, so we build the record that will have to answer for it.

Typical delivery
4–6 weeks
Deployment
Your merchants
Shopper authorises
Per-order
After launch
Agent Care available

Nestack Agents · Agentic checkout and buy-for-me

AGT-RT-03 · Agent Care available after launch

Build a checkout agent

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

About you
Your orders

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