Advertising AI agent · Retail media
Retail-Media Campaign Copilot (Multi-Network)
Run sponsored-product, brand and display campaigns across retail networks, keep spend off products that cannot be bought, and report each network's numbers as that network's — bids and budgets stay with a named person.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Bid release
- Named owner
- After launch
- Agent Care
What this agent does
Runs the campaigns, not the commercial calls
In
- 01
Read each network's campaigns, keywords, product targets and released budgets, in that network's own taxonomy.
- 02
Pull catalogue, stock, buy-box and product-page state for every line a campaign points at.
Reason
- 03
Set each network's reported spend and attributed sales beside the brand's own sales and margin data.
- 04
Name the window and the sale definition behind every figure, and never add two networks together.
- 05
Draft bid, budget, keyword and product-target changes with the limit each one uses and its reversal.
Decide
- 06
Hold spend on anything pointing at a product that is out of stock, unbuyable or newly suppressed.
- 07
Route every bid, budget, margin trade-off and any copy carrying a claim to the named owner.
Out
- 08
Apply released changes on one network at a time, and write the same change back to the plan.
- 09
Retain the state replaced, the approval, each network's reported figures and the date they were read.
Product statement
The agent manages campaigns inside limits a person set. It does not set bids or budgets, and it does not decide when revenue is worth less than margin.
Example workflow
One bid change, end to end
- 1Campaign state pulledCampaigns, ad groups, keywords and product targets from each network in scope, in that network's own taxonomy
- 2Catalogue and stock checkedStock, buy-box, price and product-page state for every line the campaign points at, stamped with the time read
- 3Numbers set side by sideEach network's reported spend and attributed sales against the brand's own sales and margin, with the window each network used
- 4Change draftedThe bid, budget, keyword or product-target change, the limit it uses, the margin on the line and the reversal that puts it back
No human action required
Stages 1 to 4 run without a person in the loop — the campaign pull, the stock check and the drafted change are finished before the owner is asked to look at anything.
Reaches the owner ready to release.
Held with the reason named, nothing applied.
Named campaign owner
The owner checks the change against the plan, the margin on the line and what the network actually reported, then releases it.
Release · Amend · Send back
Released — handed back
- 6Applied and watchedOnly on the networks and lines the owner released; stock and spend are re-checked on a fixed interval
- 7Outcome evaluatedOut-of-stock spend, cross-network reconciliation, window correctness, taxonomy mapping and margin on what sold
Restatements
When a network revises a figure the agent acted on, the change is reopened.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Setting, raising or moving any bid or budget.
- Trading margin away for network-reported revenue.
- Running copy that carries an unsubstantiated claim.
- Bidding on a term the brand already ranks for.
Automation boundary
Agent acts unaided
- ✓Read campaigns, keywords and product targets from every network.
- ✓Check stock and buy-box state on every targeted line.
- ✓Hold spend on a pre-authorised line that becomes unbuyable.
- ✓Report each network's figures as its own, with the window used.
Write actions run only inside the approval boundaries agreed during implementation. Setting a bid, moving a budget and trading margin are not among them. A hold needs a named owner's written pre-authorisation, a ceiling and a reversal.
- Adding a competitor term or a restricted product.
- Turning on a network's automated bidding feature.
- Carrying a network's restated figures into finance.
- Changing the limits, category mappings or margin floors.
Example output
One campaign line, annotated
Everything the agent proposes is attached to the line it came from.
Retail-media output · single campaign line
Illustrative example- Network
- Grocery retailer, onsite
- Target
- Sponsored product, one line
- Released budget
- $12,000
- Proposed change
- Hold the target
- Confidence
- 83%
- Status
- Held for the owner
As received
The campaign line and the budget released against it, in the network's own taxonomy — nothing on this side is inferred.
Evidence used
- Stock feed, timestamped
- Network report, dated
- Margin on the line
Why it is held
The line went out of stock; the network still reports sales in its own window.
What the score decides
How closely the owner reads before releasing — not whether the change makes or loses money.
Action
ReleaseAmendSend backValue
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 campaign line
From each network in scope
Keep spend pointed at buyable products
Stock, buy-box and product-page state are checked against every targeted line before a change is drafted, and again while it runs.
Take the cross-network assembly off the team
The campaign pull, the category mapping and the window-by-window read are done before a person opens a console, so the owner decides rather than collates.
Show whose number it is
Every attributed sale carries the network that reported it and the window it used, and a network's own ROAS reaches the owner as that network's measure — not an independent one.
Build an evidence trail
Retain the state replaced, the limit used, who released it, each network's figures with the date they were read, and every later restatement — on both paths.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Retail-media networks
- Amazon Ads
- Walmart Connect
- Criteo
- Instacart
- Roundel
- retailer APIs
Catalogue & stock
- Product feeds
- stock and buy-box
- Price and promotions
- PDP content
Sales & margin
- Brand ERP
- POS and syndicated data
- Cost and margin by line
Agent
Retail-media campaign management
- Reads every network
- Proposes changes
- Holds unbuyable lines
Reporting & finance
- Warehouse
- BI and dashboards
- Invoices
- trade finance
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 your networks
Each control wraps the one inside it. A change clears every layer before it reaches a network, and the bid itself sits outside all six.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Decide whether the change may stand
- L3
- Decides who sets the bid and the budget
- L4 – L5
- Keep networks apart and figures dated
- L6
- Pulls automation back when signals degrade
How Nestack evaluates it
Evaluate the spend and the reconciliation — not the network's dashboard.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the campaigns and changes an owner releases
- E1Final-output evaluationDid the change reach the right campaign, keyword and product target?
- E2Step-level evaluationDid it read stock, buy-box and each network's own window correctly?
- E3Tool evaluationDid it write to the correct network, account and line, and nothing else?
- E4Reconciliation and driftDo network-reported sales still tie to your own data after a restatement?
- E5Slice evaluationHow does performance change across networks, categories and stock states?
- E6Business outcomeOut-of-stock spend, the reconciliation gap, and margin on what actually sold.
Floor — spend that reached buyable products and reconciled
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 →
01 · Network read
2 modes
- RM-01
Catalogue change unseen
An item ID moves and the target matches nothing.
- RM-02
Wrong taxonomy map
One network's category maps to another wrongly.
Stage gathers
Campaigns, targets, stock and each network's window
02 · Bid & target
2 modes
- RM-03
Cannibalised organic
A bid is raised on a term the brand already won.
- RM-04
Margin ignored in bids
The line it pushes loses money when it sells.
Stage proposes
The bid, budget, keyword or product-target change
03 · Network write
1 mode
- RM-05
Unbuyable line spends
Stock ran out and the campaign keeps paying.
Stage applies
Only on the networks and lines the owner released
04 · Reconciliation
1 mode
- RM-06
Same sale counted twice
Two networks and a channel each claim one order.
Stage ties
Each network's figures back to the brand's own data
05 · Change / Version
1 mode
- RM-07
Network changes its rules
A window or sale definition moves under you.
Stage tracks
Model, prompt, limit and network-definition changes
- Sev-1 · money is spent that no one released
- Sev-2 · a decision rests on a wrong number
- Sev-3 · the read is stale before anything moves
Affected slices
Most of the wasted spend sits in a few cohorts
Aggregate campaign quality can look acceptable while a few network and stock-state cohorts carry most of the wasted spend and most of the reconciliation rework. Nestack reports performance by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Out-of-stock and suppressed lines | 5.7% | 2.9× | Review | |
| Networks with unlike windows | 4.7% | 2.4× | Review | |
| Newly mapped categories | 3.5% | 1.8× | Watch | |
| Single-network steady campaigns | 1.0% | 0.5× | Normal |
Bar: failure-rate lift vs. steady-campaign baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
The number you improved against can be rewritten
Networks restate attributed sales and change their own definitions after the fact. A cycle closed against last month's report can be reopened by this month's.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Out-of-stock spend, a reconciliation gap or margin moves on one network.
- 02
Diagnose
Traced to the stock read, the category map, a window or the figure trusted.
- 03
Improve
The mapping, rule or limit is changed, re-approved by the owner and version-linked.
- 04
Verify
Re-run against held-out campaigns from that network, including reversed ones.
- 05
Learn
The gap becomes a regression case and the network's quirk enters the runbook.
Learn → Detect
The return edge. Every cycle re-reads each network's current definitions — a figure that reconciled last month may not reconcile now.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, network and catalogue data, proposals and limits, evaluation, reconciliation, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Workflow discovery and automation-boundary definition.Week 1
- 02Network scope and API assessment.Week 1, Week 2
- 03Catalogue, stock and buy-box feed integration.Week 2
- 04Cross-network category mapping.Week 2, Week 3
- 05Attribution-window and metric register.Week 2, Week 3
- 06Bid, budget and product-target proposals.Week 3
- 07Buyability rules and pre-authorised holds.Week 3, Week 4
- 08Margin data and review sampling.Week 4
- 09Named-owner release and approval workflow.Week 4
- 10Evaluation suite, regression cases and network slices.Week 4, Week 5
- 11Cross-network sales reconciliation.Week 5, Week 6
- 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.
| Capability✓ in scope · — not at this tier | PilotOne network, one category | ProductionProduction network integration | AdvancedMulti-network / multi-brand |
|---|---|---|---|
| Introduced at Pilot | |||
| Campaign and target management | ✓ | ✓ | ✓ |
| Buyability checks and hold proposals | ✓ | ✓ | ✓ |
| Per-network reporting, as reported | ✓ | ✓ | ✓ |
| Named-owner release | ✓ | ✓ | ✓ |
| Baseline evaluation | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Cross-network category mapping | — | ✓ | ✓ |
| Attribution-window register | — | ✓ | ✓ |
| Released changes on network | — | ✓ | ✓ |
| Cross-network reconciliation | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Pre-authorised holds with stated limits | — | — | ✓ |
| Multi-brand and enterprise controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on the networks and categories in scope, managed spend, catalogue and stock feed integrations, margin-data access, 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
01The networks, brands and categories in scope
→Network scope and retail-media API assessment
Week 102Catalogue, stock and buy-box feeds you can expose
→Catalogue and stock integration, then the buyability rules
Week 203How your categories line up across each retailer
→Cross-network category mapping and the window register
Week 204Who may set a bid or move a budget, and up to what
→Spend limits, buyability holds and named-owner release
Week 305Cost and margin by line, and who is allowed to see it
→Margin data and the margin-aware review sample
Week 406Campaigns you would have paused, and ones you kept
→Evaluation suite, regression cases and failure-mode testing
Week 407Your own sales data and the report finance signs off
→Cross-network reconciliation against your sales data
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. Week 5 carries both the network slices and the first changes released on live campaigns.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Networks and categories in scope; who may set a bid
W2Network APIs, catalogue and stock feeds, category mapping
W3Window register, change proposals and buyability holds
W4Owner release workflow, margin sampling and evaluation suite
W5First released changes on live campaigns, and targeted corrections
W6Cross-network reconciliation, validation and Agent Care handover
- Reading the band
- The buyability hold is built and drilled in week 3, before a change touches a live campaign in week 5. The bars show that dependency, not a smooth ramp.
- At the end of W6
- Changes have run under named-owner release and each network's figures have been tied back to your own sales data, then Agent Care takes over monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Advertising AI agent
Build a retail-media agent around your networks and your margin.
Show us the networks you buy on, how stock and catalogue reach them, and who is allowed to set a bid today. We'll put one network's reported figures beside your own sales data for one category, and start from whatever that gap turns out to be.
- Typical delivery
- 4–6 weeks
- Deployment
- Your retail-media stack
- Bid release
- Named owner, always
- After launch
- Agent Care available
Nestack Agents · Retail-media campaign management
AGT-AM-12 · Agent Care available after launch