Advertising AI agent · Audiences
Audience-Targeting Copilot (Permitted-Use)
Define, document and maintain segments, lookalikes and suppression lists with the permission behind every input on file — a named person clears the legal basis and any restricted category before an audience ships.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Activation
- Human sign-off
- After launch
- Agent Care
What this agent does
Proposes the audience, not the delivery
In
- 01
Read the segment brief, the fields available in the CDP and the permission recorded against each one.
- 02
Pull the consent, preference and opt-out state for every source, stamped with the date it was read.
Reason
- 03
Draft inclusion and exclusion rules from fields the client is permitted to use for advertising, and no others.
- 04
Test every rule for proxies — the neutral-looking field that reconstructs a protected attribute.
- 05
Size the segment, model any lookalike from a named seed, and state how far the model sits from it.
Decide
- 06
Hold anything that reads as housing, employment or credit, or that rests on a sensitive-category inference.
- 07
Route every segment to a named approver with its inputs, permission record and proxy result attached.
Out
- 08
Push only the approved definition, apply the suppression list, and record the match rate as it came back.
- 09
Retain the brief, the rules, the permission evidence, the approver and the composition actually delivered.
Product statement
The agent proposes and documents audiences. It does not clear a legal basis, does not decide a restricted category, and does not control how a platform delivers.
Example workflow
One segment, end to end
- 1Brief receivedA segment brief from a planner, or a scheduled refresh of an audience already running
- 2Permission checkedEvery field the segment would use, set beside the consent, contract and stated purpose behind it
- 3Rules drafted and sizedInclusion and exclusion rules, the seed for any lookalike, projected size and overlap with what is already live
- 4Proxy and category screenEach rule tested for proxies of a protected attribute, and the brief tested against the restricted-category list
No human action required
Stages 1 to 4 run without a person in the loop — the sourcing, the rule draft and the proxy screen are finished before anyone is asked to sign anything.
Reaches the approver ready to sign.
Held with the field named, not built.
Audience approver and privacy reviewer
A named approver confirms the legal basis for every input and clears any restricted category. Nothing reaches a platform before that.
Approve · Narrow the rule · Send back
Approved — handed back
- 6Activated and suppressedPushed only to the platform and account the approver named, with the suppression list applied and the match rate recorded
- 7Delivery reviewedDelivered composition against intended, reach and frequency, and the slices where the two diverge
Rule changes
Every rule the approver narrows or removes is counted.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Shipping any audience in a restricted category.
- Deciding the legal basis for a segment's inputs.
- Adding a field the client cannot use for advertising.
- Building a lookalike from an unapproved seed.
Automation boundary
Agent acts unaided
- ✓Draft segment rules from fields already cleared for advertising.
- ✓Check consent, purpose and opt-out state per input.
- ✓Test each rule for proxies of a protected attribute.
- ✓Size the segment, apply suppression and record the match rate.
Write actions run only inside the approval boundaries agreed during implementation. Pushing an audience to a platform is not among them.
- Inferring health, religion, politics or sexual orientation.
- Targeting an audience that may contain known minors.
- Changing or switching off a suppression list.
- Onboarding first-party data to a new platform or partner.
Example output
One segment definition, annotated
Everything the agent proposes is attached to the brief it came from.
Audience output · single segment
Illustrative example- Segment brief
- Renters near new listings
- Sources
- CDP plus a partner list
- Permission
- Read today
- Category
- Housing — restricted
- Confidence
- 91%
- Status
- Held, not built
As received
The brief as the planner wrote it, the sources it would draw on, and the date the consent and opt-out state were read.
Evidence used
- Consent record, dated
- Stated purpose on file
- Partner list, unpermitted
Why it is held
It is a housing brief, and one source has no advertising permission. Both are named.
What the score decides
It decides how hard the approver reads the rule set — not whether the audience is lawful to run.
Action
ApproveNarrow the ruleSend 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 segment request
From planners, the CDP and scheduled refreshes
Carry the permission with the field
Every field in a segment arrives with the consent, contract and stated purpose behind it, read at build time and stamped with the date.
Cut the sourcing and sizing work
Sourcing, rule drafting, sizing and overlap are finished before a person opens the brief, so the approver reads a definition instead of building one.
Surface the proxy before the push
Rules that reconstruct a protected attribute, and briefs that fall in a restricted category, are held now instead of found in a delivery report later.
Build an evidence trail
Retain the brief, the rules, the permission evidence, the proxy result, the approver, the match rate and the composition delivered — on both paths.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
CDP & first-party data
- Segment
- Tealium
- Adobe
- BigQuery
- warehouse exports
Platform audience APIs
- Meta
- The Trade Desk
- Amazon Ads
- retail-media APIs
Clean rooms & identity
- Ads Data Hub
- Amazon Marketing Cloud
- LiveRamp
- ID resolution
Agent
Audience targeting & suppression
- Reads permission
- Drafts segments
- Routes for sign-off
Consent & preferences
- OneTrust
- Didomi
- Preference centres
- suppression lists
Measurement & observability
- OpenTelemetry
- Langfuse
- Supported measurement/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the platform
Each control wraps the one inside it. A segment clears every layer before it is pushed, and how the platform then delivers it sits outside all six.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Decide whether the segment may stand
- L3
- Decides who clears a restricted brief
- L4 – L5
- Keep exclusions applied and the trail intact
- L6
- Pulls automation back when signals degrade
How Nestack evaluates it
Evaluate the definition and what was delivered against it.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the segment definition a person approves
- E1Final-output evaluationDid the segment match the brief, and was every input permitted?
- E2Step-level evaluationDid it read the current consent and opt-out state before drafting a rule?
- E3Tool evaluationDid it push the approved definition to the right platform and account?
- E4Category and proxy detectionDoes it catch a restricted brief and a proxy rule, and how often does it miss?
- E5Slice evaluationHow does performance change across sources, platforms and segment types?
- E6Business outcomeDelivered-versus-intended composition, match honesty and audiences pulled at review.
Floor — the composition the platform actually delivered
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 · Sourcing
2 modes
- AT-01
Unpermissioned input
A field the client may not use for advertising.
- AT-02
Sensitive category inferred
Behaviour is read as a health or belief signal.
Stage gathers
Fields, consent state, opt-outs and partner permissions
02 · Definition
1 mode
- AT-03
Proxy rebuilds a protected attribute
A neutral field stands in for race, age or sex.
Stage proposes
Inclusion, exclusion, the lookalike seed and its size
03 · Activation
2 modes
- AT-04
Suppression not applied
Someone excluded by name is targeted anyway.
- AT-05
Match rate not verified
An inflated overlap hides a broken join.
Stage pushes
Only the definition and platform a person approved
04 · Delivery
1 mode
- AT-06
Audience collapses at delivery
The approved segment reaches almost nobody.
Stage reviews
What the platform delivered against what was intended
05 · Change / Version
1 mode
- AT-07
Lookalike drifts from seed
A refresh moves the model off the seed.
Stage tracks
Model, rule, seed and suppression-list changes
- Sev-1 · the audience should not run at all
- Sev-2 · the wrong people are reached or missed
- Sev-3 · the segment is unusable as built
Affected slices
Most of the exposure sits in a few kinds of brief
An aggregate failure rate can look acceptable while a small number of brief types carry nearly all of the proxy findings, the permission gaps and the audiences pulled after sign-off. 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 |
|---|---|---|---|---|
| Housing, employment and credit briefs | 6.2% | 3.9× | Review | |
| Onboarded partner and second-party data | 4.3% | 2.7× | Review | |
| Lookalikes built from small seeds | 2.9% | 1.8× | Watch | |
| Repeat first-party refreshes | 1.0% | 0.6× | Normal |
Bar: failure-rate lift vs. first-party refresh baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
An approved audience does not stay approved
Consent is withdrawn, opt-outs arrive and platforms change what a category may target. A segment nobody re-checks drifts out of permission while it runs.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
A proxy finding, a permission gap or a delivery skew shows up in one kind of brief.
- 02
Diagnose
Traced to a source field, a rule, the seed, the suppression list or a platform policy change.
- 03
Improve
The field, rule or list is changed, re-cleared by the named approver and version-linked.
- 04
Verify
Re-run against held-out briefs, including the audiences that were pulled after sign-off.
- 05
Learn
The pulled audience becomes a regression case and the rule enters the targeting standard.
Learn → Detect
The return edge. Every cycle also re-reads consent, opt-outs and the platforms' category rules — an audience cleared last quarter is not cleared 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, permission mapping, rules and proxy testing, evaluation, activation, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Workflow discovery and automation-boundary definition.Week 1
- 02CDP, warehouse and source assessment.Week 1, Week 2
- 03Permitted-use, consent and contract mapping.Week 1, Week 2
- 04Restricted-category rules and named approvers.Week 2
- 05Segment rule drafting and sizing logic.Week 2, Week 3
- 06Proxy audit and protected-attribute testing.Week 3
- 07Suppression, opt-out and preference wiring.Week 3, Week 4
- 08Approval workflow and sign-off record.Week 4
- 09Evaluation suite, slices and regression briefs.Week 4, Week 5
- 10Platform audience APIs and clean-room joins.Week 5
- 11Match-rate and composition reporting.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 platform, one segment set | ProductionProduction CDP integration | AdvancedMulti-platform / multi-brand |
|---|---|---|---|
| Introduced at Pilot | |||
| Segment definition and sizing | ✓ | ✓ | ✓ |
| Permitted-use and consent checks | ✓ | ✓ | ✓ |
| Restricted-category flagging | ✓ | ✓ | ✓ |
| Named-approver sign-off | ✓ | ✓ | ✓ |
| Baseline evaluation | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Proxy audit on every rule | — | ✓ | ✓ |
| Suppression and opt-out enforcement | — | ✓ | ✓ |
| Platform audience APIs — approver pushes | — | ✓ | ✓ |
| Match-rate and composition reporting | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Clean-room joins and identity resolution | — | — | ✓ |
| Multi-brand and enterprise controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on platforms and data sources in scope, consent and preference systems, clean-room and identity integrations, segment 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
01The platforms and segment types you want in scope
→Platform audience APIs and activation scoping
Week 102Your data sources and what each was collected for
→Permitted-use, consent and contract mapping
Week 103Access to the CDP, warehouse and consent platform
→Source assessment, consent and preference wiring
Week 204Suppression lists, opt-outs and do-not-target rules
→Suppression, opt-out and preference enforcement
Week 305The proxies and protected attributes you already watch
→Proxy audit and protected-attribute testing
Week 306Segments you have had to pull, and the reason
→Evaluation suite, regression briefs and slice reporting
Week 407Named approvers for legal basis and restricted briefs
→Approval workflow, then activation under sign-off
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 the first audiences pushed under sign-off.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Platforms and segment types in scope; who clears a restricted brief
W2Source assessment, permitted-use mapping and consent wiring
W3Segment rules, proxy audit and suppression enforcement
W4Approval workflow, evaluation suite and slice reporting
W5Platform APIs, first audiences pushed under sign-off
W6Live audiences under approval, then Agent Care monitoring
- Reading the band
- The proxy audit and the suppression wiring land in week 3, before any audience can be pushed in week 5. The bars show that dependency, not a smooth ramp.
- At the end of W6
- Audiences have gone live under sign-off with the permission behind every field on file, 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 an audience-targeting copilot around your data stack.
Show us the segments you build most often, where their data comes from and what each source was collected for, and who signs off a restricted brief today. We'll rebuild one live audience from its brief, show you the permission behind every field and the proxies it would have carried, then scope from there.
- Typical delivery
- 4–6 weeks
- Deployment
- Your data and ad stack
- Activation
- Named human sign-off
- After launch
- Agent Care available
Nestack Agents · Audience targeting & suppression
AGT-AM-04 · Agent Care available after launch