Main site

Marketing AI agent · risk assessments

Marketing Risk Assessment AI Agent

Catalogue every activity that sells or shares personal information for cross-context behavioural advertising, assess each one before it starts under § 7150, and hold the summary a named executive attests to.

Typical delivery
4–6 weeks
Deployment
Your stack
Named executive
One attestation
After launch
Agent Care

What this agent does

Assembles the register, never the attestation

In

  1. 01

    A tag is added, and § 7150 wants the assessment finished before the processing starts.

  2. 02

    An audience is exported, and selling or sharing for behavioural advertising is the trigger.

Reason

  1. 03

    An SDK ships inside an app, and the assessment weighs purpose against harm and safeguards.

  2. 04

    A CTV integration goes live, and the register has to show what leaves and where it lands.

  3. 05

    A conversions API is wired, and the opt-out has to reach it, not only your own ad platform.

Decide

  1. 06

    An activity changes materially, and the re-review clock runs from the change itself.

  2. 07

    A pixel predates the register, and no assessment can be conducted after the fact it was owed before.

Out

  1. 08

    A reporting period closes, and § 7157 wants the summary with the CPPA by 1 April 2028.

  2. 09

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

Product statement

The agent maintains the register and drafts the assessments. A named business executive attests to the § 7157 submission under penalty of perjury.

Example workflow

One activity, register to attestation

  1. 1Activity evidence receivedTag manager exports, SDK manifests, CTV integrations or audience destinations
  2. 2Activity context assembledThe activity, whether it sells or shares, the purpose it serves and the day it went live
  3. 3Assessment evidence draftedThe activity, its purpose, the negative impacts, the safeguards and completeness
  4. 4Controls appliedTrigger checks, pre-processing date checks, re-review clock checks and completeness confidence

No human action required

Stages 1 to 4 run unaided, and nothing is attested at any of them — the agent is assembling, and the privacy lane opens at the completeness gate.

5DecisionSplits at the completeness gate
Evidence sufficient

Goes to the attesting executive to approve.

Anything thin

Adds a privacy counsel read first.

Privacy review

The assessment is held with its activity, its purpose and the safeguards behind it.

Attest · Append evidence · Send to privacy review

Attested — by a named executive

  1. 6Register and activity records updatedOnly where write access and records policy allow it
  2. 7Outcome evaluatedTrigger accuracy, assessment coverage, reviewer corrections and what the read found

Corrections

Each privacy correction is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Attesting to the § 7157 summary.
    • Deciding an activity needs no assessment.
    • Certifying the cybersecurity audit under § 7124.
    • Telling the CPPA an activity was lawful.

  • Automation boundary

    Agent acts unaided

    • ✓Maintain the register of every sell-or-share activity.
    • ✓Run the § 7150 assessment before processing begins.
    • ✓Track the re-review clock that a material change starts running.
    • ✓Flag the activity whose stated purpose the record cannot evidence.

    Nothing is attested or filed except by a named person, inside the agreed boundaries.

    • Judging whether an assessment may be attested.
    • Telling the CPPA the register is complete.
    • Setting the lawful purpose an activity runs on.
  • Changes to tags, audiences or destination records.

Example output

One activity, annotated

A marketing department with a direct relationship to its customers is not a data broker, and a sale is still a sale; this record is what one activity carried into the register.

Register entry · single activity

Illustrative example
Activity
Audience export, § 7150
Recorded as
Assessed before launch, safeguards recorded
Trigger
Sell or share
Evidence of record
Register entry, 3 August 2026
Confidence
Held unattested
Held for
The attesting executive, by name

As received

Taken from the tag manager export and the destination record — it reaches as far as those sources do.

What the record holds

  • Tag manager export
  • Destination record
  • Opt-out signal log

Why no attestation here

Whether an assessment may be attested is reserved to a named executive.

What the score decides

Below the configured threshold an entry picks up a privacy read before the executive sees it.

Action

AttestAppend evidenceSend to privacy 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 activity

From the system that runs it

03Evidence

Where the evidence is used

Our audience targeting copilot yields an audience — internal, continuous and per-segment; this one yields a dated external filing with a named signer, working on processing activities rather than people.

01Approved path

The register is the assessment

An assessment attaches to a processing activity, not to a segment, and § 7150 wants it settled before that activity begins.

02Human review

What was checked, and not found

The verbatim § 7157 wording and the title it demands of its signer, the revenue thresholds deciding who is caught, a retention period put at five years by some sources and three by others, the tally of comprehensive state privacy laws, the states whose statutes compel a browser opt-out signal and the per-day penalty sums were each checked against the CPPA consolidated text, which truncated on every fetch, and none was confirmed.

04

Build an evidence trail

The activity, the purpose it serves and the executive who attested stay on the record.

Integrations

Typical integrations

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

  • Tag and pixel sources

    • Tag manager
    • containers
    • Site and app SDKs

  • Ad platforms and destinations

    • Conversions APIs
    • CTV
    • Audience export records

  • Consent and opt-out

    • Consent platform
    • GPC log
    • Preference and signal records

Agent

Marketing risk assessment

  • Reads the activities
  • Assembles the register
  • Holds for the executive

  • Records and case systems

    • Privacy tooling
    • ticketing
    • Assessment and review records

  • Observability & evaluation

    • OpenTelemetry
    • Langfuse
    • Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six gates between the model and the executive

Six gates set in order, the narrowest last. Whatever clears them all is drawn in the map below.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeNarrow the agent to evidence assembly when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt and trigger rules; on 3 December 2025 the CPPA fined ROR Partners for selling custom audiences while unregistered, which is why every destination is named.Track
L4TraceabilityRecord each activity, the assessment behind it, the register it feeds and every read of the file.Record
L3Executive releaseHold the summary for the attesting executive; the hold governs release, not whether an activity should ever have started.Gate
L2Scope guardrailsTest the register against 11 CCR §§ 7150–7157 as configured; behavioural advertising was cut from the final ADMT scope, which reaches significant decisions in finance, housing, education, employment and healthcare alone.Restrict
L1Confidence thresholdsRoute a thin assessment to a privacy read first; the PlayOn order of 3 March 2026 shows liability arising out of a single targeted campaign.Require review
Model coreEvidence assembled — the activities, their purposes, the safeguards and completeness
L1 – L2
Test whether an assessment may stand
L3
Puts the attestation in a person's hands
L4 – L5
Keep the activity and the purpose behind it
L6
Holds the assessment undrafted when signals degrade

How Nestack evaluates it

Evaluate the whole assembly — not only the register entry that comes out.

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

Surface — the summary a regulator reads

  1. E1Final-output evaluationDid the entry record what the activity actually does?
  2. E2Step-level evaluationDid the agent read the right activity, the right purpose and the live destination?
  3. E3Tool evaluationDid it read and write the correct activity and the correct assessment?
  4. E4Confidence calibrationDo low-confidence assessments actually attract more privacy corrections?
  5. E5Slice evaluationHow does performance change across specific activity types?
  6. E6Business outcomeHow many entries needed a correction before the executive attested?

Floor — the register the business answers for

Failure modes

Where each failure originates in the agent

Seven failure modes, each placed at the stage where it first shows.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • KD-03

      Stale activity read

      The destination read is not the one now in service.

    Stage gathers

    The tags, the purpose, the dates and the register

  2. 02 · Reasoning

    2 modes

    • KD-04

      Trigger missed on launch

      An activity starts before its assessment exists.

    • KD-06

      ADMT scope read too widely

      Ad targeting is worked as a significant decision.

    Stage proposes

    Activities, their purposes and completeness

  3. 03 · Tool / write

    2 modes

    • KD-02

      Thin assessment passed forward

      An entry moves on without the privacy read.

    • KD-05

      Bound to the wrong activity

      An assessment is filed against the wrong tag.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • KD-01

      Attested, evidence unrecorded

      The record shows an attestation but not what supported it.

    Stage returns

    The summary an executive attests and a regulator reads

  5. 05 · Change / Version

    1 mode

    • KD-07

      Silent trigger regression

      A configuration change moves the trigger, not the register.

    Stage tracks

    Model, prompt, trigger rules and register fields

  • Sev-1 · an activity live with no assessment
  • Sev-2 · wrong evidence reaches the register
  • Sev-3 · source degrades, entry held unattested
See the complete Marketing failure-mode catalogue

Affected slices

Third-party tags absorb the corrections

An activity-level assessment-coverage figure can read clean while third-party tags and SDKs carry most of the rework. Nestack reports the correction rate by activity type, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Third-party tags and SDKs10.4%3.6×Review
CTV and in-app integrations7.4%2.6×Review
Conversions APIs and server feeds4.6%1.6×Watch
First-party owned channels2.2%0.8×Normal

Bar: correction-rate lift vs. first-party channel baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

What an unassessed activity costs

A cycle shuts when the unassessed pixel is a regression case. That suite is what the next register built is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Correction rate rises on third-party tags and SDKs.

  2. 02

    Diagnose

    The pixel a campaign team added in an afternoon, three years before anyone had to describe it, is read back until one cause remains.

  3. 03

    Improve

    The change ships numbered, and the activities that forced it ride with it.

  4. 04

    Verify

    Nothing releases while one touched activity case is still red.

  5. 05

    Learn

    It is retained for good, and the assessment rules are amended in that same commit.

  6. Learn → Detect

    The return edge. The next cycle is measured against a suite one case longer.

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, register assembly, evaluation, integration, then production validation and handover.

Workstream

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

  1. 01Sell-or-share trigger discovery and boundary work.Week 1
  2. 02Tag, SDK and destination sources.Week 1, Week 2
  3. 03Activity-to-assessment and re-review clock mapping.Week 1, Week 2
  4. 04Activity and assessment ingestion.Week 2
  5. 05Activity, purpose and record binding.Week 2, Week 3
  6. 06Completeness scoring and review routing.Week 3
  7. 07Executive attestation workflow.Week 3, Week 4
  8. 08Tag and destination-system integration.Week 5
  9. 09Trigger and materiality cases.Week 4, Week 5
  10. 10Guardrails and attestation controls.Week 4
  11. 11Activity-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 activity type, one cycleProductionProduction assessment workflowAdvancedMultiple brands / entities
Introduced at Pilot
Register assembly to your activities✓✓✓
Attesting executive release✓✓✓
Sell-and-share inventory baseline✓✓✓
Introduced at Production
Reporting by processing activity—✓✓
Attestation workflow in your systems—✓✓
Approved write-back—✓✓
Tag-and-SDK integration—✓✓
Introduced at Advanced
Multi-brand registers——✓
Cross-period evidence packs——✓
Large tag estates——✓
Multi-clock retention controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on integrations, workflow complexity, activity 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 activities and the destinations each one feeds

    Activity inventory mapping and evidence capture

    Week 1

  2. 02Representative tag, SDK and destination records

    Record binding, trigger logic and the register baseline

    Week 2

  3. 03Your purposes and the safeguards behind them

    Activity mapping, trigger binding and the automation boundary

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Tag, SDK and destination-source assessment, then integration setup

    Week 2

  5. 05Registers you would not want examined

    Trigger cases and failure-mode testing

    Week 4

  6. 06What no risk assessment may establish

    Completeness scoring, review routing, guardrails and release controls

    Week 3

  7. 07A named executive to attest

    Attestation 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

Every band below is worked time rather than drawn space, which is why the fifth of them holds two.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Privacy workflow discovery, activity mapping and the automation boundary

W2Source integration and the assessment-coverage baseline

W3Register assembly, trigger logic and release controls

W4Evaluation suite, trigger cases and failure-mode testing

W5Destination integration, pilot activities and targeted corrections

W6One assessment cycle run under the privacy lead, then Agent Care handover

Reading the band
Each bar covers only the weeks its own work is named for. The fifth takes a second because the work does.
At the end of W6
Validation closes on live registers, and Agent Care picks up the watch.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Marketing AI agent

Build a risk assessment agent around the activity running today that a 1 April 2028 filing has to count.

Show us one activity and the record behind it. If your opt-out is honoured per device rather than per account, then that is the fact pattern the Disney settlement of 11 February 2026 turned on. Cybersecurity audits sit with the CISO and are a different lane.

Typical delivery
4–6 weeks
Deployment
Your activities
Named executive
One attestation
After launch
Agent Care available

Nestack Agents · risk assessments

AGT-MK-01 · Agent Care available after launch

Build a risk assessment agent

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

About you
Your activities

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