Main site

Advertising AI agent · Attention measurement

Attention-Measurement AI Agent

Pull attention scores from the vendors you use, tag each with the method that produced it, apply them inside the rules your team set, and check whether the outcome you're optimizing for actually moved.

Typical delivery
4–6 weeks
Deployment
Your stack
Planner review
Pre-adjustment
After launch
Agent Care

What this agent does

Reads the score, not the decision to reweight the plan

In

  1. 01

    A score arrives from a vendor, carrying the method that produced it — panel, camera, survey or model.

  2. 02

    Each score is tagged to its measurement method before anything else runs against it.

Reason

  1. 03

    A plan is built inside rules your planning lead set, and a score can move it only within them.

  2. 04

    Scores from different methods are held apart on the record, never averaged into one figure.

  3. 05

    Prior plan history and each vendor's accreditation status inform the check, dated and named.

Decide

  1. 06

    A cross-method comparison or an overstated accreditation claim is flagged before it reaches the plan.

  2. 07

    Every rule-triggered plan change is routed to the planning lead, never applied unattended.

Out

  1. 08

    The score, its method, the rule applied and the planning lead's decision are kept against the plan.

  2. 09

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

Product statement

The agent applies scores inside the rules the planning lead set — never blending methods, never touching spend on its own; the planning lead decides when a rule changes, and the outcome, not the score, is what the plan answers for.

Example workflow

One score, method to decision

  1. 1Score receivedA vendor score feed, a plan build, an accreditation update or a renewal check
  2. 2Method assembledScore, source method, vendor accreditation status and prior plan history, each with its source
  3. 3Match proposedMatched rule, method tag, flagged comparisons and confidence
  4. 4Controls appliedMethod checks, accreditation checks, cross-vendor comparison flags and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing changes the plan at any of them — the agent matches, and the planning lead's lane opens at the confidence gate.

5DecisionBranches at the confidence threshold
High confidence

Goes to the planning lead to approve.

Low confidence

Adds a second method read first.

Planner review

The match is held with its method tag, its flagged comparisons and the confidence.

Approve · Edit · Escalate

Approved — applied to the plan

  1. 6Planning systems updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedEdit distance, flagged-comparison outcomes, plan acceptance and post-flight corrections

Edits

Every planning-lead edit is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Buying media or committing spend on an attention score.
    • Reweighting a media plan on its own.
    • Converting one vendor's score into another vendor's scale.
    • Reporting an attention score as a campaign outcome.

  • Automation boundary

    Agent acts unaided

    • ✓Read each vendor's score against the method that produced it.
    • ✓Hold scores from different methods apart, never merged.
    • ✓Apply scores inside the planning rules configured for your team.
    • ✓Flag a cross-vendor comparison or an outcome.

    Any write happens inside the boundaries agreed at implementation, never ahead of approval.

    • Averaging or blending scores from different measurement methods.
    • Confirming a vendor's current MRC accreditation status.
    • Deciding that attention predicts a sales or conversion result.
  • Changes to method rules, accreditation status or approval thresholds.

Example output

One score, annotated

Everything the agent applies is attached to the score it was drawn from.

Attention-measurement output · single score

Illustrative example
Placement
Mid-flight video buy
Score
Modelled score reported for a non-panel CTV placement
Method
Signal-based (modelled)
Evidence source
This week's vendor feed
Confidence
81%
Attribution
Planning-lead name on file

As received

Taken from the vendor's feed and its published method note — nothing on this side is converted by the agent.

Evidence checked

  • Vendor score feed
  • Method taxonomy tag
  • Accreditation record

Why it's flagged

This score comes from a modelled, signal-based method.

What the score decides

Below the configured threshold the match picks up a second method read before it.

Action

ApproveEditEscalate

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 score

From the vendor feed

03Matching

Match every score to its method

Match each vendor's score to its stated method and the rule configured for it.

01Approved path

Two scores, two different things

A panel eye-tracking score and a modelled signal-based score can share a scale and still measure different things — nothing in either file makes them the same unit.

02Human review

Send review to the flagged

Cross-method comparisons and accreditation gaps are marked, so the planning lead's read starts where the risk concentrates.

04

Build an evidence trail

The score, the method it came from and the planner who applied it stay on the buy.

Integrations

Typical integrations

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

  • Attention vendors

    • Adelaide
    • Lumen
    • TVision
    • DoubleVerify

  • Delivery & verification

    • IAS
    • MOAT
    • Campaign Manager 360

  • Media planning & buying

    • MediaOcean
    • Prisma
    • DSP reporting exports

Agent

Attention measurement

  • Reads the score
  • Matches the method
  • Holds for the planner

  • Reporting & workflow

    • BI dashboards
    • Looker
    • Slack
    • email escalation

  • 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 plan

The layers wrap one another. What gets past them all is set out in the map below.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeFall back to viewability-only reporting when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, method-taxonomy and accreditation-status changes.Track
L4TraceabilityRecord the score, the method, the rule applied and the planner's decision.Record
L3Planner approvalHold a rule-triggered change for the planner; it governs application, not whether the score is right.Gate
L2Policy guardrailsTest changes against configured planning rules and method limits; a failure returns the match.Restrict
L1Confidence thresholdsRoute low-confidence matches to a second method read before the planner sees them.Require review
Model coreMatch proposed — score, method tag, flagged comparisons and confidence
L1 – L2
Test whether a match may stand
L3
Puts the change in the planner's hands
L4 – L5
Keep the score and the method behind it
L6
Holds at viewability reporting when signals degrade

How Nestack evaluates it

Evaluate the matching workflow — not only whether attention rose.

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

Surface — the plan the client sees

  1. E1Final-output evaluationDid the matched method reflect how the vendor actually measured the score?
  2. E2Step-level evaluationDid the agent use the current method taxonomy and accreditation record?
  3. E3Tool evaluationDid it read the correct vendor feed and the correct plan line?
  4. E4Confidence calibrationDo low-confidence matches actually attract more planning-lead escalations?
  5. E5Slice evaluationHow does match accuracy change across specific format types?
  6. E6Business outcomeHow many matches needed a planning-lead correction after the plan ran?

Floor — the outcome the planner answers for

Failure modes

Where each failure originates in the agent

Seven failure modes, placed at the stage each one originates.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • TN-03

      Stale accreditation status

      A vendor's MRC status is read from an earlier quarter.

    Stage gathers

    Vendor score feed, method taxonomy and plan rules

  2. 02 · Reasoning

    2 modes

    • TN-04

      Cross-method averaging

      Two vendors' scores, from different methods, become one blended number.

    • TN-06

      Accreditation overstated

      A pre-2025 grant is read as proof the current guideline is met.

    Stage proposes

    Method match, rule applied and confidence

  3. 03 · Tool / write

    2 modes

    • TN-02

      Auto-shift on attention alone

      Budget moves to a higher score before the outcome is checked.

    • TN-05

      Attention shown as an outcome

      A score appears in the outcome line, next to conversion and ROAS.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • TN-01

      Method attribution dropped

      The score reaches the plan with no method tag attached.

    Stage returns

    The score and method the planner applies to the plan

  5. 05 · Change / Version

    1 mode

    • TN-07

      Silent panel drift

      A vendor's panel or model changes without a break flagged.

    Stage tracks

    Model, prompt, method rules and accreditation

  • Sev-1 · match applied outside the boundary
  • Sev-2 · a mismatched score reaches the plan
  • Sev-3 · method degrades, routes to review
See the complete Advertising failure-mode catalogue

Affected slices

A portfolio average can hide a bad match

A single attention index can look steady across a plan while a handful of placements carry scores that were never comparable to begin with. Nestack reports the comparability-defect rate by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Scores mixed across two vendors7.4%3.8×Review
Modelled scores on non-panel inventory5.6%2.9×Review
Cross-format plans3.5%1.8×Watch
Single-vendor, single-format plans1.9%0.8×Normal

Bar: comparability-defect rate lift vs. the single-vendor baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

The loop closes on the method, not the score

A cycle shuts when the mixed method is a case the next release has to catch. That suite is what the next plan scored is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Comparability-defect rate rises in a plan slice.

  2. 02

    Diagnose

    The two scores that agreed on nothing are checked against the method each came from and the plan they were applied to until one cause explains it.

  3. 03

    Improve

    The change ships against a version, with the buys that exposed it attached.

  4. 04

    Verify

    A failing comparability case holds the release until it clears.

  5. 05

    Learn

    The case joins the standing suite and the method rules move with it.

  6. Learn → Detect

    The return edge. Detection next time runs 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, matching workflow, evaluation, integration, then production validation and handover.

Workstream

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

  1. 01Planning discovery and boundary scoping and boundary definition.Week 1
  2. 02Vendor feed and ad-server assessment.Week 1, Week 2
  3. 03Method taxonomy and rule mapping and it is logged..Week 1, Week 2
  4. 04Score intake and method normalisation.Week 2
  5. 05Matching logic and rule binding.Week 2, Week 3
  6. 06Confidence scoring and flag routing.Week 3
  7. 07Planning-lead approval workflow.Week 3, Week 4
  8. 08Ad-server and DSP integration.Week 5
  9. 09Comparability regression cases.Week 4, Week 5
  10. 10Guardrails and application controls.Week 4
  11. 11Buy-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 vendor, one formatProductionProduction vendor feedsAdvancedMultiple vendors / formats
Introduced at Pilot
Matching to your methods and rules✓✓✓
Planner approval✓✓✓
Comparability baseline✓✓✓
Introduced at Production
Reporting by format—✓✓
Escalation workflow in your tools—✓✓
Approved rule write-back—✓✓
Measurement-vendor integration—✓✓
Introduced at Advanced
Multi-method planning rules——✓
Multi-stage planner approvals——✓
High plan volume——✓
Multi-vendor score 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. 01Your attention vendor list and feed formats

    Vendor ingestion and method mapping

    Week 1

  2. 02Representative scored plans

    Matching baseline and method-tag binding

    Week 2

  3. 03Your planning rules and approval thresholds

    Planning-rule and approval-boundary mapping

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Vendor feed and ad-server assessment, then integration setup

    Week 2

  5. 05Scores you would not want optimised to

    Method cases and the evaluation suite

    Week 4

  6. 06What no score may stand in for

    Confidence scoring, flag routing, guardrails and approval controls

    Week 3

  7. 07Named planning leads to review matches

    Planner approval 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 sits on the weeks the work occupies, so week 5 runs evaluation and pilot together.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Planning discovery, method mapping and the automation boundary

W2Vendor integration and the matching baseline

W3Matching workflow, confidence logic and approval controls

W4Evaluation suite, guardrails and failure-mode testing

W5Ad-server integration, pilot plans and targeted corrections

W6One planning cycle scored under the planning lead, then Agent Care handover

Reading the band
A bar covers only the weeks its work is named in — the week 5 overlap is real, not padding.
At the end of W6
The final checks clear on live buys and monitoring moves to Agent Care.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Advertising AI agent

Build an attention-measurement agent around your planning workflow.

Show us the attention vendors you use, how scores feed your planning, and who signs off on a plan change. You'll get every score tied to its method, applied inside your rules, and checked against the outcome it's meant to predict.

Typical delivery
4–6 weeks
Deployment
Your attention vendors
Planner review
Pre-adjustment
After launch
Agent Care available

Nestack Agents · Attention measurement

AGT-AM-20 · Agent Care available after launch

Build an attention-measurement agent

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

About you
Your plans

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