Marketing AI agent · campaign analytics
Campaign Analytics AI Agent
Fix each reported figure to the method, window, attribution setting and extraction time that produced it, hold the residual against the warehouse of record, and announce a restatement rather than smoothing it.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Named analyst
- Method-bound
- After launch
- Agent Care
What this agent does
Reconciles the figure, never signs it off
In
- 01
A figure is pulled, and the method, the window and the extraction time are pulled with it.
- 02
A platform backfills overnight, and yesterday's number is not what the same query returns today.
Reason
- 03
A method changes, and the series before that change is never quietly joined to the series after.
- 04
A window shifts from click to view, and the figure moves for a reason somebody has to write down.
- 05
A platform total and the warehouse disagree, and the residual is reported rather than chosen between.
Decide
- 06
A modelled conversion arrives, and it is labelled modelled wherever the number is later quoted.
- 07
A period locks in finance, and what follows the lock is a restatement rather than a quiet edit.
Out
- 08
A figure is restated, and the movement is dated, explained and kept beside the number it replaced.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent assembles, reconciles, labels and flags. A named analyst signs the figure off, and finance owns the close.
Example workflow
One figure, extract to sign-off
- 1Figure evidence receivedPlatform exports, warehouse tables, tagging feeds or finance postings
- 2Figure context assembledThe campaign, the method behind the number, the window it covers and when it was extracted
- 3Reconciliation draftedThe figure, its sources, the residual against the warehouse and completeness
- 4Controls appliedMethod checks, window checks, residual thresholds and completeness confidence
No human action required
Stages 1 to 4 run unaided, and nothing is signed off at any of them — the agent is reconciling, and the analyst lane opens at the completeness gate.
Goes to the named analyst to sign off.
Adds a second analyst read first.
Analyst review
The number is held with its method, its residual and the window it covers.
Sign off · Append evidence · Send to analyst review
Signed off — by a named analyst
- 6Reporting and warehouse records updatedOnly where write access and records policy allow it
- 7Outcome evaluatedResidual size, method coverage, analyst corrections and what the read found
Corrections
Each analyst correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Signing a figure off for a board pack.
- Closing or reopening a finance period.
- Deciding which attribution model is right.
- Restating a number after the period locks.
Automation boundary
Agent acts unaided
- ✓Hold the method, the window and the extraction time beside the figure.
- ✓Reconcile platform totals against the warehouse.
- ✓Label a modelled conversion as modelled wherever that number travels.
- ✓Flag the number that moved between two decks without.
Nothing is signed off or booked except by a named person, inside the agreed boundaries.
- Judging whether a residual is acceptable.
- Telling the board what a campaign returned.
- Setting the attribution window we report on.
- Changes to warehouse tables or finance postings.
Example output
One figure, annotated
Nothing here is a fresh measurement; this record is what one campaign number carried on the day it reached the warehouse.
Reconciliation entry · single figure
Illustrative example- Figure
- Paid social conversions, campaign level
- Recorded as
- Reconciled to the warehouse, residual open
- Method
- Modelled
- Evidence of record
- Warehouse extract, 3 August 2026
- Confidence
- Held unsigned
- Held for
- The reporting analyst, by name
As received
Taken from the platform export and the warehouse table — it reaches as far as those sources do.
What the record holds
- Extraction time
- Attribution window
- Warehouse residual
Why no sign-off here
Whether a residual may stand is an analyst judgement, not a model output.
What the score decides
Below the configured threshold a number picks up a second read before the analyst sees it.
Action
Sign offAppend evidenceSend to analyst reviewValue
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 figure
From the source that made it
Where the evidence is used
Our advertising analytics agents answer to a client report that closes with the campaign; this one answers to the warehouse of record and a finance period that locks behind it.
A number needs its method
A figure without its method is a rumour: the window, the attribution setting and the extraction time travel beside it, or it does not travel at all.
What was checked, and not found
No platform publishes a reconciliation of its own reported totals against advertiser warehouse data, no shared definition of an attributed conversion holds across platforms, and no vendor figure has been borrowed for this page.
Build an evidence trail
The figure, the method that produced it and the analyst who signed off stay together.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Platform and channel sources
- Ad platforms
- web analytics
- Delivery and conversion exports
Warehouse of record
- BigQuery
- Snowflake
- Observed and modelled tables
Finance and planning
- ERP
- budget systems
- Booked revenue and period status
Agent
Campaign analytics reconciliation
- Reads the figures
- Reconciles the sources
- Holds for the analyst
Reporting and case systems
- BI tooling
- ticketing
- Restatement 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 gauges between the model and the analyst
Six gauges read in order, the last the finest. Whatever holds steady is set out in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a number may stand
- L3
- Puts the sign-off in a person's hands
- L4 – L5
- Keep the figure and the method behind it
- L6
- Reverts to raw extracts when signals degrade
How Nestack evaluates it
Evaluate the whole assembly — not only the number that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the figure a board pack quotes
- E1Final-output evaluationDid the entry record the method and window that produced it?
- E2Step-level evaluationDid the agent read the right table, the right window and the live extract?
- E3Tool evaluationDid it read and write the correct campaign record and the correct figure?
- E4Confidence calibrationDo low-confidence figures actually attract more analyst corrections?
- E5Slice evaluationHow does performance change across specific campaign types?
- E6Business outcomeHow many numbers needed a correction before the analyst signed off?
Floor — the numbers the business books
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 →
01 · Retrieval
1 mode
- LJ-03
Stale extract read
The extract read predates the backfill now sitting in the table.
Stage gathers
The figures, the methods, the windows and the sources
02 · Reasoning
2 modes
- LJ-04
Method dropped in transit
A window is lost when the number is copied across.
- LJ-06
Modelled shown as observed
A modelled conversion is carried as an observed one.
Stage proposes
The figures, their residuals and completeness
03 · Tool / write
2 modes
- LJ-02
Thin reconciliation forwarded
It reaches the analyst ahead of its second read.
- LJ-05
Bound to the wrong campaign
A residual is filed against a campaign that never carried it.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- LJ-01
Signed off, method unrecorded
The record shows a sign-off but not what produced the number.
Stage returns
The figure an analyst signs and a board pack quotes
05 · Change / Version
1 mode
- LJ-07
Silent restatement
A refit moves the number and the record keeps the old story.
Stage tracks
Model, prompt, method rules and figure fields
- Sev-1 · a restated figure shipped unannounced
- Sev-2 · wrong number reaches the board pack
- Sev-3 · source degrades, figure held unsigned
Affected slices
Paid social absorbs the corrections
A campaign-level reconciliation figure can read clean while paid social carries most of the rework. Nestack reports the correction rate by campaign type, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Paid social campaigns | 9.5% | 3.6× | Review | |
| Retail media placements | 6.8% | 2.6× | Review | |
| Programmatic display | 4.2% | 1.6× | Watch | |
| Search and brand campaigns | 2.3% | 0.9× | Normal |
Bar: correction-rate lift vs. search and brand baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What a silent restatement costs
A loop ends when the silently restated metric has become a case the next release must pass. That suite is what the next report issued is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Correction rate rises on paid social campaigns.
- 02
Diagnose
The number that changed between two decks and never said it had is read back until one cause remains.
- 03
Improve
Number the change; the figures that drove it are filed underneath it.
- 04
Verify
Each touched figure case is run once more, and one red holds it back.
- 05
Learn
It stays on as a standing test, and the restatement rules move with it.
Learn → Detect
The return edge. The next report 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, reconciliation, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Method discovery and automation-boundary scoping.Week 1
- 02Platform, warehouse and finance sources.Week 1, Week 2
- 03Figure-to-method and residual-threshold mapping.Week 1, Week 2
- 04Figure and extract ingestion.Week 2
- 05Campaign, method and window binding.Week 2, Week 3
- 06Completeness scoring and review routing.Week 3
- 07Analyst sign-off workflow.Week 3, Week 4
- 08Warehouse and finance integration.Week 5
- 09Reconciliation and restatement cases.Week 4, Week 5
- 10Guardrails and sign-off controls.Week 4
- 11Figure-trail instrumentation.Week 5, Week 6
- 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 tier | PilotOne campaign type, one quarter | ProductionProduction reporting workflow | AdvancedMultiple brands / markets |
|---|---|---|---|
| Introduced at Pilot | |||
| Reconciliation to your warehouse | ✓ | ✓ | ✓ |
| Named analyst sign-off | ✓ | ✓ | ✓ |
| Reportable-metric baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by campaign | — | ✓ | ✓ |
| Sign-off workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Warehouse-and-platform integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-brand estates | — | — | ✓ |
| Cross-market figure packs | — | — | ✓ |
| Large campaign estates | — | — | ✓ |
| Multi-source reconciliation controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, reporting 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
01Your campaigns and the platforms behind them
→Source mapping and figure capture
Week 102Representative exports, warehouse tables and postings
→Record binding, method logic and the reconciliation baseline
Week 203Your method, window and labelling rules
→Figure mapping, method binding and the automation boundary
Week 104Access to relevant APIs, feeds or exports
→Platform, warehouse and finance-source assessment, then integration setup
Week 205Reports you would not want defended
→Restatement cases and failure testing
Week 406What no attribution model may prove
→Completeness scoring, review routing, guardrails and release controls
Week 307A named analyst to sign the figure off
→Sign-off 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
Read the widths as weeks truly worked rather than as layout, which forces two phases into the fifth.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Reporting workflow discovery, method mapping and the automation boundary
W2Source integration and the reconciliation baseline
W3Figure assembly, method logic and sign-off controls
W4Evaluation suite, restatement cases and failure-mode testing
W5Warehouse integration, pilot figures and targeted corrections
W6One reporting year run under the analytics lead, then Agent Care handover
- Reading the band
- A bar spans only the weeks its own work is named in. Two phases land in the fifth because that is where they fall.
- At the end of W6
- Validation closes on live reports, 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 campaign analytics agent around the number your board pack has to defend.
Bring us one figure and both decks it appeared in. If your platform exports and your warehouse tables are squared by hand once the period closes, then that residual is carried on trust. Attribution and media-mix modelling stay with your analysts; the record is ours.
- Typical delivery
- 4–6 weeks
- Deployment
- Your warehouse
- Named analyst
- Method-bound
- After launch
- Agent Care available
Nestack Agents · campaign analytics
AGT-MK-07 · Agent Care available after launch