Advertising AI agent · Performance reporting
Campaign Performance-Reporting AI Agent
Pull delivery and conversion data from every platform, reconcile it against the warehouse, draft the commentary and flag what looks wrong — an analyst signs the report and owns every causal claim in it.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Report sign-off
- Named analyst
- After launch
- Agent Care
What this agent does
Assembles the report, a person explains it
In
- 01
Pull spend, delivery and conversion data from each platform, measurement source and the warehouse.
- 02
Record the attribution setting, lookback window, currency and time zone each figure arrived under.
Reason
- 03
Reconcile platform figures against the warehouse and size every difference rather than averaging it away.
- 04
Deduplicate conversions more than one platform has claimed, by the rule agreed with the client.
- 05
Draft commentary on what moved, against which comparison — as movement, not as cause.
Decide
- 06
Hold any period still inside its restatement window, or where a definition or window has changed.
- 07
Flag anomalies for a person and route the whole report to the analyst who signs it.
Out
- 08
Deliver to the deck, dashboard or client portal, with every figure traced back to its pull.
- 09
Retain each source pull, reconciliation difference, restatement and edit made to the commentary.
Product statement
The agent assembles, reconciles and drafts. It does not decide what caused a change, and no report reaches a client without a named analyst signing it.
Example workflow
One report, end to end
- 1Reporting period closesThe week, month or flight the client contracted for reaches its end date
- 2Sources pulledEach platform API, measurement source and warehouse table, stamped with pull time, currency and time zone
- 3Figures reconciledPlatform figures set against the warehouse, overlapping conversions deduplicated, every difference sized
- 4Commentary draftedWhat moved and against which comparison, with the attribution setting and window stated beside it
No human action required
Stages 1 to 4 run without a person in the loop — the pull, the reconciliation and the first draft of the commentary are finished before anyone is asked to read anything.
Reaches the analyst ready to sign.
Held with the gap named, not sent.
Analyst and account lead
The analyst checks the figures against the warehouse and rewrites any line that asserts a cause; the account lead signs what goes to the client.
Approve · Rewrite the line · Send back
Approved — handed back
- 6Report deliveredInto the deck, dashboard or client portal, only after the named analyst has signed it
- 7Outcome evaluatedFigure agreement, restatement stability, commentary lines rewritten in review, and anomaly precision by client
Commentary edits
Every line the analyst rewrites is counted, and causal claims are counted separately.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Stating that a campaign caused a business result.
- Signing or sending a client-facing report.
- Changing an attribution model, window or comparison.
- Restating a figure the client has already seen.
Automation boundary
Agent acts unaided
- ✓Pull each source on schedule and stamp what it arrived as.
- ✓Reconcile platform figures against the warehouse and size every difference.
- ✓Deduplicate conversions more than one platform claims, by the agreed rule.
- ✓Draft commentary describing movement, and flag anomalies for a person.
Write actions run only inside the approval boundaries agreed during implementation. Sending a report and asserting a cause are not among them.
- Choosing the metric a bonus is judged on.
- Reporting a partial period as if it were final.
- Closing an anomaly or ruling it not material.
- Changing a metric definition in the reporting layer.
Example output
One reporting week, annotated
Everything the agent states is attached to the pull it came from.
Reporting output · single client week
Illustrative example- Client and period
- Retail brand, one week
- Sources
- Four platforms, warehouse, CRM
- Pull stamp
- Read today
- Reconciliation
- Conversions do not agree
- Confidence
- 92%
- Commentary
- Drafted, not signed
As received
Each source exactly as it returned the data, with the window, currency and time zone it arrived under.
Evidence used
- Warehouse order table
- Platform pull, stamped
- Deduplication rule
Why it is held
Two platforms claim the same orders, and the period is still inside restatement.
What the score decides
It decides how much the analyst re-reads before signing, not whether the media worked.
Action
ApproveRewrite the lineSend 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 source, every period
From platform APIs and the warehouse
Reconcile against one agreed source
Set every platform figure against the warehouse under the client's own definitions, and size each difference instead of averaging it away.
Reduce the build time on every report
The pull, the joins, the deduplication and the first draft are done before an analyst opens the file, so the hour goes on reading rather than building.
Surface the disagreement before the client does
Figures that do not reconcile, periods still restating and definitions that have moved are raised now, instead of in the meeting where the client asks.
Build an evidence trail
Retain every source pull, the setting and window each figure arrived under, each reconciliation difference, each restatement and every edit to the commentary — on both paths.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Ad platforms
- Google Ads
- Meta
- DV360
- The Trade Desk
- Amazon Ads
- retail media
Measurement & verification
- IAS
- DoubleVerify
- MMM and lift outputs
- brand-lift vendors
Warehouse & conversions
- BigQuery
- Snowflake
- dbt
- GA4
- CRM and order systems
Agent
Performance reporting & reconciliation
- Pulls sources
- Reconciles figures
- Drafts commentary
Reporting & delivery
- Looker
- Power BI
- Tableau
- Decks and slides
- client portals
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 client
Each control wraps the one inside it. A figure clears every layer before it reaches a deck, and the causal explanation sits outside all six.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Decide whether the draft may stand
- L3
- Decides who signs it and sends it
- L4 – L5
- Keep the definitions and the trail intact
- L6
- Pulls automation back when signals degrade
How Nestack evaluates it
Evaluate the whole report — not only the headline number.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the report and the commentary the client reads
- E1Final-output evaluationDo the delivered figures agree with the warehouse, row by row?
- E2Step-level evaluationDid it read the right window, setting, currency and time zone?
- E3Tool evaluationDid it pull the correct account, client, channel and date range?
- E4Commentary and anomaly reviewHow often is a line flagged for cause, and is a flagged anomaly real?
- E5Slice evaluationHow does performance change across clients, channels and periods?
- E6Business outcomeFigures corrected after delivery, and reports restated in front of a client.
Floor — the decision the client makes from it
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 · Source pull
2 modes
- PR-01
Partial period read as final
A window still restating is reported as settled.
- PR-02
Currency or time-zone drift
Days and totals shift against the client's own books.
Stage gathers
Platform, measurement, conversion and warehouse data
02 · Reconciliation
1 mode
- PR-03
Double-counted conversions
Two platforms claim one order and both are summed.
Stage matches
Every platform figure against the warehouse row
03 · Commentary
2 modes
- PR-04
Cause asserted without a test
A movement no test measured is given a reason.
- PR-05
Commentary tilted to the agency
The window that flatters the work is chosen.
Stage drafts
What moved, against which comparison and window
04 · Delivery
1 mode
- PR-06
Right in the warehouse, wrong in the deck
The figure changes on the last hop into the slide.
Stage returns
The deck, dashboard or portal the client reads
05 · Change / Version
1 mode
- PR-07
Definition moved under the trend
A metric changes meaning and the series absorbs it.
Stage tracks
Model, prompt, definition and pipeline changes
- Sev-1 · a decision is made on a false reason
- Sev-2 · a wrong figure reaches the client
- Sev-3 · the report is held or delivered late
Affected slices
Agreement in total is not agreement by slice
A report can reconcile in aggregate while a few clients, channels and periods carry nearly all of the differences, the restatements and the commentary lines a reviewer has to rewrite. 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 |
|---|---|---|---|---|
| Periods still inside restatement | 5.9% | 3.1× | Review | |
| Multi-platform deduplicated conversions | 4.9% | 2.6× | Review | |
| Cross-currency, multi-market clients | 3.2% | 1.7× | Watch | |
| Single-platform steady-state accounts | 1.1% | 0.6× | Normal |
Bar: failure-rate lift vs. steady-state baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
The figure you signed last week has since moved
Conversions land days after the click and modelled figures settle later. A cycle that never re-reads a reported period closes the wrong thing.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Figure agreement, restatement or reviewer edits move on one client or channel.
- 02
Diagnose
Traced to the pull, a join, a deduplication rule, a definition or the commentary.
- 03
Improve
The rule, mapping or wording is changed, re-approved by the analytics lead and version-linked.
- 04
Verify
Re-run against closed periods, including the ones the client queried.
- 05
Learn
The query becomes a regression case and the definition enters the reporting glossary.
Learn → Detect
The return edge. Each cycle re-opens periods already sent, so a restatement is caught by the agent rather than by the client.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, source connections, reconciliation, commentary and evaluation, delivery, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Reporting discovery and automation-boundary definition.Week 1
- 02Platform, measurement and warehouse APIs.Week 1, Week 2
- 03Metric definitions and attribution settings.Week 1, Week 2
- 04Scheduled pulls, currency and time-zone rules.Week 2
- 05Reconciliation logic and difference reporting.Week 2, Week 3
- 06Conversion deduplication across platforms.Week 3
- 07Restatement handling and period-close rules.Week 3, Week 4
- 08Commentary drafting and causal-language controls.Week 4
- 09Anomaly detection and reviewer routing.Week 4, Week 5
- 10Evaluation suite, slices and regression cases.Week 4, Week 5
- 11Delivery into decks, dashboards and portals.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 client, one report | ProductionProduction data pipeline | AdvancedMulti-client / multi-market |
|---|---|---|---|
| Introduced at Pilot | |||
| Scheduled platform pulls | ✓ | ✓ | ✓ |
| Warehouse reconciliation | ✓ | ✓ | ✓ |
| Drafted commentary | ✓ | ✓ | ✓ |
| Named-analyst sign-off | ✓ | ✓ | ✓ |
| Baseline evaluation | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Conversion deduplication rules | — | ✓ | ✓ |
| Restatement and period-close handling | — | ✓ | ✓ |
| Anomaly detection and routing | — | ✓ | ✓ |
| Delivery to decks and dashboards | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-market currency and time zones | — | — | ✓ |
| Client portals and enterprise controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on the platforms and measurement sources in scope, reconciliation depth, reporting cadence, delivery surfaces, review 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 reports you send today, and who signs them
→Reporting discovery and automation-boundary definition
Week 102Your metric definitions, attribution settings and comparisons
→Definition, attribution and comparison-period mapping
Week 103Access to platform, measurement and warehouse APIs
→Scheduled pulls with currency and time-zone normalisation
Week 204The table or view you treat as the source of truth
→Reconciliation logic and difference reporting
Week 205How you want overlapping conversions counted
→Deduplication rules and restatement handling
Week 306Reports a client queried, and the numbers you had to correct
→Evaluation suite, slice reporting and regression cases
Week 407Named analysts and the surfaces clients actually read
→Reviewer routing and delivery into decks, dashboards and portals
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 evaluation slices and the first reports delivered under signature.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Reports in scope, metric definitions and who signs
W2Source pulls, normalisation and the reconciliation baseline
W3Deduplication, restatement handling and commentary drafting
W4Causal-language controls, anomaly routing and evaluation suite
W5Delivery surfaces, first signed reports and targeted corrections
W6Production validation, restatement re-runs and Agent Care handover
- Reading the band
- The causal-language control and the reviewer routing are built in week 4, before a report reaches a client in week 5. The bars show that dependency, not a smooth ramp.
- At the end of W6
- Reports have gone out under a named analyst's signature and a closed period has been re-read for restatement, 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 performance-reporting agent around your measurement stack.
Show us the reports you send, the sources behind them and the last few numbers you had to correct in front of a client. We'll reconcile one client's reporting week end to end against your warehouse and show you where the figures disagree, before anything is scoped.
- Typical delivery
- 4–6 weeks
- Deployment
- Your reporting and data stack
- Report sign-off
- Named analyst, always
- After launch
- Agent Care available
Nestack Agents · Performance reporting & reconciliation
AGT-AM-03 · Agent Care available after launch