Treat a bid change as the spending commitment it is: record what changed, how far it could move spend, who approved it, and the standing authority it ran under.
2Change context assembledThe line, the budget behind it, the authority it runs under and the day it was raised
3Change proposal draftedThe change, its exposure, the authority it needs and completeness
4Controls appliedAuthority checks, exposure ceilings, reversibility checks and completeness confidence
No human action required
Stages 1 to 4 run unaided, and nothing is approved at any of them — the agent is proposing, and the finance lane opens at the completeness gate.
5DecisionSplits at the completeness gate
Inside the standing limit
Goes to the paid media lead to approve.
Anything past it
Adds a finance manager read first.
Finance review
The change is held with its exposure, its authority and the lines it would touch.
Approve · Append evidence · Send to finance
Approved — by a named manager▼
6Budget and bid records updatedOnly where write access and records policy allow it
7Outcome evaluatedExposure accuracy, authority coverage, reviewer corrections and what the read found
Corrections
Each finance correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required8 items
Approving a change above the standing limit.
Setting the spending limits themselves.
Committing the brand to a platform contract.
Deciding what the media budget is there for.
Automation boundaryAgent acts unaided
✓Record what a proposed bid change could move before anyone applies it.
✓Apply only inside the standing authority your finance.
✓Hold the change that exceeds a threshold until a named.
✓Show the part of a proposal that reversing it will never give back.
Nothing is approved or committed except by a named person, inside the agreed boundaries.
Judging whether a spend was worth making.
Telling finance the quarter is on plan.
Setting who may approve which changes.
Changes to budgets, bids or platform accounts.
Example output
One change, annotated
Our media buying and budget assistant builds the plan and proposes the shifts, and our spend pacing triage agent separates a real movement from a reporting artefact; this page is neither of those, because its subject is who inside one company may move budget, up to what limit, and on whose approval.
Change record · single proposalIllustrative example
Change
Recorded as
Authority
Evidence of record
Confidence
Held for
Bid ceiling raised, one campaign line
Raised overnight, exposure priced first
Past the limit
Platform change log, 3 August 2026
Held unapplied
The paid media lead, by name
As receivedTaken from the platform change log and the budget record — it reaches as far as those sources do.
What the record holdsBid rule beforeExposure ceilingApproval, dated
Why no approval hereWhether this money may run is a budget-holder call, never a model output.
ActionApproveAppend evidenceSend to finance
What the score decidesBeyond the configured limit a change picks up a finance read before the manager sees it.
Value
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value landsValue 01 – 04
Every changeFrom the line it would move
03Evidence
Where the evidence is used
Our advertising pages propose to a client who holds the budget; here the keyboard and the budget often sit with one person, so the control worth having stops an unauthorised change.
01Approved path
A bid is a commitment
Automated bidding compounds while unattended, so the limits have to bind hardest overnight, at weekends and across holidays, exactly when nobody is at the desk to read an alert.
02Human review
What was checked, and not found
No platform publishes the ceiling an automated strategy can reach in an unattended window, no vendor estimate of what a reversal costs was traceable to a stated method, and no independent benchmark for approval thresholds inside brand teams was located.
04Build an evidence trail
The change, the budget it moved and the manager who approved it 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.
Bidding and buying platformsSearch · social · programmatic Bid rules and change logs
Budget and finance systemsPlanning tools · finance ledger Budget and commitment records
Approval and authorityIdentity provider · approval matrix Standing authority records
Agent
Paid media bidding authority
Reads the changes Prices the exposure Holds for the manager
Records and case systemsFinance tooling · ticketing Approval and audit records
Integration availability depends on the client's existing systems and API access.
Agent controls
Six brakes between the model and the manager
Six brakes applied one after another, the last the hardest. What still runs is named in the map below.
L6 · Outermost — last line of defenceInward → L1 · closest to the model
L6Rollback / safe modeNarrow the agent to proposals only when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt and authority rules; a platform update that moves a bid ceiling is a version change here, not a setting.Track
L4TraceabilityRecord each change, the exposure priced for it, the authority it ran under and every read of it.Record
L3Manager approvalHold the change for a named manager; the hold governs whether it is applied, not whether it was wise.Gate
L2Authority guardrailsTest each proposal against the standing limits your finance function configured, per account, per line and per unattended window.Restrict
L1Confidence thresholdsRoute a thin proposal to a finance read first; a platform suggesting more spend is a party to the deal, not an adviser.Require review
Model coreProposal assembled — the change, its exposure, the authority and completeness
L1 – L2Test whether a change may run
L3Puts the decision in a person's hands
L4 – L5Keep the change and the budget behind it
L6Drops to recommendations only when signals degrade
How Nestack evaluates it
Evaluate the whole assembly — not only the change proposal that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the record a budget holder reads
Depth of coverage ▼
E1Final-output evaluationDid the proposal price what the change could actually move?
E2Step-level evaluationDid the agent read the right line, the right budget and the live bid configuration?
E3Tool evaluationDid it read and write the correct campaign line and the correct limit?
E4Confidence calibrationDo low-confidence proposals actually attract more finance corrections?
E5Slice evaluationHow does performance change across specific campaign lines?
E6Business outcomeHow many proposals needed a correction before the manager approved?
Floor — the budget 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 lifecycleDirection of processing →
01 · Retrieval1 mode
LK-03
Stale budget read
The budget read is not the one now in force.
Stage gathersThe lines, the budgets, the rules and the limits
02 · Reasoning2 modes
LK-04
Exposure priced too low
A change is applied on an understated ceiling.
LK-06
Platform advice read as neutral
A vendor recommendation is relayed unlabelled.
Stage proposesThe changes, their exposure and completeness
03 · Tool / write2 modes
LK-02
Thin proposal passed forward
A change moves on without the finance read.
LK-05
Bound to the wrong line
A change is filed against another budget.
Stage writesOnly where write access and approval policy allow it
04 · Output1 mode
LK-01
Applied, approval unrecorded
The record shows a change but not who allowed it.
Stage returnsThe change a manager approves and finance reads
05 · Change / Version1 mode
LK-07
Silent authority drift
A platform update moves the ceiling, not the record.
Stage tracksModel, prompt, authority rules and change fields
Sev-1 · unauthorised spend ran overnightSev-2 · wrong budget line was movedSev-3 · source degrades, change held back
A line-level pacing-accuracy figure can read clean while automated bidding lines carry most of the rework. Nestack reports the correction rate by campaign line, not only in total.
Slice performance — reported separately, not only in aggregateIllustrative example
Slice
Failure rate
Lift
Lift vs. threshold
Status
Automated bidding lines
9.1%
3.6×
Review
Broad-match search lines
6.5%
2.6×
Review
Retargeting and remarketing
4.0%
1.6×
Watch
Brand search lines
1.7%
0.7×
Normal
Bar: correction-rate lift vs. brand search baseline · scale 0–4.0× · tick marks the 2.0× review threshold2 of 4 slices over threshold
Evidence-linked improvement
What a runaway bid rule costs
The loop shuts when the runaway bid change is a regression case. That suite is what the next change applied is measured against.
Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn
01Detect
Correction rate rises on automated bidding lines.
02Diagnose
The bid rule that spent a quarter's budget while nobody was reading alerts is worked back until one cause remains.
03Improve
Any change goes out numbered, with the budgets that caused it attached.
04Verify
One budget case still failing is enough to hold the release back.
05Learn
One case joins the suite, one line joins the approval record.
Learn → DetectThe return edge. The next change applied is weighed 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, proposal assembly, evaluation, integration, then production validation and handover.
WorkstreamWeek 1Week 2Week 3Week 4Week 5Week 6
01Spending-authority and automation-boundary work.
02Bidding, budget and approval sources.
03Change-to-authority and exposure-ceiling mapping.
04Change and budget ingestion.
05Line, budget and authority binding.
06Exposure scoring and review routing.
07Manager approval workflow.
08Bidding-platform integration.
09Pacing and authority cases.
10Guardrails and approval controls.
11Change-trail instrumentation.
12Deployment, documentation and Agent Care handover.
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallelFinal 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 campaign line, one quarterProductionProduction approval workflowAdvancedMultiple brands / markets
Introduced at Pilot
Proposal assembly to your limits✓✓✓
Named manager approval✓✓✓
Managed-budget baseline✓✓✓
Introduced at Production
Reporting by campaign line—✓✓
Approval workflow in your systems—✓✓
Approved write-back—✓✓
Bidding-platform integration—✓✓
Introduced at Advanced
Multi-market estates——✓
Cross-account approval chains——✓
Large account structures——✓
Multi-account authority controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build priceConfirmed after discovery based on integrations, workflow complexity, spend volume, approval controls and deployment requirements.
Separate from buildBuild 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 bringWe build with it
01Your campaign lines and the budgets behind them→Budget mapping and change captureWeek 1
02Representative bid rules, changes and approvals→Record binding, authority logic and the exposure baselineWeek 2
03Your standing authority and its thresholds→Authority mapping, limit binding and the automation boundaryWeek 1
04Access to relevant APIs, feeds or exports→Bidding, budget and approval-source assessment, then integration setupWeek 2
05Changes you would not want reversed→Pacing cases and the evaluation runWeek 4
06What no bid change may guarantee→Exposure scoring, review routing, guardrails and approval controlsWeek 3
07A named manager to approve the change→Approval workflow, then pilot and production validationWeeks 5–6
Nothing else is requiredDeployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
The spans below are honest working weeks rather than drawn space, so two of them share week five.
PhaseW1W2W3W4W5W6
DiscoveryW1
BuildW2 – W3
EvaluateW4 – W5
Pilot & LaunchW5 – W6
Week focusW1Paid workflow discovery, authority mapping and the automation boundaryW2Source integration and the exposure baselineW3Proposal assembly, authority logic and approval controlsW4Evaluation suite, pacing cases and failure-mode testingW5Platform integration, pilot changes and targeted correctionsW6One media year run under the paid-media lead, then Agent Care handover
Reading the bandEach bar covers only the weeks its own work is named for. Week five doubles up because the work does.
At the end of W6When the change record validates, Agent Care takes the agent on.
DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Marketing AI agent
Build a bidding agent around the change that moved spend before anyone holding the budget had approved it.
Show us one bid change and the authority it ran under. Send a fortnight of change logs and we will hand back the ones nobody authorised. Undoing a change gives back neither the spend nor the learning period, so the irreversible part is shown before it is applied.
