Main site

Advertising AI agent · Media buying

Media-Buying & Budget AI Agent

Build and maintain the media plan, set up buys, propose budget shifts and pacing changes, and reconcile delivered against billed — inside spend limits a named buyer set and can revoke.

Typical delivery
4–6 weeks
Deployment
Your stack
Spend release
Named buyer
After launch
Agent Care

What this agent does

Works the plan inside limits you set

In

  1. 01

    Read the approved plan, the flight dates, the committed IOs and the budget released against each line.

  2. 02

    Pull spend and delivery from every platform in the buy, in the plan's currency and day boundary.

Reason

  1. 03

    Compare planned against delivered against billed, line by line, and name where the gap came from.

  2. 04

    Set the platform's own numbers beside the measurement the client decides on, and flag disagreement.

  3. 05

    Draft a reallocation, bid or pacing change with the limit it consumes and its reversal written first.

Decide

  1. 06

    Hold anything that breaches a committed IO, a market minimum, a change window or a spend limit.

  2. 07

    Route every release of new spend and every shift across a client or flight to the named buyer.

Out

  1. 08

    Apply released changes on the platform, one line at a time, and write the same change to the plan.

  2. 09

    Retain the state it replaced, the approval, the change made and the reversal path for every action.

Product statement

The agent proposes and executes inside limits a person set. It does not release spend, and no shift it recommends moves money until a named buyer approves it.

Example workflow

One reallocation, end to end

  1. 1Plan and flight in viewThe approved plan, the flight dates, the committed IOs and the budget released against each line
  2. 2Delivery pulledSpend, impressions and pacing from every platform in the buy, in the plan's currency and day boundary
  3. 3Variance readPlanned against delivered against billed, line by line, with the platform's numbers set beside independent measurement
  4. 4Change draftedThe shift, bid or pacing change, the limit it consumes, the IO it touches and the reversal that puts it back

No human action required

Stages 1 to 4 run without a person in the loop — the delivery pull, the variance read and the drafted change are finished before the buyer is asked to look at anything.

5DecisionSplits on the size of the shift and the limits it touches
Inside every limit

Reaches the buyer ready to release.

Breaches a limit or a commitment

Held with the breach named, nothing applied.

Named buyer

The buyer checks the shift against the plan, the committed IO and the client's mandate, then releases the spend.

Release · Amend · Send back

Released — handed back

  1. 6Applied and watchedOnly on the lines the buyer released; spend and pacing are checked against the plan on a fixed interval
  2. 7Outcome evaluatedPlan-versus-delivered variance, guardrail breaches, reversal time and reconciliation against what was billed

Reversals

Every change carries the state it replaced, and the time taken to put it back is counted.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Releasing spend against a new flight or client.
    • Moving budget between clients, markets or committed IOs.
    • Raising any total budget, cap or daily limit.
    • Cancelling, pausing or short-rating a committed IO.

  • Automation boundary

    Agent acts unaided

    • ✓Read the plan, the committed IOs and delivery from every platform in the buy.
    • ✓Reconcile planned against delivered and billed, and name where the gap came from.
    • ✓Draft a reallocation with the limit it consumes and the reversal that puts it back.
    • ✓Hold anything that breaches a limit and route it to the named buyer.

    Write actions run only inside the approval boundaries agreed during implementation. Releasing spend and touching a committed IO are not among them.

    • Changing a bid strategy, target or pacing mode.
    • Making any change outside the agreed change window.
    • Turning on a platform auto-feature that re-spreads budget.
  • Changing the spend limits, guardrails or approval thresholds.

Example output

One line in the plan, annotated

Everything the agent proposes is attached to the plan line and the delivery it came from.

Reallocation output · single line item

Illustrative example
Line item
Online video, one market
Flight
Day 19 of 30
Released budget
$180,000
Proposed shift
Reallocate within IO
Confidence
71%
Status
Held for the buyer

As received

The plan line and the budget released against it — nothing on this side is inferred.

Evidence used

  • Platform delivery, dated
  • Committed IO minimum
  • Independent measurement

Why it is held

Efficient in the platform's own numbers, flat in the client's measurement.

What the score decides

How closely the buyer reads before releasing — not whether the shift is the right call.

Action

ReleaseAmendSend back

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 line in the plan

From the approved media plan

03Plan of record

Keep the plan as the record

Every platform change is written back to the plan it came from, with the limit it consumed and the state it replaced.

01Approved path

Cut the gathering before the decision

The delivery pull, the currency and day-boundary conversion and the variance read are done before a person opens the plan, so the buyer decides rather than gathers.

02Human review

Show the disagreement before money moves

Where the platform's own numbers and the client's independent measurement point different ways, the shift is held and both are put in front of the buyer.

04

Build an evidence trail

Retain the state replaced, the limit consumed, who released it, the change applied, the reversal path and what was finally billed — on both paths.

Integrations

Typical integrations

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

  • Buying platforms

    • Google Ads
    • DV360
    • Meta
    • The Trade Desk
    • Amazon
    • Retail media

  • Planning & order management

    • Media plans
    • flowcharts
    • IOs and POs
    • Order management

  • Finance & billing

    • Invoices
    • vendor billing
    • Purchase orders
    • Finance systems

Agent

Media buying & budget

  • Reads the plan
  • Proposes shifts
  • Holds breaches

  • Measurement sources

    • Ad servers
    • verification
    • MMM and incrementality
    • Warehouse

  • 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 buy

Each control wraps the one inside it. A change clears every layer before it reaches a platform, and the release of spend sits outside all six.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeRevert to the last released plan and hand buying back to the team.Roll back
L5TraceabilityRecord the state replaced, who released it, the change and the reversal path.Record
L4Blast radiusOne line and one account at a time, with a ceiling on changes per run.Limit
L3Named-buyer gateNo spend is released and no budget crosses a client or flight without a buyer.Gate
L2Commitment checksCommitted IOs, minimums and change windows are tested first.Check
L1Spend limitsEvery change is sized against the budget released for that line.Cap
Model coreProposal — the shift, the bid or pacing change, the limit it consumes and confidence
L1 – L2
Decide whether the change may stand
L3
Decides who releases the spend
L4 – L5
Keep the blast radius small and reversible
L6
Pulls automation back when signals degrade

How Nestack evaluates it

Evaluate the change that moved money — not only the plan.

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

Surface — the plan and the shift a buyer releases

  1. E1Final-output evaluationDid the buy deliver where it was booked, at the budget released?
  2. E2Step-level evaluationDid it read the right flight, currency, day boundary and committed IO?
  3. E3Tool evaluationDid it change the correct account, campaign and line, and nothing else?
  4. E4Buyer agreement and driftDo proposed shifts match a buyer's call, and is the gap widening?
  5. E5Slice evaluationHow does performance change across platforms, markets and buy types?
  6. E6Business outcomeGuardrail breaches, reversal time, and reconciliation against what was billed.

Floor — the spend that was delivered and billed as planned

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 →

  1. 01 · Plan lookup

    1 mode

    • MB-01

      Currency and time-zone error

      The day boundary and rate come from the account, not the plan.

    Stage gathers

    Flight, released budget, committed IOs and delivery

  2. 02 · Variance & shift

    2 modes

    • MB-02

      Optimising to an artefact

      Money follows a number independent measurement cannot see.

    • MB-03

      Committed-IO breach

      A reallocation takes a booked line below its minimum.

    Stage proposes

    The reallocation, the bid or pacing change and its limit

  3. 03 · Platform write

    2 modes

    • MB-04

      Spend on the wrong flight

      Budget is released against the wrong flight or client.

    • MB-05

      Duplicate line double-spend

      The same buy runs twice because a line was created, not amended.

    Stage applies

    Only on the lines and limits a named buyer released

  4. 04 · Pacing watch

    1 mode

    • MB-06

      Unattended change runs on

      A wrong change spends for hours before anyone sees it.

    Stage follows

    Spend against plan, and every change already live

  5. 05 · Change / Version

    1 mode

    • MB-07

      Auto-feature overrides plan

      A platform setting re-spreads budget the buyer had fixed.

    Stage tracks

    Model, prompt, limit and platform-setting changes

  • Sev-1 · money moves that no one released
  • Sev-2 · a commitment or the plan is broken
  • Sev-3 · the read is wrong before anything moves
See the complete Advertising failure-mode catalogue

Affected slices

The variance is not spread evenly across the buy

Aggregate plan-versus-delivered variance can look acceptable while a few platform and market cohorts carry most of the guardrail breaches, most of the reversals and nearly all of the reconciliation rework. Nestack reports performance by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Cross-market multi-currency buys6.1%3.4×Review
Retail-media and closed platforms4.5%2.5×Review
Final week of flight3.2%1.8×Watch
Single-platform steady buys1.3%0.7×Normal

Bar: failure-rate lift vs. steady-buy baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

The proof arrives after the money has gone

A wrong shift spends in real time; the invoice that settles it lands weeks later. Every cycle is dated twice — when it ran, and when billing confirmed it.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Variance, a guardrail breach or a reversal moves on one platform or market.

  2. 02

    Diagnose

    Traced to the plan read, the day boundary, a limit or the measurement it trusted.

  3. 03

    Improve

    The limit, rule or measurement source is changed, re-approved by the buyer and version-linked.

  4. 04

    Verify

    Re-run against held-out changes from that platform, including the ones that were reversed.

  5. 05

    Learn

    The reversal becomes a regression case and the limit enters the buying runbook.

  6. Learn → Detect

    The return edge. Every cycle re-checks against what was finally billed — a change that looked right on the day can still be wrong on the invoice.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, plan and delivery data, limits and proposals, evaluation, reconciliation, then production validation and handover.

Workstream

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

  1. 01Workflow discovery and automation-boundary definition.Week 1
  2. 02Plan, flight and committed-IO model.Week 1, Week 2
  3. 03Platform and DSP API assessment.Week 1, Week 2
  4. 04Currency and time-zone normalisation.Week 2
  5. 05Delivery ingestion and variance against plan.Week 2, Week 3
  6. 06Reallocation, bid and pacing proposals.Week 3
  7. 07Spend limits, change windows and blast-radius rules.Week 3, Week 4
  8. 08Named-buyer release and approval workflow.Week 4
  9. 09Buyer-agreement and guardrail evaluation.Week 4, Week 5
  10. 10Reversal path and safe-mode drill.Week 4, Week 5
  11. 11Order-management and billing reconciliation.Week 5, Week 6
  12. 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 tierPilotOne platform, one clientProductionProduction platform integrationAdvancedMulti-market / multi-platform
Introduced at Pilot
Plan and delivery reconciliation✓✓✓
Variance and pacing reporting✓✓✓
Reallocation proposals✓✓✓
Named-buyer release✓✓✓
Baseline evaluation✓✓✓
Introduced at Production
Spend limits and change windows—✓✓
Released changes on platform—✓✓
Committed-IO and minimum checks—✓✓
Reversal path and safe mode—✓✓
Introduced at Advanced
Multi-market and multi-currency——✓
Cross-client and enterprise controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on the platforms and markets in scope, managed spend, order-management and finance integrations, spend limits and 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. 01The plans, flights and committed IOs in scope

    Plan, flight and committed-IO model

    Week 1

  2. 02Access to the platform and DSP APIs you buy on

    Platform and DSP API assessment, then delivery ingestion

    Week 2

  3. 03Your markets, currencies and reporting day boundary

    Currency and time-zone normalisation

    Week 2

  4. 04Who may release spend, and up to what

    Spend limits, change windows and blast-radius rules

    Week 3

  5. 05The measurement you actually decide on

    Measurement sources, and the rule that holds a shift when they disagree

    Week 3

  6. 06Reallocations you made, and ones you refused

    Evaluation suite, regression cases and failure-mode testing

    Week 4

  7. 07Named buyers, and the invoices to reconcile against

    Named-buyer release workflow, then billing reconciliation

    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 buyer-agreement evaluation and the first changes applied to live budget.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Plans, flights and committed IOs; who may release spend

W2Platform APIs, delivery ingestion and currency normalisation

W3Variance read, reallocation proposals and spend limits

W4Buyer release workflow, the reversal drill and evaluation suite

W5First released changes on live budget, and targeted corrections

W6Billing reconciliation, production validation and Agent Care handover

Reading the band
The reversal path is built and drilled in week 4, before a single change touches live budget in week 5. The bars show that dependency, not a smooth ramp.
At the end of W6
Changes have run under named-buyer release and reconciled against what was billed, 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 media-buying agent around your plan and your limits.

Show us one client's plan, the platforms it runs on and who is allowed to release spend today. We'll take one flight end to end, agree the limits and the reversal, then scope from there.

Typical delivery
4–6 weeks
Deployment
Your buying stack
Spend release
Named buyer, always
After launch
Agent Care available

Nestack Agents · Media buying & budget

AGT-AM-02 · Agent Care available after launch

Build a media-buying agent

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

About you
Your setup

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