Main site

Advertising AI agent · Supply-path & curation

Supply-Path Optimisation & Curation AI Agent

Map every path a bid can take to one impression, price the fees on each hop, and model what consolidating costs in reach, not just fees — the programmatic lead approves every change.

Typical delivery
4–6 weeks
Deployment
Your stack
Buyer approval
Pre-cutover
After launch
Agent Care

What this agent does

Maps the paths, not the decision to cut one

In

  1. 01

    A bid's path is pulled from the DSP log, every SSP report and each declared hop between them.

  2. 02

    The schain object is read alongside it, as the payment path it shows and nothing more.

Reason

  1. 03

    A curated deal is checked against its own terms and the curator's tie to each SSP inside it.

  2. 04

    Every fee the path discloses is attributed to its hop; what isn't visible is marked, not estimated.

  3. 05

    A proposed path change is modelled for the fee it moves and, separately, the reach it would cost.

Decide

  1. 06

    A curator that is also the SSP running the auction is named, never modelled away.

  2. 07

    Every proposed path change is routed to the named programmatic lead, never applied alone.

Out

  1. 08

    The paths compared, the fees found and the buyer's decision are retained against the deal.

  2. 09

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

Product statement

The agent maps paths and prices what it can see; the programmatic lead decides what changes, and the agent never cuts a path or renegotiates a fee itself.

Example workflow

One path change, signal to approval

  1. 1Signal receivedPath-performance report, a curator renewal, a consolidation request or a new deal ID
  2. 2Evidence assembledDSP logs, SSP reports, curator terms and schain data, each with its source
  3. 3Change draftedPath map, fee attribution, reach estimate and confidence
  4. 4Controls appliedFee-visibility checks, curator-conflict checks, reach modelling and confidence threshold

No human action required

Stages 1 to 4 run unaided, and no path changes at any of them — the agent models, and the buyer's lane opens at the confidence gate.

5DecisionBranches at the confidence threshold
High confidence

Goes to the programmatic lead to approve.

Low confidence

Adds a second desk review first.

Buyer approval

The change is held with its evidence, its reach estimate and the confidence.

Approve · Edit · Escalate

Approved — cut over live

  1. 6DSP and SSP settings updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedFee accuracy, reach delta, curator performance and post-cutover corrections

Edits

Every buyer edit is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Cutting a supply path or ending a curator relationship.
    • Changing a bid floor, deal ID or SSP fee arrangement.
    • Deciding a curator's fee is undisclosed or improper.
    • Negotiating terms with an SSP, exchange or curator.

  • Automation boundary

    Agent acts unaided

    • ✓Map every declared path a bid can take to one impression.
    • ✓Attribute each visible fee to the hop it was charged on.
    • ✓Model the reach a proposed path cut would cost, not only the fee.
    • ✓Flag a curator's tie to the SSP it curates, and hold the change for the buyer.

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

    • Asserting a partner took an undisclosed margin.
    • Presenting a modelled fee saving as money recovered.
    • Adding a new SSP or curator to the buy.
  • Changes to fee-visibility rules, reach models or approval thresholds.

Example output

One path map, annotated

Everything the agent models is attached to the deal it was drawn from.

Supply-path output · single deal

Illustrative example
Deal
Curated video deal
Finding
Curator fee bundled into the deal's CPM, not itemised on the terms
Fee status
Fee not itemised
Evidence source
This week's DSP log
Confidence
81%
Attribution
Programmatic-lead name on file

As received

Taken from this week's DSP log and the curator's published deal terms — nothing on this side is written by the agent.

Paths checked

  • DSP delivery log
  • Curator deal terms
  • SSP fee disclosure

Why it's flagged

The CPM matches the insertion order, but the curator's own terms don't break out its fee.

What the score decides

Below the configured threshold the flag picks up a second desk review before it reaches.

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 path

From the DSP and SSP feeds

03Mapping

Map every bid's path

Trace each hop a bid can take against the DSP log, the SSP report and the curator's terms.

01Approved path

A modelled saving is not a saving

Curated does not mean clean: the industry's own benchmark data still finds made-for-advertising inventory inside curated PMPs — smaller than the open exchange's rate, never zero.

02Human review

Send review to the costly paths

Undisclosed fees, conflicted curators and reach-losing cuts are marked, so the buyer's read starts where risk concentrates.

04

Build an evidence trail

The path, the fees read against it and the buyer who cut it stay on the deal.

Integrations

Typical integrations

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

  • DSP & bidding

    • The Trade Desk
    • DV360
    • Amazon DSP
    • Yahoo DSP

  • SSPs & exchanges

    • Magnite
    • PubMatic
    • Index
    • OpenX
    • Xandr
    • Equativ

  • Curation platforms

    • Scope3
    • Audigent
    • Illuma
    • Sincera

Agent

Supply-path optimisation & curation

  • Maps the path
  • Prices the hop
  • Holds for the buyer

  • Verification & measurement

    • DoubleVerify
    • IAS
    • MOAT
    • Human

  • 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 path

The layers sit inside one another. What none of them catches is listed in the map below.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeNarrow back to path reporting when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, fee-model and curator-terms changes.Track
L4TraceabilityRecord the paths compared, the fees found, the reach estimate and the buyer's decision.Record
L3Buyer approvalHold the change for the named buyer; it governs cutover, not whether the model is right.Gate
L2Policy guardrailsTest every change against configured fee-visibility and curator-conflict rules; a failure returns it.Restrict
L1Confidence thresholdsRoute low-confidence path changes to a second desk review.Require review
Model coreChange proposed — path map, fee attribution, reach estimate and confidence
L1 – L2
Test whether a change may stand
L3
Puts the cutover in the buyer's hands
L4 – L5
Keep the path and the fees behind it
L6
Narrows to path reporting when signals degrade

How Nestack evaluates it

Evaluate the path-mapping workflow — not only the final number.

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

Surface — the path the buyer sees

  1. E1Final-output evaluationDid the attributed fee match a disclosed line at that hop?
  2. E2Step-level evaluationDid the agent use the current DSP log, SSP report and curator terms?
  3. E3Tool evaluationDid it read the correct deal ID and the correct SSP account?
  4. E4Confidence calibrationDo low-confidence path changes actually attract more buyer escalations?
  5. E5Slice evaluationHow does fee visibility change across specific supply types?
  6. E6Business outcomeHow many proposed changes needed a buyer correction after cutover?

Floor — the outcome the buyer 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

    • SO-03

      Stale path snapshot

      The DSP log or SSP report read is a cached copy, not this hour's data.

    Stage gathers

    DSP logs, SSP reports, curator terms and schain data

  2. 02 · Reasoning

    2 modes

    • SO-04

      Blended MFA average cited

      A network PMP average is reported as this deal's own MFA rate.

    • SO-06

      Schain read as full custody

      Payment-flow data is treated as proof of the full technical path.

    Stage proposes

    Path map, fee attribution, reach and confidence

  3. 03 · Tool / write

    2 modes

    • SO-02

      Uniform log-level assumed

      Fee attribution assumes equal log-level detail from every SSP.

    • SO-05

      Saving reported as realised

      A modelled fee reduction is shown as money already recovered.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • SO-01

      Reach impact omitted

      A proposed path cut reaches the buyer with no measured reach estimate attached.

    Stage returns

    The path change the buyer approves

  5. 05 · Change / Version

    1 mode

    • SO-07

      Silent fee-model drift

      A model or terms update widens what the agent will treat as a disclosed fee.

    Stage tracks

    Model, fee model, curator terms and schain version

  • Sev-1 · path cut outside the boundary
  • Sev-2 · a wrong fee reaches the buyer
  • Sev-3 · log degrades, change routes to review
See the complete Advertising failure-mode catalogue

Affected slices

Aggregate savings can hide one lost path

A blended fee number can look efficient while a handful of paths carry most of the rework. Nestack reports the path-rework rate by hop, not only in total Nestack reports it by slice rather than in aggregate..

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Impressions reachable on one path only7.7%4.0×Review
Curated deals built by a third party5.3%2.8×Review
Long-tail and open-exchange supply3.6%1.9×Watch
Direct SSP paths to major publishers1.9%0.7×Normal

Bar: path-rework rate lift vs. the direct-SSP 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 a path, not a projection

A cycle is done when the lost reach has become a case the next release must pass. That suite is what the next path cut is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Path-rework rate rises in a supply slice.

  2. 02

    Diagnose

    The path that was cut and took reach with it is checked against every hop and deal until one cause explains it.

  3. 03

    Improve

    Version-stamp the change and attach the paths that exposed it.

  4. 04

    Verify

    The affected cases run again, and a failure stops the release.

  5. 05

    Learn

    It becomes a standing test, and the path rules change with it.

  6. Learn → Detect

    The return edge. The next model 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, path-mapping workflow, evaluation, integration, then production validation and handover.

Workstream

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

  1. 01Desk discovery and boundary definition and it is logged..Week 1
  2. 02DSP, SSP and curator source assessment.Week 1, Week 2
  3. 03Fee and curator-conflict rule mapping and rule mapping.Week 1, Week 2
  4. 04Path intake and deal-ID normalisation.Week 2
  5. 05Path-mapping logic and fee attribution.Week 2, Week 3
  6. 06Confidence scoring and reach routing.Week 3
  7. 07Buyer approval workflow.Week 3, Week 4
  8. 08DSP and SSP integration.Week 5
  9. 09Path and reach regression cases.Week 4, Week 5
  10. 10Guardrails and cutover controls.Week 4
  11. 11Deal-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 desk, one DSP seatProductionProduction DSP and SSP accessAdvancedMultiple markets / brands
Introduced at Pilot
Path mapping to your rules✓✓✓
Buyer approval✓✓✓
Fee-visibility baseline✓✓✓
Introduced at Production
Reporting by supply path—✓✓
Escalation workflow in your tools—✓✓
Approved cutover actions—✓✓
DSP and SSP integration—✓✓
Introduced at Advanced
Multi-market curator rules——✓
Multi-stage desk approvals——✓
High path volume——✓
Multi-market path 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 DSP seat list and current SSP / curator setup

    Path ingestion and deal-ID mapping

    Week 1

  2. 02Representative path and fee history

    Path-mapping baseline and fee attribution

    Week 2

  3. 03Your fee-visibility and curator-conflict rules

    Fee-visibility and approval-boundary mapping

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    DSP and SSP assessment, then integration setup

    Week 2

  5. 05Paths you would not want cut

    Reach cases and the evaluation suite

    Week 4

  6. 06What no path may be cut without

    Confidence scoring, reach routing, guardrails and approval controls

    Week 3

  7. 07Named buyers to review path changes

    Buyer 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

The phases are drawn on the weeks they occupy, so week 5 genuinely doubles rather than padding the plan.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Desk discovery, fee-rule mapping and the automation boundary

W2Source integration and the path-mapping baseline

W3Path-mapping workflow, confidence logic and approval controls

W4Evaluation suite, guardrails and failure-mode testing

W5DSP and SSP integration, pilot paths and targeted corrections

W6One buying cycle run under the programmatic 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 paths 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 a supply-path agent around your trading desk's approval chain.

Show us your DSP seats, your SSP and curator list, and who signs off on a cut. Not a lower blended number. A mapped path, its visible fees and the reach a cut would cost — named before the buyer approves it.

Typical delivery
4–6 weeks
Deployment
Your DSPs and SSPs
Buyer approval
Pre-cutover
After launch
Agent Care available

Nestack Agents · Supply-path optimisation & curation

AGT-AM-14 · Agent Care available after launch

Build a supply-path agent

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

About you
Your supply paths

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