The best DV360 skills for an AI agent
DV360 gets two built-in skills instead of one, because DV360 itself splits cleanly into two disciplines: operating the campaign → insertion order → line item entity hierarchy, and pulling performance or forecasting reach through the separate Bid Manager reporting vocabulary. DV360 Campaign Operations covers the first; DV360 Reporting & Reach Planning covers the second. Together they're the same reference role Google Ads' GAQL Best Practices plays — the card the orchestrator consults before it touches a DV360 entity or report.
DV360 Campaign Operations
- The ENTITY_STATUS lifecycle — DRAFT creates, PAUSED stops, ARCHIVED before DELETE — so an agent doesn't attempt an invalid state transition.
- Money fields expressed in micros, with worked examples for converting to and from major currency units.
- PATCH + updateMask semantics for partial entity updates, and insertion-order creation requirements.
- The targeting playbook — keyword, device, geo and placement assigned-targeting-options, plus the name-to-ID lookup for resolving a human-readable target into the ID an entity actually needs.
- First-party audience (customer match) create and update, including the SHA-256 hashing contacts require before upload.
DV360 Reporting & Reach Planning
- Bid Manager's own FILTER_*/METRIC_* vocabulary and dataRange enums — distinct from GAQL, since DV360 reporting runs through the separate Bid Manager API.
- Reading dates and money in the advertiser's native timezone and currency, so numbers match the DV360 UI rather than a silently-converted approximation.
- The reach-planning chain: listing plannable products, then generating a reach forecast for a proposed budget and targeting combination — see the full reach-forecasting walkthrough for this in practice.
- Bid Manager's daily query quota discipline — batching everything a question needs into one report call rather than looping per entity.
Why two skills instead of one
Operating entities and pulling reports are genuinely different DV360 API surfaces — different base APIs (Display & Video 360 v4 vs. Bid Manager v2), different vocabularies, and different failure modes. Splitting them means a run that only needs reach forecasting doesn't load entity-mutation detail it won't use, and vice versa — the same reason Google Ads splits its 24 skills by discipline rather than shipping one giant reference card.
How they get used, and adding your own
Both load automatically when a DV360 account is connected and your goal touches the matching surface — campaign/line-item work loads Campaign Operations, reporting or planning questions load Reporting & Reach Planning, and a goal touching both loads both. You can also pin either one directly in the Workspace composer, and save your own DV360 workflow as a skill the same way as any other platform — see Skills, memory & teams for the general mechanism.
Frequently asked questions
- How many built-in DV360 skills does Agent Planners have?
- Two: DV360 Campaign Operations (the entity CRUD hierarchy and targeting) and DV360 Reporting & Reach Planning (Bid Manager reporting and reach forecasting).
- Why are DV360's skills split by operations vs. reporting?
- Because DV360 itself splits that way at the API level — entity operations run through the Display & Video 360 API while reporting runs through the separate Bid Manager API, with different vocabularies and quota rules.
- Does the reach-forecasting skill cost real budget to use?
- No — reach forecasting is a simulation against DV360's own forecasting API; it answers the budget-vs-reach question before any line item is created or any spend commits.
- Do both DV360 skills always load together?
- Only when your goal touches both surfaces — a pure reporting question loads just the reporting skill, and a pure entity-creation goal loads just campaign operations.
- Can I add a custom DV360 skill for my own workflow?
- Yes — the same mechanism as any platform: ask the agent to save a proven run, approve a proposed skill card, or import a playbook from a link.