Standards
AdTech data architecture
Plan advertising data around source systems, record grain, identifiers and metric definitions so reports support clear decisions.
AdTech data architecture maps how advertising data moves from source systems into interpretable measures. Start with the decisions the data must support. For each required measure, document its source, record grain, identifiers, definition and owner. Pooling figures in one warehouse does not make differently defined impressions comparable.
Design around the questions
A buyer may need daily spend by line item. A publisher may need delivered impressions by ad unit. An operations team may need to trace a delivery stage. Each question needs a source and level of detail suited to it.
| Layer | What to record | Decision it supports |
|---|---|---|
| Source | Platform objects, available files and reports | Which system supplies each value? |
| Ingestion | Extract time, source version and corrections | Can late or revised data be handled? |
| Model | Object relationships and the grain of each table | What does one row mean? |
| Reporting | Metric definitions and permitted breakdowns | Which figures can be compared? |
| Control | Access and review owners | Who resolves an unexplained value? |
These are design responsibilities, not features every platform supplies. Google Ad Manager provides event-level Data Transfer files via a separately enabled, paid feature. Display & Video 360 instant reports use selected dimensions and metrics. Before designing around them, confirm the account's available files, fields and permissions.
Set a source contract before ingestion
Google Ad Manager Data Transfer provides non-aggregated event-level data from ad server logs, with a separate file for each event type. Event data is accurate to the second, and configurations can include contextual fields such as device and geography. Ad units must be approved by partners to appear in Data Transfer files.
Data Transfer is an additional-cost feature enabled under Advanced features; enabling it requires Admin permissions. Check the account's available files, pricing and access before treating event-level data as a committed architecture input.
Data Transfer files are available in CSV and Parquet. Google notes that operating the feed requires ETL capability, handling large files and manipulating text files, a mid-sized data store and scripting. Organisations without those capabilities can consider an approved Google Marketing Platform partner.
Separate objects, events and summaries
A campaign object describes configuration. An event record describes an occurrence at a specified delivery stage. A report groups values under selected dimensions and counting rules. Model them separately.
Google Ad Manager orders contain line items; ad units represent inventory locations. A Display & Video 360 line item belongs to an insertion order. Similar labels across systems do not establish a shared ID or business relationship. Retain each native ID with its platform and account or network, and record any approved cross-system relationship separately.
For each report extract, retain its metric definition, dimensions, filters, period, time zone, currency where relevant, and extraction time. A grouped total cannot be turned back into a complete event history.
Preserve configuration that explains delivery
Configuration records can explain why delivery differs even when campaign labels look alike. In Google Ad Manager, an Order can hold general information such as currency or salesperson. Its LineItems describe requirements including when and how ads show, creative sizes, priority and cost structure.
Keep the source fields needed to interpret those settings alongside the reporting model, rather than relying on a display name alone. This helps an analyst understand what a line item was configured to do when reading operational results.
Define the measures and their limits
An impression measure needs its counting stage and source; spend needs a currency and cost definition. A rate needs a numerator and denominator; a daily measure needs a rule for assigning events to dates. Mark data that is provisional, late or corrected.
Ad Manager documents late Data Transfer events. An ingestion design should retain enough source detail to identify and process late-arriving data without silently counting it twice.
Make ingestion operationally supportable
Allow for normal delivery latency when setting refresh expectations: Google says Data Transfer delays of up to 15 hours after the recorded hour are normal. A report for near-real-time operations should state its freshness expectation and identify the source feed that can meet it.
Assign responsibility for maintaining the extraction and interpreting the measure. For Data Transfer, document who manages the feed, its access permissions and the dependencies needed to keep it operational. This keeps an operational report from relying on an unowned feed.
Plan a first usable report
Choose one decision, identify the available source records, and define one row in each input and output. Record native IDs and any dated business mappings. Compare a small period only after aligning the event stage, inventory, filters and time zone. Keep unexplained differences visible and assign an owner to the measure.
If the source cannot provide the detail the question requires, use a supported aggregate measure or change the question. The architecture should show which figures can be joined, which can only be compared as aggregates and which are unavailable.
For Display & Video 360 instant reports, record the saved report settings and the exported file. Scheduled delivery specifies a time zone, frequency and date range; reports are delivered at midnight in the selected time zone.
A shared instant report gives another user the report files, not access to run the report. Treat the delivered file and the runnable report definition as distinct inputs when deciding who can refresh a metric and how a change to its settings will be reviewed.
In this guide
- Mapping campaign data across advertising systemsBuild a campaign-to-platform map using native IDs, dated relationships and report grain, then resolve uncertain joins.
- Distinguishing raw events from aggregated ad reportsUnderstand the grain and limits of event-level advertising records and grouped reports, and compare like with like.
- Planning a stable advertising data identifierDefine stable campaign, platform-object and source-event keys with clear namespaces, dated mappings and validation checks.
