High-accuracy time-series forecasting requires clean, granular inputs. PaceCast processes daily campaign-level advertising records. This guide details required columns, export configurations for Meta and Google Ads, and how our automated ingestion pipeline cleans currency symbols and attribution lags.
Data Specifications
Daily campaign-level. Exporting at the ad set or ad creative level is unnecessarily large and creates sparse conversion counts.
Minimum 90 days. 180 to 365 days is recommended to calibrate weekday purchasing velocity and multi-month seasonal trends.
Standard comma-separated value (.csv) or tab-separated value (.tsv) UTF-8 encoded files.
Required Column Schema
PaceCast includes an intelligent column resolver that maps platform-specific naming conventions to internal model fields:
| Model Field | Type | Definition & Recognized Aliases |
|---|---|---|
daterequired | YYYY-MM-DD | Daily reporting timestamp. Aliases: 'Day', 'Date', 'Reporting Starts', 'Period', 'Time'. |
campaignrequired | string | Campaign identifier or name. Aliases: 'Campaign name', 'Campaign', 'Campaign ID', 'id'. |
spendrequired | numeric float | Total daily advertising cost. Currency symbols ($ € £) and commas are automatically stripped. Aliases: 'Amount spent', 'Cost', 'Spend (USD)', 'Spend'. |
impressionsrequired | integer | Total impressions delivered. Aliases: 'Impressions', 'Impr.', 'Total Impressions'. |
clicksrequired | integer | Total clicks or link clicks. Aliases: 'Clicks', 'Link clicks', 'Clicks (all)'. |
conversionsrequired | numeric float | Primary conversion event (purchases, leads, or orders). Aliases: 'Purchases', 'Results', 'Conversions', 'Orders'. |
conversion_valueoptional | numeric float | Optional gross sales or conversion value used to model blended ROAS. Aliases: 'Purchase conversion value', 'Conv. value', 'Total Conversion Value'. |
Sample Ingestion Format
A minimal valid CSV export contains headers matching the required dimensions:
Date,Campaign,Spend,Impressions,Clicks,Conversions,Conversion value
2026-05-01,Prospecting - Broad,1250.40,48200,840,42,3570.00
2026-05-01,Retargeting - Catalog,420.00,12100,310,28,2100.00
2026-05-02,Prospecting - Broad,1310.00,51000,890,39,3315.00
2026-05-02,Retargeting - Catalog,395.50,11400,280,24,1920.00Meta Ads Manager Export Guide
To export from Meta Ads Manager (Facebook & Instagram):
- Open Meta Ads Manager and select the Campaigns tab.
- Set the date range in the top-right corner to the last 90–365 days. Ensure you end the range yesterday (exclude today to prevent incomplete day counts).
- Click Breakdown → Time → By Day. This step is required; omitting daily breakdown produces aggregate totals that cannot be modeled as a time series.
- Ensure your columns include Amount spent, Impressions, Link clicks, and your primary conversion event (e.g. Purchases).
- Click Reports → Export table data, choose .csv, and download the file.
Google Ads Export Guide
To export from Google Ads:
- Open Google Ads and navigate to Campaigns.
- Set the date filter to the trailing 90–365 days (ending yesterday).
- Click Segment → Time → Day above the performance statistics table.
- Confirm columns include Cost, Impressions, Clicks, and Conversions.
- Click the download icon in the table toolbar and select .csv.
Attribution Embargo Handling
Ad networks credit conversions back to the ad view or click date rather than the purchase date. Consequently, the most recent 7 days in any fresh export systematically underreport conversions as customer journeys finish converting.
If models train on incomplete recent days, they perceive an artificial plunge in conversion efficiency and project a false downward trend. PaceCast automatically excises the trailing 7 calendar days before training the model:
Common Pre-Flight Errors
- Missing daily breakdown: Exporting totals without selecting breakdown by day. PaceCast will reject files containing fewer than 60 distinct daily timestamps.
- Zero conversions across dataset: If your export does not include the primary conversion event column (e.g. only link clicks are present), CPA and forecast intervals cannot be calibrated.
- Negative spend or currency anomalies: While PaceCast handles currency prefixes, credit adjustments that create negative daily spend rows are zero-clipped automatically.