Looker Studio and Meta Ads: Building Custom Performance Dashboards

A dashboard can look polished while multiplying spend or averaging incompatible ratios. Build one that reconciles to a small export before attaching a live connector. This guide includes fictional input rows, the expected failed join and a corrected report specification. It is a reproducible worksheet, not a deployed native report or observed customer result.

Freeze the source contract

Record account ID, date range, account timezone, currency, reporting grain, conversion definition and attribution window, export timestamp and connector refresh rules. Keep IDs as text to preserve them. Campaign names are labels, not stable keys. Separate platform-attributed purchases from settled store orders and document why they differ.

Use the Google connector catalog to verify the actual connector you plan to authorize. Partner connectors can have their own subscription, quota and permission requirements; a report interface does not imply a free built-in Meta connection. Existing Supermetrics reporting guidance describes the external connector workflow. Start with a controlled spreadsheet or export if live access is unavailable.

Copy this fictional export and dimension table

Dataset A has one fact per account, date and campaign. All amounts below are USD, the timezone is UTC, and purchases use one consistent illustrative definition. These are test fixtures, not campaign performance.

Dataset B is a fictional creative dimension with three distinct records belonging to camp_A. It has no spend facts. Include creative_id as a dimension when reproducing the join so pre-grouping does not collapse the rows.

Reproduce USD 100 becoming USD 300

Left-join A to B on account_id, date and campaign_id, retaining creative_id. The camp_A fact matches three creative rows. The resulting table repeats spend 100 and purchases 10 three times; camp_B remains one unmatched row. Naively summing the joined table yields USD 1,200 and 60 purchases, although the original totals are USD 1,000 and 40 purchases. For camp_A alone, USD 100 becomes USD 300. Adding account_id protects against cross-account collisions but does not fix this one-to-many relationship.

Google’s blending documentation explains that included dimensions govern pre-grouping and that joined tables inherit underlying freshness and credentials. Adding a unique row ID can preserve source rows; it does not prevent duplicated facts after a one-to-many join. Hiding repeated join fields affects fields, not the multiplication of spend values.

Correction for a campaign report: keep spend and purchase totals at campaign grain, or pre-aggregate B to one row per account/date/campaign, such as creative_count = 3 for camp_A. Join only after validating that this key is unique on both sides. If the question needs creative-level spend, obtain a creative-level fact export; do not allocate campaign spend evenly without explicitly labeling that allocation as an assumption.

Expected report after correction

Use a total ratio: SUM(spend) / SUM(purchases). The simple average of campaign CPAs is (10 + 30) / 2 = USD 20 and answers a different question. If purchases total zero, display “No purchases; CPA unavailable” and retain spend; do not display zero CPA or infinity as a usable recommendation. With mixed currencies, show separate currency totals or apply an explicitly dated exchange-rate source before summing. USD 100 plus EUR 100 is not USD 200.

Check stale data and permissions

1. Reconcile a chosen closed day against the same Meta export settings; verify row counts, duplicate keys and totals before enabling report sharing.
2. Display source export time and last successful refresh. If refreshed data is stale or a connector fails, mark the decision panel unavailable instead of silently showing cached values as current.
3. Test the intended viewer account with the chosen source credentials model. Confirm exactly which data it can see, whether raw records can be downloaded, and whether another account is accidentally exposed.
4. Verify denied access using an unauthorized test identity and repeat after permission changes. Keep credentials outside the report; record the responsible owner and recovery process.
5. Test zero purchases, missing campaign dimension, renamed campaign, timezone boundary and mixed currency fixtures. Save expected and actual outcomes internally.

This acceptance plan has not been executed against a live Google report here. A shared report and its source have separate access implications; confirm the actual credentials and connector policy. Remove personal data from demonstration exports and avoid granting broad source access merely to solve a display error.

Use the report for one bounded decision

If a reconciled mature campaign row exceeds your contribution-based acquisition limit, inspect its conversion definition, return lag, budget and landing experience before proposing a pause or a new test. A chart alone cannot tell whether Meta caused those purchases. Prepare a reviewed campaign draft in NovaStorm if that is your workflow; this external dashboard does not establish a built-in Looker connector, automatic budget adjustment or causal attribution feature.