Marketing Analytics — Guide · 2026
A worked campaign performance analysis on a downloadable Google Ads and GA4 sample: make the two sources agree, compare CPA and ROAS, test for noise, and decide what to scale, fix or cut.
13 min readQuick answer — campaign performance analysis
Campaign performance analysis compares what each campaign cost with what it produced, against a target set before launch, so you can decide what to scale, fix or cut. First make Google Ads and GA4 count the same thing. Then compare cost per conversion and ROAS by campaign, check each gap against normal weekly noise, and write down the decision. The worked example reconciles both sources, number by number, on a sample you can download.
It is Sunday night and the budget meeting is on Monday. On the default click-date columns, the Google Ads overview says last week’s return on ad spend (ROAS) fell to 2.48, against 3.73 for the previous twelve weeks. Someone has already suggested pausing the two most expensive campaigns.
This guide shows how to analyze campaign performance before that meeting. The running example is a fictional six-person outdoor-gear store with eight Google Ads campaigns, plus email and paid social, over 13 weeks. Most of the “drop” is incomplete data. One campaign has a real leak, and one that looks expensive is only noise. Every number comes from three synthetic files you can download.
Who this is for: marketers, agency analysts and founders who set ad budgets from Google Ads, GA4 and other exports.
Run in an Anomaly test workspace on 8 October 2026. AI-assisted draft, fact-checked by [editor, date].
A campaign report says what happened: spend, clicks, conversions. Campaign performance analysis asks whether each campaign earned its budget against the target, whether the difference is real, and what should change. In the example below, the report shows ROAS of 2.48; the analysis decides whether Tents, Display or nothing should lose budget.
Campaign analytics usually means the tracking and tools that collect the numbers. Marketing performance analysis is the wider view across every channel and goal. A marketing campaign analysis, or campaign analysis, is the same job for one campaign or a set.
To measure campaign performance, start from the decision and work back to the metric. ROAS is conversion value divided by spend. Cost per acquisition (CPA) is spend divided by conversions. Click-through rate (CTR) and cost per click (CPC) explain why those moved, but rarely settle a budget question.
Campaign performance metrics and where they come from
| Decision | Metric | Source |
|---|---|---|
| Is the campaign worth its budget? | ROAS = conversion value ÷ spend | Google Ads, conversion-time columns |
| Is each sale affordable? | CPA = spend ÷ conversions | Google Ads |
| Do the clicks reach the site? | Sessions compared with clicks | GA4 traffic acquisition, by session campaign |
| Does the platform’s count hold up elsewhere? | GA4 purchases compared with Ads conversions | GA4 and Google Ads, joined on campaign ID |
| Is the difference real? | Conversions in the window (at least 30) | Both |
| Is a cut even needed? | Pacing: spend to date against budget | Google Ads and your budget plan |
Download the three-file sample
Synthetic sample data for 29 June to 27 September 2026: Google Ads by day (4,095 rows), GA4 by session campaign (1,274 rows), and email and paid social (364 rows). The README lists the planted effects and an answer key for every number in this guide.
campaign-google-ads-daily.csv · campaign-ga4-session-campaign-daily.csv · campaign-other-channels-export.csv · README
Decide what counts as success before you see the numbers, or every result can be argued either way. We saved these rules as project knowledge, so every later answer used them.
To analyze Google Ads data alongside GA4, bring the files into one place. Connect the sources directly (see how to analyze GA4 data in Anomaly and Google Ads and Search Console data analysis), or upload exports, as we did.
Before any analysis, check the data itself. Are email and paid-social links UTM-tagged, so GA4 can match each visit to its campaign? Is the last day complete? Did a campaign change its name? Does every Google Ads campaign ID appear in GA4?
Result: every email and paid-social row in GA4 has a campaign name. 27 September holds only 482 Google Ads clicks, against 1,126 to 1,261 on each of the seven days before, because the export was pulled mid-morning. Campaign 2007 was renamed from “Search - Hiking Boots” to “Search - Footwear” on 17 August, so a join on name would split it in two. Joined on ID, all eight campaigns match GA4 on every date.
The two tools never show the same conversion count, and Google explains why. Google Ads “reports conversions on the ad impression date,” while other tools “attribute them to the conversion date” (Google Ads Help). In this export, that date is the click date. Conversions can arrive up to 90 days after the click. Google Ads counts clicks; GA4 counts sessions (Google Ads Help). GA4 also credits each purchase under its own attribution model, so some go to email or organic visits (Analytics Help).
Google suggests the “Conversions (by conv. time)” column for comparisons with GA4. It moves each conversion to the day it happened. What is left is the attribution and measurement gap.
Result: over the twelve complete weeks, Google Ads reports 2,710 conversions by click date and 2,722 by conversion time. GA4 records 2,283 purchases: 15.8% and 16.1% fewer. The dating matters only in the incomplete week: 136 conversions by click date, 187 by conversion time and 156 in GA4. That is why last week’s ROAS looked so bad: recent clicks had not converted yet, and the last day was partial.
The account’s normal gap is about 16%. A campaign far outside it is a tracking question, not a performance answer: Display remarketing is 47.1% lower in GA4. If your GA4 figures also disagree with themselves, see why GA4 channel and medium numbers disagree.
How it was checked: a separate script recomputed every total from the CSV files; all matched, including week 13.
Compare campaigns on one basis: conversion-time columns, complete weeks, joined by ID. Over twelve weeks the account spent $84,542.09 for 2,722 conversions worth $316,950.77. That is a CPA of $31.06 and a ROAS of 3.75 (3.73 on the default click-date columns), above the 3.0 target.
Result: Search - Tents takes 21.5% of spend for 10% of conversions, and its CPA of $67.16 is the account’s highest. Yet its ROAS is 3.21, above target, because tents sell for more. Judge on CPA alone and you would cut a profitable campaign. Display remarketing is the laggard: 11.5% of spend, 6.2% of conversions, ROAS 1.98.
On complete data the dip is real but small: weekly ROAS fell to 3.23 in week 9 and 3.10 in week 11. Steps 5 and 6 test whether it is noise or a change worth acting on.
Search - Trail Running looks like a cut. In its last complete week it spent $257.26 for 3 conversions: a CPA of $85.75 and a ROAS of 2.43. But a small campaign swings a lot; one extra sale moves its CPA by a quarter.
Widen the window, then put a rough range around the result. With 20 conversions, chance alone can move the count by about twice its square root either way. That gives a last-four-week CPA range of $34.74 to $88.95, which includes the 12-week CPA of $43.85.
Result: over twelve weeks Trail Running spent $3,069.38 for 70 conversions at a ROAS of 3.28, above target. Its last four weeks hold 20 conversions, below the minimum of 30. The decision is hold. To know whether a campaign worked, you need enough conversions, a window longer than a week, and a result outside the range chance can produce. The same rule applies before you call an A/B test.
Account ROAS fell from 3.94 in weeks 1 to 8 to 3.38 in weeks 9 to 12. To find out why, split every campaign by network and device, and rank each piece by how much it pulled ROAS down. Add landing page if your export has it; this sample does not. This is where ad performance analysis by segment pays off.
Result: the outlier is the Tents campaign on Google’s Search partners network. From week 9 its weekly spend rose from $107.61 to $612.45, while its conversion rate fell from 2.1% to 0.3% and its ROAS from 3.21 to 0.52. That pair explains 50.8% of the decline; without it, weeks 9 to 12 ROAS is 3.65.
To improve performance, fix the segment, not the whole campaign: Tents on Google search still returns 3.21. Turn off Search partners for that campaign and watch it for two weeks. The search terms report in Google Ads shows which searches triggered those clicks.
Apply the step 1 rules to every campaign and write each decision down with its numbers.
The decisions for Monday’s meeting
| Decision | Campaigns | Why |
|---|---|---|
| Scale | Search - Backpacks, Shopping - Clearance, Search - Brand | ROAS of 3.5 or more in both windows, with enough conversions |
| Fix | Search - Tents | Last-4-week ROAS of 2.31; the drop sits on Search partners |
| Cut | Display - Remarketing | ROAS 1.98 over 12 weeks and 1.78 over 4; GA4 sees 47.1% fewer purchases |
| Hold | Search - Trail Running, Shopping - All Products, Search - Footwear | Too few conversions (Trail Running), or between the thresholds |
A scheduled report set to “Only when something happens” checks, on every scheduled run, for the change, threshold or exception you describe. The report goes out, by email, Slack or both, only when it is met. Ours runs every Monday at 9am. It sends only if a campaign’s 4-week ROAS falls below 2.5 on 30 or more conversions, or Search partners take over 25% of its spend. Anomaly suggested an eight-day lag, so each report covers four finished weeks. Scheduled reports need the Pro plan ($25 a month, or $20 billed yearly) or above.
For a one-off copy for the meeting, export the dashboard as a PDF.
If a cost spike reaches the board, see how to explain a CAC spike to your board. If people need a fixed weekly read-out, decide whether they need a dashboard or a reviewed report, automate a weekly report from Google Sheets, or follow the step-by-step weekly marketing report tutorial. For a dashboard people check between reports, see these marketing dashboard examples.
Cross-channel marketing analytics needs one referee. Each platform counts conversions its own way, with its own window, dating and view-throughs. Add them together and you count the same sale twice.
In the sample, the platforms report 3,759 conversions over twelve weeks: 2,710 from Google Ads, 443 from email and 606 from paid social. GA4 credits 2,963 purchases to the same channels, 21.2% fewer. The gap is 15.8% for Google Ads, 26.9% for email and 41.3% for paid social. Compare channels on the conversion credit GA4 assigns under your attribution model, and use each platform’s own numbers only to manage that platform. Meta and TikTok Ads can be connected directly; other networks can come in as CSV or Sheets exports, as paid social does here.
Performance also varies by platform for honest reasons. Paid social can reach people early; if a later search or email visit closes the sale, GA4 may credit that visit. A weaker ROAS in GA4 is not always a weaker campaign.
This analysis shows what happened and where it changed, not what would have happened without the ads. That question is incrementality. It needs a held-back group, such as a geographic experiment where some regions see the ads and others do not (Vaver and Koehler, Google, 2011). It is also not media mix modelling or attribution modelling.
That matters for the Cut row. Remarketing can reach people who would have bought anyway, or GA4 could be missing sales the ads drove. Before switching it off, consider pausing it in some regions first.
Watch for four traps: mixed date ranges, a partial last day or unconverted recent clicks, small samples, and renamed campaigns joined by name.
What people call AI-powered campaign analysis is mostly faster joins and faster follow-up questions. In our run, each step was one request:
Speed is not accuracy. Ask how each number was counted, which dates it covers and which rows it excludes, then check the important ones against the source rows. We asked for clearer output three times: rounded ROAS, readable chart labels and a tighter chart axis.
Anomaly is an AI data analyst for large datasets and spreadsheets. Here it joined the Google Ads, GA4 and channel exports, applied the saved rules, and built the dashboard, decision table and conditional weekly report. Each table and chart opens to its query, and the project’s saved calculations keep the SQL behind each result.
Set a ROAS or CPA target and a minimum number of conversions before launch. Then compare each campaign’s spend with the conversions and value it produced over complete weeks, using one source for decisions and another as a cross-check.
ROAS and CPA decide budget, because they tie spend to results. CTR, CPC and conversion rate explain why ROAS or CPA moved. Use ROAS rather than CPA alone when products sell at very different prices.
Google Ads dates a conversion to the ad, while GA4 dates it to the day it happened. GA4 also counts sessions, not clicks, and credits purchases under its own attribution model. Compare on the conversion-time columns and watch for unusual gaps.
Reporting shows what happened: spend, clicks and conversions. Analysis compares the results with a target, tests whether differences are real, finds what changed, and ends with a decision to scale, fix, cut or hold.
Check and report weekly, but decide budgets on four or more complete weeks so small campaigns have enough conversions. Run a post-campaign analysis after the conversion window closes, not on the campaign’s last day.
Yes, for the joins, comparisons and follow-up questions: it can line up Google Ads and GA4 exports, rank segments and draft a decision table. A person should still set the target, check how each number was counted, and decide.
Founder, Anomaly AI (ex-CTO & Head of Engineering)
Abhinav Pandey is the founder of Anomaly AI, an AI data analysis platform built for large, messy datasets. Before Anomaly, he led engineering teams as CTO and Head of Engineering.
LinkedInContinue exploring AI data analysis with these related insights and guides.
One weekly marketing question, asked from Claude Code through Anomaly’s MCP server and traced to its SQL and rows, with the daily-users trap worked out in the open.
Marketing dashboard vs marketing report: when to give stakeholders a reusable dashboard, when to send a reviewed report, and when the workflow needs both.
Client reporting tools for agencies, compared on one two-client test file: what each tool did for client #2, what it costs at your roster size, and where we could not get in.
