Marketing performance root cause analysis: a repeatable method

What is root cause analysis in marketing?

Root cause analysis (RCA) in marketing is a structured search for the change that made a metric move, instead of reacting to the symptom. The goal is to name one cause you can prove, fix or keep, and prevent the same surprise next time.

The American Society for Quality describes RCA as a systematic process to identify the underlying causes of problems, with the aim of stopping them from recurring. Its five steps translate well to paid media and analytics work:

1. Define the problem: which metric moved, by how much, from when, and where.
2. Collect and analyse data: split the metric into its drivers and segments.
3. Identify possible causes: list every change around the start date.
4. Evaluate and prioritise causes: rank hypotheses by timing, size and scope.
5. Act and follow up: fix or keep the change, then check the metric recovers or holds.

The difference between a symptom and a cause matters. "CPA went up" is a symptom. "The new landing page form added two required fields on the 12th and form completion rate fell" is a cause you can test and reverse.

How do I define the metric shift before looking for causes?

Write a one-line problem statement with the metric, the size of the change, the start date and the scope. A vague problem ("results are worse") produces vague answers; a precise one tells you which changes are even possible suspects.

Before you go further, check the drop is real. Compare the ad platform with your analytics and your CRM or back end. If sales are stable but reported conversions fell, you are looking at a measurement problem, not a performance problem. Also remember that recent days can be incomplete: Google lists delays in conversion reporting among the common reasons for fluctuations in its performance troubleshooting guide.

How do I break a metric into its drivers?

Split the metric into the simpler numbers it is made of, then see which of those moved. This narrows the search from "anything" to one part of the funnel.

Then segment each driver by campaign, device, country, audience, landing page and new versus returning users. A real cause usually has a footprint: it hits the segments it touched and leaves the others alone. If CPA rose everywhere at once, suspect something shared, like tracking, site speed or a budget change at account level.

What changed in the days before the shift?

List every change from roughly one to two weeks before the start date on a single timeline, whoever or whatever made it. Most root cause analyses stall here because the changes live in five different places.

• Your team's edits: bids, budgets, targets, keywords, audiences, creatives, schedules.
• Platform automations: auto-applied recommendations, bid strategy learning periods, automated budget moves.
• Website and tracking: releases, new forms, consent banner updates, tag or pixel edits, checkout changes.
• CRM and sales: lead scoring rules, routing, qualification criteria, pipeline stages.
• External events: holidays, competitor promotions, news, stock-outs, price changes.

In Google Ads, the change history (Campaigns menu, then Change history) lists changes to the account, campaigns and ad groups for the past 2 years, and you can filter by type of change. Google notes that changes by automated or internal systems may appear under users such as "Google Ads system". Our guide to Google Ads change history shows how to read it, and auto-applied recommendations explains one common source of changes nobody on the team made. For Meta, see Meta Ads activity history.

Platform logs only cover their own platform. Site releases, tracking edits and CRM rules have to be recorded by people, which is why teams keep a marketing change log next to the platform histories.

How do I test hypotheses one at a time?

Turn each change on your timeline into a hypothesis, rank them, and test the most likely first. Testing two at once makes the result impossible to read.

A fishbone (Ishikawa) diagram and the 5 Whys, both listed by ASQ as tools for identifying causes, adapt well to marketing. Use these fishbone branches: tracking, platform settings, automation, creative, landing page and site, CRM and sales process, market and competition. Then ask "why" until you reach something someone changed.

Rank each hypothesis with three questions:

1. Timing: did the change happen just before the shift, allowing for learning periods and reporting delays?
2. Size: could this change plausibly explain a move of this size?
3. Scope: does the affected segment match what the change touched?

A hypothesis that fails any of the three moves down the list. For the top one, test it the cheapest way: reverse the change in one campaign, compare a segment that was not affected, or check the metric in a second system. Record what you tested and what happened, even when the answer is "not this".

How do I confirm the cause with segments and platform tools?

Confirm the cause by checking that it explains the pattern in the data, not only the timing. Platform tools can speed this up, but they only see what happens inside their own product.

Google Ads explanations. Explanations give insights into large performance changes in Search, App, Performance Max, Demand Gen, Display and Video campaigns. They can point to bidding, budget, conversion settings or delays, assets and eligibility, targeting, auction competition and search interest. The comparison cannot include today, must sit within the last 90 days, and both periods must be the same length and contiguous.

GA4 anomaly detection. GA4 flags a value as an anomaly when it falls outside the range its model predicted. Per Google's anomaly detection documentation, the training period is 2 weeks for hourly, 90 days for daily and 32 weeks for weekly anomalies. Contribution analysis then surfaces the user segments that contributed most to an anomaly, which you can open in Explorations.

These tools answer "what moved" well. They cannot see a form change on your website, a new lead-scoring rule in your CRM or a competitor's sale. That part of the confirmation still depends on your own timeline. A simple rule: the cause is confirmed when the timing matches, the affected segments match, and reversing or isolating the change moves the metric back.

What does a marketing root cause analysis look like in practice?

Here is an illustrative example, not a real client. It shows how the steps narrow a vague problem to one cause in a short session.

1. Problem: cost per lead on paid search rose 38% week over week, starting on a Tuesday, across all campaigns.
2. Real or tracking? CRM leads also fell, so the drop is real.
3. Drivers: cost per click was flat; conversion rate fell from 6.1% to 4.4%. The problem is after the click.
4. Timeline: Monday, bid target change on one campaign; Tuesday, website release with a new form; Wednesday, auto-applied recommendation adding keywords.
5. Rank: the bid change touched one campaign and CPC did not move. The keywords arrived after the break. The form release matches timing, size and scope.
6. Test: form completion rate by landing page dropped only on pages using the new form.
7. Act: the team restored the old form on half the pages, conversion rate recovered there, and the fix was rolled out.

Notice that the answer was outside the ad platform. For the conversion-specific version of this method, read why did my conversions drop. If you then need to present the finding, the guide on how to explain a conversion drop to a client covers the conversation.

How do I make the next root cause analysis faster?

Log changes when you make them, not when something breaks. Most of the time in an RCA goes into reconstructing who changed what and when, across tools that keep their own partial histories.

• Record every change with a date, owner, platform, affected campaigns and the reason.
• Include changes made outside ad platforms: site releases, tracking edits, CRM rules, promotions.
• Note expected impact and a check date, so you review the result instead of forgetting it.
• Close each investigation with the confirmed cause and the fix, so the next person starts from evidence.

Changeline is a marketing change log built for this: a to-scale timeline per client where your team records the changes they make in Google Ads, Meta Ads, the website, tracking and the CRM, with owner, tags and reason. It is free during early access, and today changes are logged by your team (automatic imports are coming soon). You can start a change timeline for free. If you also use GA4, add GA4 annotations for the biggest events.