How to measure the impact of marketing changes and prove a change worked

What does it mean to measure the impact of a marketing change?

Measuring the impact of a marketing change means estimating the difference between what happened after the change and what would have happened without it. You never see the second number directly, so every method is a way of building a credible stand-in for it.

There are three levels of evidence, and you should know which one you are using before you report a result:

Most teams only do the first one, then argue in the next meeting. The rest of this article shows how to move up a level with the data you already have.

How do I set up a before/after comparison that holds up?

Use equal windows, cut out the noisy days right after the change and wait until conversions have caught up. A sloppy window can create or hide a 20% effect on its own.

1. Pin the exact change time. Date, hour and time zone. A bid strategy switched at 18:00 on a Tuesday makes Tuesday a mixed day; leave it out of both windows.
2. Use equal windows with the same weekdays. Compare 14 days with 14 days, or 4 full weeks with 4 full weeks, so Mondays are compared with Mondays.
3. Skip the ramp-up. Automated bidding needs time to settle. In its own experiments, Google discards the first 7 days of data to account for ramp-up, and its FAQ describes a 7–14 day period where data can be volatile. Apply the same logic to your manual reads.
4. Wait for conversion delay. Google Ads can report conversions up to 90 days after the click, depending on your conversion window. Recent days look worse than they will end up because some clickers have not converted yet. Check the Conversions > Days to conversion segment to see how long your customers usually take.
5. Write the expected effect first. Note which metric should move, in which direction and by roughly how much, before you look at the result. This stops you from picking whichever metric happens to look good.

Example: a team switches a Search campaign to a new landing page on a Wednesday at 10:00. They compare the 28 days before that Wednesday with days 8 to 35 after it, and only read the conversions once the click dates are older than their usual time to convert.

How do I use a control group when I can't run a test?

Pick something you did not change that normally moves with the thing you did change, and subtract its movement. What is left is a much better estimate of the impact.

Useful controls in marketing:

• A similar campaign or ad group you left untouched
• A region, country or store where the change was not rolled out
• A product line or service with the same seasonality
• The same weeks of last year, for seasonal businesses

Example with illustrative numbers:

Before/after alone says the new page lifted conversion rate by 15%. The control shows the market lifted everything by about 8%, so the page itself explains roughly the remaining 7%. This is the idea behind a difference-in-differences read.

For a more formal version, Google's open-source CausalImpact package builds a counterfactual from control time series. Its documentation is clear about the conditions: the controls must not be affected by the intervention, and their relationship with your metric must stay stable after the change. If those conditions fail, the model can over- or underestimate the effect.

When should I run a real experiment instead?

Run an experiment when the decision is expensive or hard to reverse, such as a new bid strategy, a broad match rollout or a new landing page for your biggest campaign. Running both versions at the same time removes seasonality and market shifts from the comparison.

In Google Ads, custom experiments are available for Search, Display, Video and Hotel campaigns, not for App or Shopping campaigns. You find them under Campaigns > Experiments. Points from Google's help pages that matter for a clean read:

• Google recommends a 50% budget split for the best comparison.
• A cookie-based split shows each user only one version; a search-based split can reach significance faster but users may see both.
• Run the experiment for at least 4–6 weeks, longer if you have a long conversion delay.
• Do not run several experiments at once and avoid frequent budget or target changes while it runs.
• The traffic split cannot be changed once the experiment has started.

If the result comes back inconclusive, that is a result too: the change did not produce an effect large enough to detect at your volume. Google's own advice is to use high-volume campaigns and run longer.

Why do overlapping changes ruin the measurement?

If two changes land in the same window, you cannot separate their effects, no matter how good your method is. Overlap is the most common reason a measurement fails.

Before you read any result, isolate the cause:

1. Put every change on one timeline. Include the days before and after your change: your team's edits in Google Ads, Meta Ads and other platforms, auto-applied recommendations, Smart Bidding or Advantage+ adjustments, website releases, tag and consent changes, CRM or lead-routing changes, price changes and external events.
2. Rule out tracking first. A broken tag, a duplicate conversion or a new consent banner can look like a performance change. Check that conversion counts in the ad platform, GA4 and the CRM still agree.
3. Test one hypothesis at a time. If the timeline shows a second change in the window, either shorten the window to avoid it or report both together and say so.

Your Google Ads change history covers account edits, but not the site, tracking or CRM. Seasonal swings are another overlap; see how to tell seasonality apart from your own changes.

How do I log a change so it can be measured later?

Log the change when you make it, with enough detail that someone can measure it a month later without asking you. Memory and chat threads do not survive the next reporting cycle.

Record at least:

• Date, time and time zone
• Owner and client or project
• What changed, as before → after (for example, target CPA 40 → 32)
• Scope: account, campaigns, pages or regions affected
• Hypothesis and expected effect
• Metric, window and control you will use, plus a review date

GA4 lets you add notes to reports: right-click a data point on a line graph, choose Add annotation and then Create annotation. Titles are limited to 60 characters, descriptions to 150, and a property holds up to 1,000 annotations; you need the Analyst role or above to create them. That is enough for a marker, not for the full record above, and it only covers what you look at in GA4. See the GA4 annotations guide for details.

A shared marketing change log keeps every change, from every platform, on one dated timeline per client. Changeline does this today with manual entries, owners, tags and reasons; you can start a change log for free while it is in early access.

How do I report the impact of a change to a client?

Report the verdict, the evidence level and the decision in five lines. Clients trust a clear method more than a large number.

1. Change: what changed, when, and who made it.
2. Windows: the before and after dates, with ramp-up and delay excluded.
3. Control: what you compared against, or that there was none.
4. Result: the estimated effect and the evidence level (before/after, control, experiment).
5. Decision: keep, revert, or extend the test.

When the result is mixed, say so. "No measurable effect after 6 weeks" is a useful finding that saves budget. If a metric fell instead of rising, the steps in marketing root cause analysis and the guide on how to explain a conversion drop to a client help you structure the conversation.