sporely.
Get started free

Did Your Campaign Actually Work?

By Görkem Sarptunalı · Last updated · 7 min read

Key takeaways

  • "It went up" isn't proof. Weeks swing on their own, and a bad week is often followed by a better one anyway.
  • Judge a campaign against the stage's usual range over the last 8 to 12 weeks, not against the week before.
  • Decide the metric, the size of the change and the time window before you launch.
  • Check the next stage. More leads that never become opportunities is volume, not growth.
  • The strongest evidence is a holdout: people or regions that didn't get the campaign.
Did it work, or did it just happen? A row of mushrooms of similar height sits inside a band labelled usual range. The last one rises clearly above it: judge a campaign against its stage's usual range, not against last week. Did it work, ordid it just happen?Judge a campaign against itsstage's usual range, not last week.Usual range View full size

Leads jump 30% the week the new LinkedIn campaign goes live, and the Monday meeting decides in four minutes that it works. Three weeks later leads are back where they were, the campaign is still running, and nobody wants to reopen the discussion. Most teams have been in that meeting; the question is how to leave it with an answer you'd still defend a month later.

There's no single magic metric for this. There is, however, a short routine that catches most false wins and false losses, and it works for ads, emails, events, landing pages and pricing changes alike.

Why "it went up" isn't proof

Funnel numbers move every week for reasons that have nothing to do with your campaign. A holiday, an end-of-quarter push, a newsletter that mentioned you, a competitor's outage. If you launch something in a quiet week, almost anything you launch will look like it worked.

There's also a statistical effect worth knowing by name. After an unusually low week, the next one tends to be closer to normal on its own; this is called regression to the mean. Teams usually react to a bad week by launching something, so the recovery that was coming anyway gets credited to the fix.

Finally, most campaigns don't run alone. If the new email sequence, a pricing change and a sales hire all start in the same fortnight, the numbers can't tell you which one did the work.

Step 1: Know your usual range

Before judging any week, know what a normal week looks like for that stage. Take the last 8 to 12 complete weeks, leave out weeks you know were unusual (a bulk import, a holiday), and look at two things: the average and how much a typical week moves around it.

Judge a week against its usual range Sixteen weekly mushrooms. Twelve normal weeks inside a shaded usual range set the baseline. A line marks the campaign start. The first week after launch is still inside the range, which is noise; the last weeks rise clearly above it. Heights are illustrative. Judge a week against its usual rangeWeekly leads for one stage. Heights are illustrative.Usual range: average plus or minus about two standard deviationsCampaign startsClearly above usualInside the range: noise12 normal weeks set the usual rangeIllustrative example. Leave out unusual weeks (holidays, bulk imports) when setting the usual range. View full size

A week is worth attention when it sits clearly outside that range. A common rule of thumb is about two standard deviations from the average; with weekly data, that keeps you from chasing ordinary noise while still catching real changes. For very small numbers, like three or four customers a week, even a big percentage change can be noise, so wait for more weeks before reading anything into it.

Step 2: Decide what "worked" means before you launch

Write one sentence before the campaign starts: which stage it should move, by roughly how much, and by when. For example, "LinkedIn retargeting should lift demo requests about 20% above usual within six weeks."

This feels like admin, but it protects you from the most common trap in campaign reviews, which is picking whichever number happened to go up afterwards. If the target was demo requests and only website visits rose, the campaign didn't do what it was for, even if the visits chart looks lovely.

Step 3: Give it a fair run

Every campaign has a natural delay between launch and effect. Paid ads can show up in leads within days; a nurture sequence needs weeks before its first full cohort has seen every email; anything aimed at closed deals has to wait for the sales cycle.

Judging too early produces "not working" verdicts on campaigns that simply haven't had time. Judging too late wastes budget on ones that clearly aren't moving anything. A sensible default is at least two full weeks for top-of-funnel campaigns and at least one typical sales cycle for anything aimed at opportunities or revenue. Until then, the honest answer is "too early to tell", and it's a perfectly good answer.

Step 4: Look one stage further

A campaign that fills the top of the funnel with the wrong people can make the first number look great and every later number worse. So when the targeted stage rises, check the stage right after it.

  • Leads up, qualified leads up in proportion: the campaign brought real demand.
  • Leads up, qualified leads flat: the campaign brought volume, not quality. Check targeting before scaling it.
  • Leads flat, qualified leads up: the campaign may be improving quality, which is often worth more than volume.

Conversion between stages tells you more than any single stage on its own. The same idea is behind the funnel metrics in our growth guide: a better funnel moves people further, not just in larger numbers.

Step 5: Write down what else changed

Keep a simple log of everything that starts or stops: campaigns, budget changes, form changes, pricing, new hires, outages. One line each, with a date. When a number moves, this is the first thing you check.

If two changes overlap, you can still learn something. Look for the one whose effect lines up with its own timing and its own stage. A form change should move the stage right after the form; a brand campaign should move the top of the funnel first. Where they truly can't be separated, record the result as "unclear" rather than handing the credit to the favourite.

Step 6: Compare with people who didn't see it

All of the steps above compare the campaign with the past. The strongest evidence compares it with a group that lived through the same weeks without the campaign.

Compare with people who didn't see the campaign Two groups over the same weeks. The group that saw the campaign goes from 150 to 230; the holdout goes from 150 to 175. The holdout shows what would have happened anyway, so the campaign's lift is 80 minus 25, or 55. Illustrative numbers. Compare with people who didn't see itChange in both groups over the same weeks. Heights are illustrative.Saw the campaign150Before230AfterHoldout150Before175AfterLift: +80 vs +25, so +55Illustrative example. The holdout shows what would have happened anyway; the gap between the changes is the campaign's effect. View full size

There are a few ways to set this up, from simplest to most rigorous:

  • Region or segment split. Run the campaign in some regions or segments and not in others, then compare how each changed. Comparing the change in both groups, rather than their raw levels, is a method called difference in differences.
  • Platform lift studies. Ad platforms can hold back a random share of your audience and measure the difference for you, as in Google Ads Conversion Lift.
  • A/B tests. For anything inside your product or website, a randomised test gives the cleanest answer. The book Trustworthy Online Controlled Experiments by Kohavi, Tang and Xu (experimentguide.com) is the standard reference on doing them well.

You don't need a holdout for every campaign. Use one for the big, expensive ones, or whenever a result will decide next quarter's budget.

Reading the result

Once a campaign has had its fair run, it usually lands in one of four places.

What you seeWhat it meansWhat to do
Target stage clearly above usual for several weeks, next stage followingWorkingKeep it, then test scaling one variable at a time
Target stage inside its usual range after a fair runNot workingStop or change one thing and test again
Fewer weeks than the campaign needsToo earlyKeep it running and check again on a fixed date
Numbers too small or too many overlapping changesNeed more dataRun longer, combine weeks, or set up a holdout

Common traps

  • Trusting platform-reported conversions alone. Each ad platform counts the conversions it can claim, so totals across platforms often add up to more than your CRM saw. Check against the funnel itself.
  • Picking the time window afterwards. If you choose the weeks that make the campaign look good, you can make almost anything look good.
  • Stopping the moment it looks good. Checking every day and stopping at the first nice spike inflates false wins; this "peeking" problem is well documented in experimentation literature.
  • Scaling three things at once. If budget, audience and creative all change together and results improve, you won't know which one to keep.

Quick Q&A

How long should I run a campaign before judging it?

At least two full weeks for top-of-funnel campaigns, and at least one typical sales cycle for campaigns aimed at opportunities or revenue.

What is a baseline in marketing?

The usual level of a metric before a change, usually the average and typical spread of the last 8 to 12 normal weeks. A campaign is judged against it.

What is a holdout test?

A comparison where a random share of people, regions or segments doesn't get the campaign. The difference between the two groups is the campaign's real effect, also called incrementality or lift.

Can I trust the conversions my ad platform reports?

Use them as a signal, not as proof. Platforms count conversions they can attribute to themselves, so check the result against your own funnel numbers.

My numbers are small. Can I still judge a campaign?

Yes, but more slowly. Combine several weeks, judge the stage with the most volume that the campaign targets, and treat single-week swings as noise.

Keep reading

More from the blog