The attribution matrix — creatives on one axis, landing pages on the other, CPL in each cell — is the most useful table in paid social, and the easiest to misread.
The three readings of a cell
Every cell is in one of three states:
| State | Looks like | What to do |
|---|---|---|
| Unmeasured | spend below your judgement threshold | nothing. Keep spending or stop spending, but do not conclude |
| Measurable, bad | spend above threshold, CPL above target, few or no leads | cut the combination, then find out why |
| Measurable, good | spend above threshold, CPL below target | increase budget, and check it holds at higher volume |
Most misreadings come from treating an unmeasured cell as measurable. A cell with €8 of spend and one lead shows a CPL of €8 and means nothing.
Setting the threshold
The threshold is not a matter of taste. It comes from your own target CPL:
Judgement threshold ≈ target CPL × 3
If your target CPL is €10, then a cell needs roughly €30 of spend before a zero-lead result is evidence rather than luck. Three times the target is a rough but useful rule: at that spend, the probability of seeing zero leads from a genuinely good combination is low enough to act on.
A stricter version, when the decision is expensive (killing a creative that might work elsewhere): wait until you would have expected five leads at target CPL.
The four actions a matrix supports
1. Cut the combination, keep the parts. Ad A × LP 2 is bad, but Ad A × LP 1 is your best cell. The problem is the page, not the creative.
2. Cut the creative. Every cell in the row is above target. This is the only case where the creative itself is the finding.
3. Cut the page. Every cell in the column is above target across several creatives. The page is failing independently of what points at it.
4. Scale the combination. A cheap cell with high volume. Before scaling, check the volume: CPL often rises as spend rises, and a cell that was €4.20 at €420 can be €7 at €1,400.
Reading the pattern, not the number
The shape of the matrix carries information the individual cells do not:
- A row that is uniformly good → the creative works, and the page barely matters. Scale the creative, test pages only at the margin.
- A column that is uniformly good → the page carries the performance. Worth finding out why, and applying it to other pages.
- A diagonal → the creative and the page interact. Specific combinations work. Do not generalise.
- Everything mediocre → the problem is upstream: the audience, the offer, or the traffic quality. No page edit will fix it.
What the matrix cannot tell you
- Whether the lead was any good. Lead quality lives downstream, in your CRM. A cheap cell can be the worst cell — see cost per qualified lead.
- Why. The matrix shows that LP 2 underperforms, not that its form is too long. That is the next question, and the AI agent is useful for it.
- What happens next. Every number is historical. Scaling a winning cell changes it.
A working routine
- Weekly: sort cells by spend. Ignore everything under the threshold.
- Cut the combinations above target CPL with enough spend.
- Check the row and column before cutting anything — is it the creative or the page?
- Scale the cheapest cells with volume, in steps, watching whether CPL holds.
- Feed the losers back: a page that loses across several creatives is a page to rewrite, and the automatic A/B test will handle the comparison with the rewrite.
Related
- CPL by creative — how the numbers are produced
- Cost per qualified lead — the check the matrix cannot do
- What is creative attribution — the definitions behind the table