How do I prove content produced leads when I cannot run a controlled test?
You cannot prove it the way an experiment proves it, so you build the strongest available evidence: page-level path data, query-level movement tied to publish dates, and comparison against a set of topics you deliberately did not publish. A holdout does not have to be a formal experiment. It can simply be the cluster you chose to hold back, tracked with the same metrics as the one you published.
Start by admitting the confound
Content is entangled with everything else. Seasonality moves, ad spend moves, brand awareness moves, competitors move. A chart showing organic sessions rising after you started publishing is consistent with the content working and equally consistent with three other explanations.
Saying that out loud is not weakness. It is what makes the evidence you do present believable, and it is the difference between a marketing report and an analysis.
Three levels of evidence, weakest to strongest
- Aggregate correlation. Pages published, sessions up, leads up. Nearly worthless on its own, and it is what most content reporting stops at.
- Page-level path data. Which specific pages appeared in the journeys that produced a lead, how often they were the entry page, and how often the contact action happened on the page itself. Much stronger, because it names the pages.
- A deliberate holdout cluster. Two comparable topic clusters, one published and one held back for a quarter, both tracked. This is the only version that supports a causal claim, and it is available to any business willing to not publish something for three months.
Building a holdout you can live with
Pick two clusters with similar demand, similar competition and similar commercial value. Publish one fully. Hold the other. Track impressions, ranking positions, entry sessions and leads for both. At the end of the period, the difference between them is the closest thing to an estimate of content lift you will get without a research budget.
The objection is real: you are choosing not to pursue demand for a quarter. Weigh that against the alternative, which is spending on content indefinitely with no way to know whether it does anything. Running one holdout once is usually cheaper than the ambiguity.
Measure revenue, not leads
Content-sourced leads and paid-sourced leads do not book at the same rate or carry the same average ticket, so counting leads flatters or punishes content arbitrarily. Follow the lead into the job and the invoice before you compare channels.
That requires the content data and the field service data to live in the same reporting layer, which is the same join that makes any channel comparison honest — see revenue intelligence for how the identifiers are stitched, and marketing intelligence for how the channels get compared once they are.
Topics: measurement · attribution · holdout · content ROI
Have a version of this question about your own business?
The useful answer usually depends on which systems you run and how they're connected. That's a conversation, not a blog post.