Measurement plans for a privacy-first web
Consent mode, modelled conversions and server-side tagging changed what your dashboard is really showing you. A grounded look at what still holds up.
Somebody senior will eventually ask what the content programme returned. The honest answer involves more uncertainty than most marketing teams are comfortable saying out loud — which is exactly why so many of them reach for a number they cannot defend.
Attribution models disagree by design
Start here, because everything downstream depends on it. A last-click model and a data-driven model are not two attempts at the same truth, one of which is more accurate. They are two different questions.
Last-click asks: what was the final touch before conversion? Data-driven asks: given the paths in our data, how much credit does each touchpoint statistically appear to carry? These will disagree. They are supposed to disagree. Presenting one of them as "the ROI" without saying which question it answers is not reporting; it is picking a number that suits the narrative.
The first thing we do on a measurement engagement is get everyone to agree, in writing, which question the business is asking. Everything gets easier after that.
What content genuinely can be measured on
Content in a considered purchase has three roles, and each is measurable to a different degree of confidence:
- Acquisition. Did this page bring someone to the site who was not otherwise coming? Organic entrances, new users by landing page, assisted first touches. Reasonably measurable.
- Conviction. Did engaging with this content change the probability of conversion? Measurable, with effort — cohort comparisons between content-exposed and non-exposed users. Confounded, but honestly confounded.
- Efficiency. Did content reduce cost elsewhere? Fewer support tickets, shorter sales cycles, less paid spend needed for the same pipeline. Very measurable, and consistently ignored.
Most content reporting only attempts the first, then implies the second, and never mentions the third. That is why it feels unsatisfying to the person holding the budget.
The modelled portion, stated openly
Consent mode, cross-device gaps and privacy-first browser defaults mean that a meaningful share of the conversions in your analytics platform are modelled, not observed. That is not a scandal. It is the current state of measurement, and modelling is a reasonable response to it.
What is a problem is presenting a modelled figure with the same confidence as an observed one. When we build a reporting view, we mark which figures are observed, which are modelled and roughly what proportion. Stakeholders handle this far better than marketers expect. What erodes trust is not uncertainty — it is discovering later that the uncertainty was hidden.
A defensible way to talk about content value
The approach we use has three parts, and it deliberately produces a range rather than a point.
- Contribution, not causation. "Organic pages contributed to 34% of pipeline-generating sessions in the period" is defensible. "Content generated £X in revenue" is not, unless you have run something close to an experiment.
- Comparison against a baseline. What did the equivalent period look like before the programme? What did pages we did not touch do over the same window? A control group of untouched pages is the closest most content teams will get to an experiment, and it is worth building deliberately.
- Assumptions written down. Every model rests on assumptions — attribution window, lead value, close rate, what counts as engaged. Put them on the same slide as the number. If someone disputes an assumption, change it in front of them and show the new range. This turns an argument into a conversation.
What to do when the number is bad
Say it. In the monthly review, with the data, before anyone else finds it.
The instinct is to reach for a metric that went up. Impressions, engagement, time on page — there is always something. This buys a month and costs the relationship, because the person you are presenting to is not stupid, and they will remember that you showed them impressions in the month the pipeline dropped.
A programme that reports honestly through a bad quarter is trusted in the good ones. That trust is, in the end, the thing that lets a content programme survive long enough to work at all.
The uncomfortable summary
Content ROI in a multi-touch, committee-led purchase cannot be measured with the precision that a board would like. Anyone claiming otherwise is either running a very simple business or overselling.
What can be done — and what we think is genuinely more useful — is to produce a defensible range, state the assumptions, show the direction of travel, and be specific about what would change your mind. That is a harder sell than a confident number. It is also the only version that survives scrutiny.
A note on claims. Nothing in this article should be read as a guarantee of results. Marketing outcomes depend on your market, product, budget, timing and team. We describe methods we use and what we have seen them do — not predictions of what would happen for you.
