What paid social for event registrations actually delivers once you match the file

The social platform reports 1,842 registrations for the edition. The registration file, joined on every identifier the campaign passed through, holds 774 rows that can be traced back to it.

Neither system is broken. They are counting two different things, and the media plan is about to be built on whichever one is quoted first in the meeting.

Four products, one line on the plan

What gets called paid social on a media plan is at least four different purchases, and their prices are not comparable.

Prospecting to a site registration. An impression served to somebody who has never touched the show, a click to the registration page, a form completed on your own domain. Expensive per registration and the only version that reliably adds people.

Prospecting to an in platform lead form. The same audience, but the form opens inside the app and never leaves it. Cheap per submission and it produces a lead record rather than a registration.

Retargeting to a site registration. Served to somebody who already visited the registration page. Cheap per registration and largely made of people who were already in motion.

Retargeting to an in platform lead form. Cheapest per submission of all four, and the version most likely to produce a number that flatters everybody.

Keep these as separate campaigns rather than ad sets in one campaign. Shared budgets and shared optimisation goals mean the platform will move spend towards whichever inventory converts fastest, which is always the retargeting, and your reporting will then attribute the campaign's efficiency to a decision nobody made.

What do two ad sets look like once you match the file?

Take one edition and two ad sets, one of each type, with the platform numbers and the file numbers side by side.

Ad set A is prospecting to the registration page. Spend 12,000. It served 240,000 impressions, produced 3,000 clicks at a click through rate of 1.25 per cent, and 2,400 of those clicks became landing page views, so a fifth of the clicks never loaded the page. 720 people started the registration form and 288 completed it, a completion rate of 40 per cent. Cost per completed registration is 12,000 divided by 288, which is 41.67.

Ad set B is retargeting to an in platform lead form. Spend 8,000. It produced 2,000 form opens and 900 submissions, a submission rate of 45 per cent. The platform reports 900 leads at 8.89 each, which is the figure that will be repeated in the meeting. Of those 900, 486 can be matched to a row in the registration file. Cost per matched registration is 8,000 divided by 486, which is 16.46.

So the honest comparison is 41.67 against 16.46, and the platform's version of the same comparison is 41.67 against 8.89. The gap between 16.46 and 8.89 is 414 people who submitted a lead form and never became registrations.

One more number changes how you read all of this. Of the 486 matched registrations from ad set B, 402 had a site session on the registration page before the first impression of the retargeting ad was ever served. That is 83 per cent of the ad set's output arriving from people who had already found the show and started the process.

Why does the retargeting number flatter itself?

Because the audience is defined by prior intent, and the measurement gives the ad credit for the intent.

The 402 people above visited the registration page, left, saw an ad, and came back. Some of them came back because of the ad. Some would have come back anyway, because they left the page to check a diary, ask a manager, or find a card. A cost per registration of 16.46 treats every one of them as bought.

Li and Kannan, in the Journal of Marketing Research in 2014, modelled channel consideration and visits over time on individual level data with a field experiment to validate it, and found that channels' relative contributions differed significantly from the metrics then in use. Their model also identified cases where retargeting decreased conversion probability, which is the uncomfortable version of this and not something any platform report will surface.

Gordon, Zettelmeyer, Bhargava and Chapsky put the general problem plainly in Marketing Science in 2019, working with large field experiments at Facebook: observational methods often fail to recover the treatment effects that randomised experiments produce. Any number computed by comparing people who saw ads with people who did not inherits that failure, and that includes most of the lift reporting available in the buying interfaces.

None of this means retargeting is worthless. It means the 16.46 is an accounting figure and not a purchase price, and the difference between those two is the entire reason an incrementality test on the retargeting audience is worth designing before the budget is set.

Lead forms produce leads, and the gap is measurable

The 414 people who submitted ad set B's form and never appeared in the registration file deserve their own examination, because they split into causes you can act on.

Some submitted the form with a personal email that does not match the corporate one they later registered with, so the match failed rather than the person. Some completed a form that only collected name, email and company, and were never asked the questions your registration requires, so a second step was needed and never happened. Some tapped submit on a prefilled form without much intent, which is the known cost of removing friction.

Measure the split once and the design decision becomes obvious. If most of the gap is match failure, fix the join by passing a click identifier through the lead export and matching on hashed email plus company domain, and the lead form stays. If most of the gap is people who never completed the second step, the lead form is buying you a list rather than an audience, and it should be reported and budgeted as a list building activity.

Timing matters here too. A lead submitted in March that becomes a registration in August is a success with a five month lag, and a monthly report will record it as a failure twice before it records it as a success. Report lead to registration conversion by cohort with the lag shown, not as a single ratio in the month the spend happened.

What to publish by placement and objective

Six lines, per edition, per campaign group: spend, impressions, clicks or form opens, platform reported conversions, matched registrations in the file, and cost per matched registration.

Then two derived numbers that do most of the work. The match rate, which is matched registrations divided by platform reported conversions, at 288 over some larger platform figure for ad set A and 486 over 900 for ad set B. And the prior intent share, meaning the proportion of matched registrations whose first site session predates the campaign's first impression to them.

A low match rate points at the join or at the platform counting things you would not count. A high prior intent share points at a campaign harvesting demand somebody else created. Both are answerable questions with owners.

Placement level reporting is worth the extra rows, because in feed, stories and audience network inventory fill at prices that differ by a factor of several, and a single ad set average hides which one produced the registrations. The same applies to the objective: an ad set optimised for landing page views and one optimised for conversions will buy different people at different prices from the same audience definition.

Hold the reporting period to the edition rather than the calendar month. A show that opens in October has a campaign running from February, and a monthly view chops it into eight reports that each look like a failure or a triumph depending on where the deadline fell. Edition to edition comparison on the same weeks before doors is the only version that supports a decision about next year, and it needs the week number relative to show open stored on every spend row.

Where this stops

Everything above depends on matching, and the match is weakest exactly where the audience is most valuable. Senior buyers use personal accounts on social platforms and corporate addresses on registration forms, so the rows that fail to join are disproportionately the ones the sales team cares about. A match rate of 54 per cent, as in ad set B, is not a claim that 46 per cent of those people did not register.

The platform is also measuring itself. It decides what an impression was, what a conversion was, and which of those to attribute to itself, and it has a commercial interest in the answer. No accusation of bad faith is required for that to matter. It is reason enough to hold the file as the system of record and to treat the platform figure the way you would treat the narrative on a supplier's invoice.

Cost per matched registration says nothing about who registered, and paid social is the channel most capable of producing large volumes of the wrong audience cheaply. A campaign that halves your cost per registration by finding people who will never walk the floor has made your acquisition reporting better and your show worse.

The parallel questions for search, where the substitution problem is different and better evidenced, sit with the three group split of search spend, and the question of whether an impression alone should ever earn credit sits with view through counting. Both connect back to the acquisition and attribution pillar.

Pull one retargeting ad set from your last edition this week, match its reported conversions to rows in the registration file, and then check how many of those rows had a registration page session before the first impression was served. That single percentage tells you what you have been buying.