Retargeting abandoned registrations from the point the form was left

The registration report says 3,100. The web analytics report says 4,300 people started. Somebody asks the obvious question about retargeting abandoned registrations, which is whether those 1,200 are worth chasing, and the room discovers it cannot answer because nobody knows where the 1,200 stopped.

That is the actual problem. A gap of 1,200 is not one audience. It is at least four, they behave differently, and a single ad pointed at all of them treats a person who typed an email address the same as a person who reached the payment screen and hesitated over a 495 delegate fee.

What counts as an abandoned registration?

You need a floor and a ceiling. The floor is the first step that produces evidence of intent, which is a completed field. Somebody who loaded the form and left in four seconds is a bounce and belongs in site reporting. The ceiling is confirmation.

Between those two, define the steps once and never redefine them mid-campaign. A workable set for a paid trade show runs: form opened with at least one field completed, identity captured with name and email, profile captured with company and job title, qualification answers submitted, ticket or pass selected, payment attempted, confirmed. Seven states, six of which can be the last one you saw.

The abandonment record is then one row per person per session, carrying the last completed step, its timestamp, the pass type in the basket if there was one, and the source that brought them. That row is the whole basis of the audience, and it is worth being strict that the funnel definitions are not yours to invent here. They belong to the registration funnel stages work, and this post assumes you already have them.

Splitting 1,200 abandons by where they stopped

Take the 4,300 starts and 3,100 completions. Completion is 3,100 divided by 4,300, which is 72.1 per cent, and the 1,200 who did not finish split by last completed step like this.

Two things fall out of that table immediately. The 42 people who attempted payment and failed have a payment problem, and a retargeting ad is the wrong tool for a declined card. The 430 who completed one field are barely identified, cannot be emailed, and can only be reached by a pixel-based audience.

The middle of the table is where the money goes. Between identity captured and pass selected sit 728 people who gave you enough to be recognised and who got far enough to be serious.

Abandonment at this scale is normal and it is measured elsewhere. Baymard Institute, in the version of its cart abandonment list published in 2025, puts the average documented online shopping cart abandonment rate at 70.22 per cent, averaged across 50 separate studies. Event registration is a different act with a different price and a different diary consequence, so do not import that figure as a target. Take from it the point that abandonment rates vary by an enormous margin depending on what the measurer counted as a start, which is why your own definition matters more than any benchmark.

Which of the four audiences should you pay to reach?

I would build three audiences and skip one entirely.

Identity captured, no profile. These people gave a name and an email and stopped when the form asked what they do for a living. They can be emailed, so the paid audience here exists mostly to support the email, and the ad should carry the same message as the mail.

Profile or qualification complete, no pass selected. The strongest group. They have invested several minutes and answered questions about themselves. Something about price, date or agenda stopped them. This audience gets the highest bid and the most specific creative, because you know their job title and can name the part of the programme built for it.

Pass selected, no payment. Small, hot, and worth reaching within hours instead of days. Seventy-six people is too few for most platforms to serve efficiently on its own, so it usually has to be merged with the group above and separated again in the reporting.

The one I would skip is the single-field group. You cannot tell those 430 apart from accidental traffic, and a pixel audience of that size drives the frequency up on a pool you cannot describe.

Frequency is the setting most of these programmes get wrong. A retargeting audience of a few hundred people with a daily budget sized for a prospecting campaign will serve the same person the same banner twenty times a week, which is how a registration reminder turns into an annoyance the sales team hears about. Cap it at two or three impressions a day per person and let the budget go unspent if the audience is small. The audience is the constraint here, and spending the budget is not the objective.

Working the recovery rate on the 1,200

Recovery has to be measured against the abandoned pool, at the person level, inside a fixed window. Seven days is a reasonable window for a show with a nine month sell, and the same seven days must apply to everyone in the pool.

The window is a real decision and it deserves five minutes of thought before anyone builds the query. Too short and you miss the person who abandoned on a Friday afternoon and finished on Monday morning, which on a business audience is a large group. Too long and the window swallows the deadline effects that had nothing to do with your ads, so a person who abandoned in March and registered in September when the early bird closed gets counted as recovered. Seven days is short enough to keep the causal story plausible and long enough to survive a weekend.

Run the flight for three weeks and suppose the arithmetic comes out like this. Of the 312 identity-captured abandons, 47 complete within seven days of their drop, which is 15.1 per cent. Of the 340 profile and qualification abandons, 71 complete, which is 20.9 per cent. Of the 76 pass-selected abandons, 23 complete, which is 30.3 per cent. Total recovered is 141 of the 728 you targeted, or 19.4 per cent of the addressable pool.

Now the cost. Reaching 728 people at an average frequency of 9 over three weeks is 6,552 impressions. At a 14 dollar CPM that is 91.73 in media, plus whatever the creative cost, which for a resized existing banner set is usually a couple of hours of somebody's time. Against 141 registrations that is a media cost per recovered registration of 65 cents.

Do not put that number in a board pack. It is the cost per recovered registration on the assumption that every one of the 141 was caused by the advertising, and a decent share of them would have come back anyway, because people abandon forms to find their credit card and finish twenty minutes later. The honest version of this measurement is a held-back cell, which is the subject of its own method in testing whether retargeting caused anything and is not something you can infer from a completion count.

Setting the audience so it expires when the show does

Two settings decide whether this programme stays clean over an edition.

Membership duration is the first. Google's documentation for data segments, current in 2026, allows a maximum membership duration of 540 days and requires a minimum of 100 active users in the last 30 days for a segment to serve. Both numbers matter in opposite directions. The 540 day ceiling is far longer than an edition, so leaving the default in place means people who abandoned before last year's show are still being served ads for this one. Set the duration to the days remaining until doors, or use a rolling 30 day window and rebuild.

The second is exit. A person who completes a registration must leave every abandonment audience within a day, which is a suppression job rather than a retargeting job and has its own daily push back to the ad platforms. Without it you pay to tell people who are already coming that they should come.

Where this stops

The method breaks in three specific places, and all three are worth saying out loud before anyone builds a dashboard on it.

Cross-device abandonment is invisible. Somebody starts on a phone at a client site and finishes on a laptop that evening. The phone session sits in your abandoned pool forever and the laptop registration is counted as a fresh start, so your abandonment count is too high and your recovery rate is too low. If you capture the email at step two you can stitch some of it back together, which is the argument for asking for the email early.

The step definitions decide the answer. Move the qualification questions from step four to step two and the shape of that table changes completely, with more people abandoning earlier and a lower recovery rate on a larger pool. Any comparison across editions is only valid if the form did not change, and forms change every year.

Consent is the third. In markets where a pixel needs consent before it fires, your abandoned audience is only the consenting subset, and the share of registrations that this programme can ever address is capped by your consent rate. Reporting recovery as a share of all abandons when you could only ever reach 60 per cent of them understates a programme that is working.

If you want a first step this week, pull the last completed step for every incomplete registration in the current edition and count the rows in each state. You will have the table above for your own show inside an hour, and the two numbers that matter, the size of the middle of the funnel and the share sitting at payment, will tell you whether this is a marketing programme worth running or a checkout bug worth fixing. That distinction sits underneath the whole acquisition and attribution question and it is cheap to settle.