Rules for double counted registrations across channels when three sources claim one person
When several sources claim one registration, apply a written precedence rule that assigns each registration to exactly one channel, then publish both counts: the exclusive count that sums to your registration total, and the claimed count that lets each channel see every registration it touched.
Add up the channel numbers in your post-campaign pack and you get 14,620. The registration system says 12,400 people registered. Somewhere in that gap sit the double counted registrations across channels, and the meeting is about to spend forty minutes deciding which report is wrong when neither of them is.
A single registration row can carry a paid click identifier, an email send identifier and a partner promo code at the same time, because a real person clicked an ad on Monday, opened your deadline mail on Thursday, and typed a code she got from a trade newsletter when she finally registered. Three channels each count her once, which is arithmetically correct in each report and produces 118 per cent when the reports are stacked.
Where the extra 18 per cent comes from
Take the file apart before deciding anything. Of the 12,400 registrations, suppose 10,320 carry exactly one source identifier, 1,940 carry two, and 140 carry three. The excess claims are 1,940 plus twice 140, which is 2,220, and 12,400 plus 2,220 is the 14,620 in the pack. Multi-claimed rows are 2,080 registrations, or 16.8 per cent of the file.
That distribution is worth looking at every edition, because its shape tells you something. A file where almost nothing carries two identifiers usually has a tracking problem rather than a clean funnel: identifiers are being dropped between the click and the registration row. A file where a third of rows carry three claims usually means the same person is being touched by every channel at once, which is a media planning fact. Nothing is broken in the tracking, and the overlap is what deserves the attention.
The people who appear in several channels are also the people most worth understanding, since they are the ones who needed several touches and got them.
What does a precedence rule have to decide?
Three things, in this order.
Which identifiers are admissible. A click identifier written onto the registration row at submit is evidence. A cookie read from a session that happened on a different device six weeks earlier is a guess. Decide which of the two your rule will treat as a claim before you rank anything.
Whether the session that produced the registration outranks earlier touches. My view is that it should, because the session in which somebody actually registers is the one piece of the journey you observed end to end.
How deliberate acts rank against passive ones inside that session. A person who types a promo code has done something on purpose. A person who arrived through a retargeting click may barely have noticed the ad. Those are different strengths of evidence and a rule that treats them identically is throwing away information you already have.
None of this splits credit into fractions. Assigning a third of a registration to each of three channels is a different method with different uses, and it belongs with weighting every touch in the journey.
The rule I would write
Six lines, applied top to bottom, first match wins.
First, a paid click identifier written on the registration row from a search or prospecting campaign, where the click and the registration are in the same session. That click delivered the person to the form.
Second, a promo code entered by hand on the registration row. Entering a code is deliberate, and it usually means the person is acting on something they read and kept.
Third, a paid click identifier from a retargeting campaign in the same session. Retargeting reaches people already in motion, so it sits below the deliberate act rather than above it, which is where most default settings put it.
Fourth, an email send identifier from a click inside a fixed window, commonly seven days.
Fifth, a tagged source from any other referral in the registering session.
Sixth, everything left over goes to direct, unattributed, or whatever your taxonomy calls the residual.
That third line is the one people argue about, and I would defend it hard. A retargeting click is often the last thing that happened before a registration that was already going to occur, which makes it the default winner under an unexamined last touch rule and the reason retargeting looks miraculous in most event reporting.
Google's own analytics documentation, current in 2026, follows the same logic on the residual, stating that all attribution models exclude direct visits from receiving credit unless the entire path consists of direct visits. Direct is the last resort in a published, widely used rule, and your sixth line is doing the same job.
Whether the whole rule should look at the first touch or the last is a separate argument with a genuine answer on each side, and it is fought out in choosing between the front and the back of the journey. What matters here is that one rule exists, is written down, and does not change between editions.
Publishing two counts instead of arguing about one
The precedence rule produces an exclusive count. Every registration lands in exactly one channel and the columns sum to the registration total. That is the number for finance, for cost per registration, and for the board.
It is also the number that will make a partner angry, because their code appears on 2,050 rows and your report credits them with 1,560. So publish the claimed count next to it.
| Channel | Claimed | Exclusive |
|---|---|---|
| 4,900 | 4,180 | |
| Organic and direct | 3,040 | 2,830 |
| Paid social | 2,180 | 1,730 |
| Partner codes | 2,050 | 1,560 |
| Paid search | 1,460 | 1,300 |
| Tagged referral | 990 | 800 |
| Total | 14,620 | 12,400 |
The claimed column answers what each channel touched. The exclusive column answers who gets the registration. Both are true, they are labelled, and the difference between them per channel is a measurement of how much that channel overlaps everything else.
The gap moves real money. Paid social at 68,000 of spend costs 31.19 per registration on its claimed 2,180 and 39.31 on its exclusive 1,730. Choosing the flattering denominator inside a channel report while using the exclusive total for the show is how a media plan drifts for three editions without anybody noticing.
Keep the ratio between the two columns as a metric in its own right. Email holds 4,180 of its 4,900 claims, a retention of 85.3 per cent, because a large share of mailed people arrive with nothing else on the row. Partner codes hold 1,560 of 2,050, or 76.1 per cent, so roughly a quarter of every partner claim is shared with a channel that outranks it. Paid social holds 79.4 per cent. A channel whose retention falls sharply between editions is usually being scheduled on top of everything else instead of reaching anybody new, and that is visible here long before it shows up in a cost per registration.
Partner conversations get easier with both numbers on one page, and the same discipline applies when a trade title's redemptions and tagged clicks disagree, which is handled in one code per partner per placement.
Why can the platforms not do this for you?
Because each platform de-duplicates only against itself, and it does that well.
Meta's documentation for the Conversions API, in its 2026 version, describes exactly how: a server event and a browser event are matched on event_id together with event_name, or on event_name with the fbp browser identifier or an external identifier, and events are only de-duplicated when they are received within 48 hours of the first event carrying that identifier. It is a precise rule with a stated window, and it applies to one platform's own two pipes.
Nothing in that mechanism can see a promo code redeemed on your registration form or an email send identifier from your ESP. Two platforms both reporting a conversion for the same person have no way to discover each other, so each one counts it, honestly, in isolation.
Your registration file is the only place where every identifier lands on the same row. That makes cross-channel de-duplication an organiser's job by construction, and it is why the precedence rule has to live in your warehouse rather than in anybody's ad account.
Where this stops
A precedence rule is a reporting convention and it does not measure causation. The channel that wins a row under any ordering is the channel your rule chose, and swapping lines three and four will move hundreds of registrations between two teams' numbers without a single person behaving differently.
The second limit is identifier coverage. A rule can only rank the claims that exist on the row, so a registration arriving with nothing on it goes to the residual regardless of what actually happened, and the size of that residual caps how much the rule can do for you.
The third is that the claimed count is only complete for channels that write an identifier. Print, radio, an exhibitor mentioning the show on a call, and a colleague forwarding a link with the parameters stripped all appear as nothing. When somebody points out that the claimed column sums to 118 per cent, the fair answer is that it would sum to considerably more if the untracked channels could claim anything at all.
Start this week by counting how many registrations in the current edition carry more than one source identifier, and split that count by which pairs occur most often. Twenty minutes of that query tells you whether you have a de-duplication problem worth writing a rule for, or a tracking problem that has to be fixed before any rule is worth having, and either answer improves the rest of your acquisition and attribution reporting.
Questions people ask about double counted registrations across channels
- Why do channel registration counts add up to more than the total?
- Because one registration row can carry several identifiers at once: a paid click id, an email send id and a partner code, all written at submit. Each channel report counts the rows carrying its own identifier, so a person who clicked an ad, opened a mail and typed a code appears three times across three reports.
- What is a precedence rule for registration attribution?
- An ordered list that decides which identifier wins when a registration carries more than one. A workable order puts evidence from the session that produced the registration ahead of earlier touches, and inside that session ranks the deliberate acts above the passive ones. The order matters less than writing it down and applying it every edition.
- Should partners be shown the exclusive count or the claimed count?
- Both, on the same page. The claimed count is the honest answer to what a partner touched and is the number they will recognise. The exclusive count is what sums to your registration total and what finance can use. Showing only one of them starts an argument that the other would have settled.
Related reading
- Running multi touch attribution for events on a single registration file
- First touch versus last touch on a show that sells for nine months
- Media partner promo codes that actually tell you which title delivered