Anonymous lead and identified lead: when to merge, and when never to
The rule that joins two records into the same person is the most important decision in a CDP, and the easiest to get wrong in both directions. Merging too much mixes two people; merging too little leaves the same person in five. The pattern that works is asymmetric, and this article is the rule written down.
A visitor opens the site in a browser: an anonymous lead is born, with a cookie. They fill the form: the lead gains an e-mail. They buy on their phone, with the same e-mail: a second lead is born, on another device. They open the welcome e-mail on the work computer: a third. One person, three records. Without a rule, the CDP counts three leads and no journey.
Identity resolution is the rule that decides when two records become one. It looks like a technical detail, but it is the foundation of everything we describe in centralized leads — and the place where most implementations fail, because both mistakes look like success for a while.
The two mistakes
| Merging too much | Merging too little | |
|---|---|---|
| Typical cause | Shared device; generic company e-mail; public machine cookie | Only e-mail merges; phone and cookie do not; purchase with another e-mail |
| Symptom | One person “bought” three products they never saw; the history makes no sense | The buyer “never visited the site”; the lead who “never bought” has a purchase in another record |
| Cost | Irreversible without undoing the merge; the audience sends the wrong ad | Inflated lead count; lost attribution; the journey never fires |
| Detectable? | Only by opening a profile and finding it odd | By measuring: how many buyers exist as a lead without the purchase? |
The asymmetric rule
The pattern that survived production — the same one tools like Mixpanel adopted after years of symmetric merging — has three sentences:
- 01A strong identifier merges. A valid e-mail and a valid phone with country code are strong: two records with the same e-mail are the same person.
- 02A weak identifier ties, it does not merge. Cookie and device attach the anonymous to the identified person who showed up in that browser — but NEVER merge two different identified people with each other. If the same computer saw two people logged in, they are two people.
- 03Anonymous never merges identified. A record with no e-mail and no phone has no strength to pull anyone; it is absorbed by whichever identified person appears, and that is all.
A weak identifier has the power to tie, not to decide. What decides is the e-mail or the phone — and only after validation.
Validate before believing
An e-mail “test@test.com”, a phone “99999-9999”, a field holding the form’s placeholder: if that becomes identity, everyone who typed junk becomes the same person. Format validation has to happen BEFORE creating the identity, at every entry door, with the same rule — plus a blocklist for the values that show up a thousand times. Without it, “merging too much” walks in through the front door.
When two people become one: what happens
The merge needs to be a recorded act, not an overwrite: which record survived (the canonical), which was absorbed, when and by which identifier. The absorbed one’s timeline becomes the canonical’s, audiences recount, and the Purchase that goes out via CAPI carries the canonical id as external_id. And the merge must be reversible, because rule 1 also fails: two siblings using their mother’s e-mail are two people.
One number to calibrate expectations: in a base of almost three million leads, recorded merges are about 49 thousand. Most people are never merged; the merge is the exception that resolves the purchase coming from another device. If your rate is far above that, some weak identifier is merging.
In CrazyLeads the resolver applies the asymmetric rule on every source, validates e-mail and phone before creating identity, keeps the blocklist, records every merge in a map that audiences use to canonicalize, and allows undo.