One customer, three leads: deduplicate before you credit anyone

One customer often arrives as three leads: a form, a call and a chat, sometimes from two campaigns. Count people, not records, by matching on normalized phone and email; credit each sale once, to one lead chosen by a rule you set in advance; and report how many records were duplicates, because that number changes every channel's cost per lead.

Every business that runs more than one channel has duplicate leads, and most of them do not know how many. The duplicates are not errors in the usual sense. Each record describes a real contact: somebody really did fill in that form, really did call that number, really did open that chat. The error is in what happens next, when the records are counted as if each one were a different person.

That counting error spreads into everything built on top of it. Lead volume looks higher than it is. Cost per lead looks lower. Conversion rates look worse, because the same number of customers is being divided by more leads. And when sales are credited back to sources without a rule, a single customer can be claimed by every channel they touched.

How one person becomes three leads

A duplicate lead is a second or third record for somebody who is already a lead. It appears whenever one customer contacts you through more than one route, or through one route more than once, and each contact creates its own record in whatever system received it.

The common patterns are all ordinary customer behavior:

• Form, then call. Somebody submits a quote request from a Facebook ad, gets impatient, and calls the number on your site two days later. The form is in your CRM; the call is in your call tracker.
• Two campaigns. The same person clicks a search ad in June and a retargeting ad in July and fills in the form both times.
• Chat, then form. The chat widget captures an email address, the visitor leaves, and returns the next week to book online.
• A lead marketplace and you. A homeowner requests quotes through a lead marketplace, then finds you directly and calls.
• The same channel, twice. Somebody submits a form, hears nothing for an hour, and submits it again.

None of these is rare. The more channels you run and the slower your response, the more of them you get, which means the businesses with the most duplicates are often the ones spending the most on leads.

Why your systems do not catch them

Each system deduplicates on its own key, inside its own walls. A CRM folds a new contact into an existing one with the same email address; an ad platform counts one conversion per click; a call tracker counts callers. None of them sees the other systems' records, so the same person passes through every check as somebody new.

That rule is sensible for an email marketing tool and it leaves a gap exactly where local and sales-led businesses need it closed. A form fill carries an email. A phone call carries a phone number and nothing else. The CRM cannot know that the caller and the form-filler are the same person, because the only key it deduplicates on is missing from one of the two records. If you use HubSpot or any similar CRM, assume that calls and forms from the same person are sitting in different places.

"One per click" is deduplication within one platform and one click. The same person clicking two different ads, or clicking a Google ad and later a Facebook ad, is counted again by each. That is not a flaw in either platform; it is the boundary of what a platform can see. It is also why the sum of every platform's reported leads is always larger than the number of people who contacted you.

Every system deduplicates inside its own walls, on its own key. The only place all of a customer's contacts can be seen together is a file you assemble yourself, so that is where deduplication has to happen.

Deduplicate on the person, not the record

To deduplicate across sources, put every lead from every system into one file with a source column, normalize the phone numbers and email addresses, and treat any two records that share either identifier as the same person. Keep every original row; the deduplication is a decision about which row counts, not a deletion.

1. Combine the sources. Export forms, calls, chats and marketplace leads into one file, one row per contact, with the date and a source or channel column on every row.
2. Normalize the identifiers. Convert every phone number to one format, ideally E.164 with the country code, and lowercase and trim every email. Without this, "(555) 010-2030" and "+15550102030" look like two people. Phone normalization is the single step most likely to go wrong.
3. Link on phone or email. Two rows that share a normalized phone or a normalized email belong to the same person. Chains count: if the form shares an email with the chat and the chat shares a phone with the call, all three are one person.
4. Do not merge on names. "J. Smith" and "John Smith" might be one person or two. A name can support a match a person reviews; it should never create one on its own.
5. Pick the counting row by a written rule. Usually the earliest contact. Mark it, keep the others, and record the rule in the report.

Do this outside your CRM, or at least without merging inside it. A CRM export should arrive as the system holds it, because a merge made to tidy up attribution usually discards the losing record's source field, and with it the evidence of which channel the person came through first.

What duplicates do to cost per lead

Cost per lead divides spend by records. Once duplicates are removed, it divides spend by new people, and the channels whose leads are mostly people who already knew you get noticeably more expensive. The ranking can change even though not a single dollar of spend did.

Take a month with 400 raw lead records across three sources. Deduplicated on phone and email, they turn out to be 310 people. Each person is assigned to the source of their earliest contact.

The arithmetic: $7,000 ÷ 160 = $43.75 and $7,000 ÷ 140 = $50.00; $9,000 ÷ 150 = $60.00 and $9,000 ÷ 105 = $85.71; across the whole month, $16,000 ÷ 400 = $40.00 against $16,000 ÷ 310 = $51.61. Ninety of the 400 records, 22.5%, were people who had already contacted you.

Meta's cost per lead rises by 14%. Google's rises by 43%, because 45 of its 150 records were people who had first arrived through Meta or the website. That is a common shape: search ads catch people looking for a business they have already heard of, and so collect a larger share of second contacts. It does not mean search is wasteful; it means its raw cost per lead was describing a different thing from Meta's.

A channel full of second contacts looks cheap per record and expensive per new person. Report both figures, and the share of records that were duplicates, so nobody cuts the channel that finds people to fund the one that collects them.

First touch or last touch: which lead gets the sale

Once one person has several leads, a sale needs a rule to pick one of them. First touch credits the earliest contact, the source that brought the person to you. Last touch credits the latest contact before the sale. Both are defensible; they answer different questions, and they can move a large share of revenue between channels.

Continue the example. Of the 310 people, 40 bought, at an average of $3,000. Credit each sale to the person's earliest lead and then to their latest lead before the sale, and the channels trade places:

Total revenue is $120,000 under both rules, and that is the property that matters most: each sale is counted once. What moves is $21,000 of credit — seven sales — from Meta to Google and the website. Under first touch, Meta returns $54,000 on $7,000, or 7.7 to one. Under last touch it returns $33,000, or 4.7 to one. Neither figure is wrong. One describes who found the customer; the other describes who was last in the room.

We think first touch is the better default for a business deciding where to spend, because the expensive problem is finding new customers, and the last contact before a sale is disproportionately the main phone line, the chat window and branded search — routes people use once they have already decided. We set out the wider argument in our post on attribution models. What is not defensible is choosing after seeing which rule flatters the channel you wanted to keep.

The rule you must never use is no rule at all: crediting the sale to every lead that shares the customer's phone or email. In this example that hands out credit for more than 40 sales, and every channel's return on spend is overstated at once. It is the same arithmetic that makes the platforms' reported totals add up to more than your revenue.

First touch and last touch move credit between channels; they never change the total. A report that credits one sale to three leads has changed the total, and that is the one mistake that makes every other number in it wrong.

What deduplication cannot fix

Matching on phone and email finds the same person when they used the same identifiers. It cannot link a customer who called from a work phone and bought with a personal one, or a household where one partner made the inquiry and the other signed. Those stay separate, and the honest report says so.

This limit is worth stating plainly because the temptation is to close it with guesses: merge on surname and ZIP code, link records within the same street, treat similar names as one person. Each guess makes the duplicate count look more complete and makes the attribution less true. A sale that cannot be linked to a lead on a real identifier is better left unattributed than credited to a lead that might have been somebody else.

• Ask for the phone number on every form. It is the identifier most likely to appear in both the lead and the sale.
• Capture the caller's number on every call. A call log without numbers cannot be deduplicated or matched.
• Record which identifier made each link. An email match and a phone match are both strong; anything weaker belongs in a review queue, not in the totals.
• Watch the [match rate](/blog/match-rates). If it falls after a change to a form or a phone system, an identifier has stopped arriving.

CloseRev takes one lead file per import, so forms, calls and chats go into a single file with a source column. For each sale it looks for leads with the same email first and, failing that, the same phone number in E.164, then credits the earliest of those leads; that is first touch, and it is the only deduplication it does. It does not merge on names or households, and sales it cannot match stay in Direct / Unknown.

A short routine for every reporting period

Deduplication is not a one-off cleanup. Done every period, it takes minutes and keeps three numbers honest: how many new people each channel brought in, what each of them cost, and which single lead each sale is credited to.

1. Export every lead source for the period, plus at least one sales cycle before it, into one file with a source column.
2. Normalize phones and emails, and link records that share either one.
3. Count new people per channel by earliest contact, and report the share of records that were duplicates.
4. Credit each closed sale to one lead by the rule you chose in advance, and check that credited revenue does not exceed total revenue.
5. Leave sales with no linked lead in their own line rather than spreading them across channels.

The output is smaller than the platforms' figures, and it is supposed to be. Every lead in it is a person, every sale is counted once, and every number can be traced back to the rows that produced it.