Lead conversion rate by source: how to measure which channels' leads become customers

Lead conversion rate by source is new customers ÷ real leads, per channel, for one cohort of leads. Count leads by the month they arrived, take duplicates, spam and existing customers out first, and show the counts beside every rate. Cheap leads that rarely buy are expensive customers, and the rate is what proves it.

Most lead reports stop at the inquiry. They say how many calls and forms each channel sent and what each one cost, and the channel with the most leads for the least money wins the meeting. Whether any of those people bought is a question the report cannot answer, because the answer lives in a different system: the invoices, the signed contracts, the jobs in the field software.

This post is for businesses where that gap is the whole problem, because the sale closes on the phone, at the kitchen table or in the office: HVAC, roofing, dentists, law firms. Our post on cost per lead argues why the inquiry is the wrong finish line. This one is about measuring the right one, carefully enough that the number survives being questioned.

The lead conversion rate formula

Lead conversion rate by source = new customers from a source's leads ÷ that source's clean leads, for one cohort of leads and a fixed cutoff date. Both counts are people, not records: a customer with three invoices is one conversion, and a caller who rang twice is one lead.

Three decisions make the number, and each should be written down once and kept, because a rate that changes definition every quarter cannot show a trend:

• The cohort. Which leads are in it, by the date they first contacted you: for example, every lead that arrived between 1 May and 31 July.
• The cutoff. The date by which a lead must have bought to count, set late enough that most buyers have had time to decide.
• The match. How a sale is tied to a lead: the customer's phone or email in the sales records found in the lead records, after both are normalized, and one rule for which lead gets the credit when a customer appears more than once.

The same number goes by several names: close rate, lead-to-customer rate, lead-to-sale rate. Sales teams sometimes mean something narrower by close rate, such as won deals divided by quotes given. That is a useful number for judging the sales process, but it is not this one; a channel whose leads never get a quote should be visible as a channel that failed, not dropped from the denominator.

Sales that match no lead at all, the referrals and repeat customers and calls nobody logged, have no source and no denominator. Leave them out of every channel's rate rather than sharing them around; our post on unattributed revenue covers what that bucket can tell you instead.

Count by lead cohort, not by calendar month

A cohort rate divides the customers who came from one month's leads by that month's leads. A calendar rate divides the sales closed in a month by the leads that arrived in it. They agree only when volume is flat and sales close instantly; otherwise the calendar rate swings with every change in spend.

Take a search campaign whose leads close at a steady 14%, half of them in the month the lead arrives and half the month after. In July the budget goes up and leads rise from 100 to 150; in August it goes back down.

The arithmetic. July's 150 leads produce 21 customers, 10 in July and 11 in August. Sales closed in July are 7 from June's leads plus 10 from July's, 17, so the calendar rate is 17 ÷ 150 = 11.3%. In August, 11 from July plus 7 from August is 18, and 18 ÷ 100 = 18.0%. The cohort rate is 21 ÷ 150 = 14.0% in July and 14 ÷ 100 in August. Nothing about the leads changed. The calendar version says the extra spend attracted worse leads and that August's were the best of the year, and both are false.

How long to wait before reading a cohort depends on your trade. Measure the days from first lead to first sale in pairs you have already matched and set the cutoff where most of them have closed, the 80th or 90th percentile, not the median. A water-heater replacement closes in days; a roof, a dental implant or a personal-injury case can take months, and our post on long sales cycles covers how to report on cohorts that are still closing. Until then, label the rate provisional and say how many days the cohort has had.

Clean the denominator before you divide

The denominator is real prospects, not rows in a call log. Remove duplicates, spam and wrong numbers, job seekers and vendors, and existing customers calling about work they already bought. Keep calls nobody answered. Every row removed raises the rate, so the removal rules are written once and applied to every channel alike.

• Duplicates. The same person calling twice, or filling in a form and then calling, is one lead. Group rows by normalized phone and email first; our posts on duplicate leads and phone normalization cover how, and why a number written three ways is the commonest failure.
• Existing customers. Someone you already serve who calls the number on your ad to book a repair is a real call and a real sale, but not an acquisition. Take them out of both counts, or the channels customers use to come back, usually branded search and mail, look far better at converting than they are.
• Not prospects at all. Robocalls, wrong numbers, job applicants, suppliers. Some channels attract far more of these than others, and fake form submissions cluster on cheap lead forms.
• One person, two channels. Someone who called from a postcard and later clicked a search ad is one lead, counted in one channel's denominator: the same channel that would get the credit if they bought. If the numerator credits the earliest lead and the denominator counts the person in both channels, every rate is wrong in a different direction.

Missed calls stay in

A first-time caller who reached voicemail and never called back is a lead your business failed to convert, not a lead that did not exist. Drop missed calls from the denominator and the channels with the most unanswered calls, often the ones that ring outside office hours, get a better rate exactly where the business performed worst. Count them, and report the contact rate beside the close rate; our post on missed calls sizes what they cost.

Small numbers: show counts before rates

A conversion rate built on a few customers is mostly chance. With 6 customers from 40 leads, the true rate could plausibly be anywhere from about 7% to 29%. Quote a percentage only on at least 50 leads and, ideally, 20 or more customers; below that, show the counts and let nobody rank channels on them.

The uncertainty in a count shrinks with its size, roughly as one over its square root. Six customers carry a relative error of about 100 × √(1 ÷ 6) = 41%; twenty carry 22%; fifty, 14%. That is why the numerator matters as much as the denominator: 2,000 leads with 4 customers is still a rate built on 4 events.

A range says the same thing more usefully than a warning. The Wilson interval is a standard way to put a 95% range around a proportion, and unlike the textbook p ± 1.96 × √(p(1 − p) ÷ n) it never runs below zero for a small count.

Read the ranges before the rates. Mail and Local Services Ads overlap almost completely, so calling one the better converter is a coin toss. Yelp's range covers search, mail and most of Local Services Ads; it could be any of them. Facebook's whole range sits below everyone else's, so that gap is real even on 14 customers. Ranking channels on a column of bare percentages treats all five rows as equally certain, and they are not.

The ad platform's conversion rate is a different number

An ad platform's conversion rate is conversions divided by clicks or other interactions. For a business that closes offline, the conversion it counts is usually a call or a form, so it measures click-to-lead. Lead conversion rate is lead-to-customer. A campaign can lead on the first and trail on the second.

The two numbers multiply. Clicks × click-to-lead rate × lead-to-customer rate = customers. The platform optimizes the middle term because it is the one it can see, which is why campaigns drift toward whatever produces the most inquiries for the money: shorter forms, broader audiences, lead ads that pre-fill a form in one tap. Each of those can raise the platform's rate and lower yours. Platforms can bid on later stages if you send them back, and Google's own advice is to keep the stages apart.

Revenue per lead combines rate and deal size

Revenue per lead = conversion rate × average sale from that channel's new customers, or simply the cohort's revenue ÷ its clean leads. It is the number that compares channels with different deal sizes, and multiplied by gross margin it gives the most a lead from that channel can cost before it loses money.

A channel can win on rate and lose on revenue: branded search and referrals close at high rates because the people calling have already chosen you, often for small jobs. Another can close fewer leads at larger tickets. Revenue per lead settles which is worth more per inquiry. Break-even cost per lead = revenue per lead × gross margin; at a 40% margin, a channel whose leads are each worth $2,000 in revenue can pay up to $800 a lead, and one whose leads are worth $131 can pay $52.50. That ceiling, not the platform's suggested bid, is what a lead from each source is worth to you.

Cost per customer: where conversion rate meets cost per lead

Cost per customer = cost per lead ÷ lead conversion rate, which is the same as spend ÷ new customers. A $40 lead that converts 2.9% of the time costs $1,371 per customer; a $200 lead that converts 25% costs $800. The rate is the bridge between what a lead costs and what a customer costs.

One consequence is worth knowing before anyone argues about cleaning. Removing duplicates and junk raises cost per lead and raises conversion rate by exactly the same factor, so cost per customer does not move: spend and customers are the same before and after. If a channel's cost per customer changes when you deduplicate, a customer was double-counted somewhere. Our post on customer acquisition cost by channel covers that figure in full, including fully loaded costs and payback, and our post on marketing ROI covers the return on margin.

A worked example: five channels, one quarter of leads

An HVAC company takes every lead from May to July 2026 and every sale closed through 30 September, matched on phone and email. Figures are illustrative. Facebook lead ads send the cheapest leads by far and convert the worst; direct mail sends the most expensive leads and converts the best.

First the denominator. The lead export has 1,391 rows across five channels. Duplicates, existing customers and junk come out by the rules above.

The arithmetic. Clean leads are rows minus the three removals: 410 − 60 = 350, 175 − 25 = 150, 600 − 120 = 480, 160 − 40 = 120, 46 − 6 = 40. Facebook loses a fifth of its rows, mostly form resubmissions and fake numbers; mail loses a quarter, mostly existing customers answering a postcard sent to a list that included them.

The arithmetic. Cost per lead is spend ÷ clean leads: $35,000 ÷ 350 = $100, $21,000 ÷ 150 = $140, $19,200 ÷ 480 = $40, $24,000 ÷ 120 = $200, $6,400 ÷ 40 = $160. Conversion rate is customers ÷ clean leads: 49 ÷ 350 = 14.0%, 36 ÷ 150 = 24.0%, 14 ÷ 480 = 2.9%, 30 ÷ 120 = 25.0%, 6 ÷ 40 = 15%. Revenue per lead is revenue ÷ clean leads: $840, $1,200, $131.25, $2,000, $825. Cost per customer is spend ÷ customers: $714.29, $583.33, $1,371.43, $800, $1,066.67. Check one the other way: $40 ÷ (14 ÷ 480) = $1,371.43.

What the rankings say

By cost per lead the order is Facebook, search, Local Services Ads, Yelp, mail. By conversion rate it is nearly reversed: mail, Local Services Ads, then search, with Facebook last by a distance. By revenue per lead mail is first at $2,000, fifteen times Facebook's $131, because mail converts more often and its customers buy bigger systems, $8,000 on average against Facebook's $4,500. By cost per customer Local Services Ads is cheapest at $583 and Facebook dearest at $1,371. The cheapest lead in the business is its most expensive customer.

Now the cleaning. Measured on raw rows, Facebook's leads cost $19,200 ÷ 600 = $32 and convert 14 ÷ 600 = 2.3%; cleaned, $40 and 2.9%. Its cost per customer is $1,371.43 either way. The raw report flattered the price of its leads and understated their quality by the same factor, and only the customer count was ever real.

What not to conclude

Facebook is not necessarily losing money. Its revenue per lead of $131.25 at a 40% margin allows a lead to cost up to $52.50, and it pays $40, so on these numbers it clears break-even with a thin margin; the other four clear it easily. Yelp's 6 of 40 is too few to rank, and mail against Local Services Ads is a tie inside the noise. And none of this says where the next dollar should go, because conversion rates usually fall as a channel is pushed harder; our post on how to allocate marketing budget covers judging the next dollar rather than the average.

CloseRev does the matching that the numerator depends on. It reads two CSV files you export, closed sales and leads, matches each sale to leads on normalized phone and email, and by default credits it to the source of the earliest matched lead. Only high-confidence matches count automatically; medium- and low-confidence ones stay reviewable, and unmatched sales stay in Direct / Unknown. Its reports put each channel's lead count, rows in the lead file contacted in the date range, beside the matched sales and revenue closed in that range. That is a calendar comparison of raw rows, so the cleaned cohort rate described here stays a spreadsheet step.

When a channel's conversion rate drops

A falling conversion rate is a symptom with four usual causes: fewer leads reached, a different mix of leads, a sales team with less capacity, or a source column recording the wrong channel. Only the second is about the channel itself, so check the others before blaming the leads.

• Contact rate. Compare the share of the channel's leads that reached a person, this cohort against the last. A campaign that moved into evenings can lose half its rate to voicemail.
• Lead mix. Split the channel by campaign or service. A rate can fall because a cheap new campaign added low-intent leads while every old campaign converted as before.
• Capacity. In a peak season, slow quotes and full calendars lower every channel's rate at once. If all of them fall together, the cause is probably not marketing.
• Tagging. A rate that collapses overnight often means calls started landing under a different source name, or none. Our post on lead source tracking covers keeping that column honest.

A quarterly routine for measuring conversion rate by source

Once a quarter, take the lead cohort that has had time to close, clean it by the written rules, match it to sales, and divide. Publish counts beside rates, ranges beside small counts, and revenue per lead and cost per customer beside both. The first pass takes a day; after that, a couple of hours.

1. Export every lead from the cohort months and every sale closed through the cutoff, plus earlier sales history to spot existing customers.
2. Normalize phones and emails, group duplicate leads into people, and remove existing customers and junk by the same rules for every channel.
3. Match sales to leads, credit each new customer to one channel by one written rule, and count customers and clean leads per channel.
4. Report rate, count and range together, then revenue per lead and cost per customer, and mark any cohort still closing as provisional.

A lead is a cost and a customer is the result; conversion rate by source is the honest bridge between them. Measure it on cohorts of clean leads, never quote it on a handful of customers, and read it beside revenue per lead and cost per customer. The channel that wins on cheap leads rarely wins on any of the three.