Most established service businesses know their cost per lead to the cent and their cost per customer not at all. Here's why that's backwards, with the example that makes it obvious.
What's the difference between cost per lead, cost per acquisition and cost per customer?
Three terms get used as if they're interchangeable. They measure three different things.
| Metric | Formula | What it actually measures | Who usually reports it |
|---|---|---|---|
| Cost per lead (CPL) | Ad spend ÷ leads | How cheaply you can collect contact details | The ad platform, the agency |
| Cost per acquisition (CPA) | Ad spend ÷ "conversions" | Whatever event you told the platform to count as a conversion. Often just a lead; sometimes a booked call; almost never a paying customer | The ad platform |
| Cost per customer (CAC) | (Fees + media + sales time) ÷ new paying customers | What it costs to win a client, all in | Almost nobody, unless you build it |
The trap is in the middle row. "Cost per acquisition" sounds like cost per customer, and on most ad dashboards it isn't. A Meta campaign optimised for "leads" reports a CPA that is simply your CPL under a different name. Unless the conversion event is a closed sale fed back from your CRM, CPA tells you nothing about customers.
Cost per customer is the only one of the three that includes the leak. It counts the agency fee and the media, but also the sales hours consumed and, crucially, the leads that never became anything. A lead that's never contacted, never qualifies, never shows or never closes still cost money. CAC is where that money reappears.
Why a "healthy" cost per lead can hide a broken business
The most common line I hear from founders is some version of: "Our cost per lead is fine, but what is the cost per customer?" They're asking the right question. Here's why the first half of that sentence can be true while the business is quietly losing money.
Take a home-improvement business spending $6,000 a month on Meta ads. The agency "optimises" the account: broader targeting, a native lead form instead of a landing page, a cheaper offer hook. Cost per lead halves. The report looks like this:
| Before optimisation | After optimisation | |
|---|---|---|
| Ad spend | $6,000 | $6,000 |
| Cost per lead | $60 | $30 |
| Leads | 100 | 200 |
| Contact rate | 70% | 55% |
| Qualified (of contacted) | 50% | 35% |
| Close rate (of qualified) | 20% | 13% |
| New customers | 7 | 5 |
| Cost per customer (media only) | $857 | $1,200 |
Cost per lead halved. Cost per customer rose 40%. Benchmarks make it worse: the average Facebook cost per lead was $27.66 in 2025, so a $30 lead looks like a job well done. And that's before counting sales time: the team worked twice as many leads to close fewer deals, so the true CAC gap is wider still. Every number on the agency's dashboard improved. The only number that matters got worse, and nobody was tracking it.
This isn't a hypothetical failure mode. It's the default outcome of optimising for CPL, because the cheapest way to lower cost per lead is to lower the bar for what counts as one: wider audiences, less friction, weaker qualification. Cheap leads and good leads are usually not the same leads. CPL is a diagnostic, not the scoreboard. I've sat in the meeting where the ad team celebrated the cost per lead while the sales team quietly binned the leads. Both were telling the truth. Only one of them was looking at a number connected to revenue, and it wasn't the ad team.
When cost per lead is the right number to track
CPL isn't useless. It's mis-assigned. There are three jobs it does well, and they're all inside the ad account:
- Comparing creatives or audiences within one channel, when everything downstream is the same. If two ads feed the same form, the same follow-up and the same sales team, the one with the lower CPL is probably the better ad, though you should still check lead quality a month later.
- Early-warning. If CPL jumps 40% in a week, something broke: creative fatigue, a tracking issue, a competitor bidding up the audience. CPL moves daily; CAC moves monthly. Use the fast signal to catch problems and the slow one to make decisions.
- Early testing, when customer counts are too small to mean anything. Three customers in a month is noise; you can't judge a new channel on it. CPL and cost per qualified call give you something to steer by until the customer numbers accumulate.
The rule: CPL for decisions inside the ad account, CAC for decisions about the budget. Which creative to run is a CPL question. Whether to spend more, change agency, or move channels is a CAC question, and answering it with CPL is how businesses end up in the table above.
The cost ladder: four numbers between an ad and a customer
The reason CPL misleads is that it's the first rung on a ladder, and each rung exposes a leak the one before it hid.
- Cost per lead — spend ÷ leads. Hides everything after the form.
- Cost per qualified call — spend ÷ calls held with people who fit, have the problem and can pay. Exposes contact rate, qualification and show rate. This is the first number worth judging marketing on.
- Cost per customer — all acquisition cost ÷ paying clients. Exposes close rate and sales time.
- Payback — CAC against first-year gross profit per customer. Exposes whether the whole thing is worth doing at all.
Most businesses report rung one and argue about rung four. The two rungs in between are where the revenue actually leaks, and they're also where the fixes are cheapest: faster follow-up, a qualification gate, a booking path that sets expectations. When you only track rung one, every problem on rungs two and three gets misdiagnosed as an ad problem and "fixed" with more spend or new creative, which is why the CPL keeps improving and the business doesn't.
How to start tracking cost per customer this month
You don't need attribution software to get a usable CAC. You need five things done consistently.
- Count all the costs. Agency fees, media, software, and sales time at a loaded hourly rate. Sales time is the one everyone skips, and in the example above it's where half the damage was hiding.
- Count customers by source. A "how did you hear about us?" question at intake, and a source field in the CRM that sales must fill before a deal can be marked won. Imperfect self-reporting beats no data.
- Match by month with a lag rule. Attribute each customer to the month their lead arrived, not the month they signed. Otherwise a slow quarter looks catastrophic and a fast one looks like genius.
- Compute per channel, not just in total. Referrals, paid social, search, and outbound will have wildly different CACs. The blended number hides which one is subsidising the others.
- Put it next to first-year gross profit per customer. This turns CAC from a number into a decision: cheap at any spend, or expensive at any spend.
If you'd rather see the arithmetic before building the reporting, the Revenue Leak Calculator walks your current numbers through the ladder and shows what each rung is costing you. It takes about three minutes and usually settles the CPL argument on the spot.
What to do when CPL and CAC disagree
Once you have both numbers, they'll point in different directions surprisingly often. The combination tells you where the problem is:
- CPL fine, CAC bad. The leak is downstream: speed to lead, qualification, show rate or close rate. Don't touch the ads. Fix the follow-up and the booking path first; that's where most of the money dies.
- CPL bad, CAC fine. The "expensive" leads are converting. Leave the campaign alone and consider scaling it. This is the setup agencies get fired for while it's making the client money.
- Both bad. The problem is upstream: offer, audience, or creative. This is the one case where the ad account is genuinely the fix.
- Both good. Scale spend until CAC starts rising, then stop. That rising point is your channel's real capacity, and no benchmark can tell you where it is.
The mistake is treating every one of those four situations as an ad problem because CPL is the only number on the report. Track both, and the argument between marketing and sales mostly disappears, because they're finally looking at the same system.
Frequently asked questions
Is cost per acquisition the same as customer acquisition cost?
Usually not, despite the similar names. On ad platforms, cost per acquisition means cost per whatever conversion event you configured, and for most service businesses that event is a lead or a booked call, not a sale. Customer acquisition cost means total acquisition spend divided by new paying customers. Unless your platform is receiving closed-sale data back from your CRM, treat CPA as a cost per lead with a more flattering name.
What's a good cost per customer for a service business?
There's no universal number. A $1,500 CAC is cheap for a $30,000 kitchen renovation and ruinous for a $900 service. Judge it against first-year gross profit per customer and payback speed: as a working rule, CAC no more than a third of first-year gross profit, with payback inside a couple of quarters. The comparison that matters is between your own channels and months, not against an industry average.
Should I ask my agency to report cost per customer?
Yes, and their answer tells you a lot. A good agency will ask for access to your CRM or a monthly closed-deals list so they can build it. An agency that says it can't be measured, or that it's "sales' job," is telling you it's optimising for a number disconnected from your revenue. They may still be good at the ad account, but you'll have to own the CAC reporting yourself.
Why does my agency only report cost per lead?
Because it's the number the ad platform produces automatically and the number the agency controls most directly. Cost per customer depends on your follow-up and your sales team as well as their ads, so it's harder to build and harder to take sole credit for. That's a reasonable explanation and a poor excuse. The businesses that grow predictably insist on the second number anyway.
Sources
- WordStream by LocaliQ, Facebook Ads Benchmarks 2025: the average Facebook cost per lead across industries was $27.66 in 2025, which is the kind of benchmark the CPL argument tends to lean on, and exactly the number this post argues against judging by.
