Cost per lead is the most expensive number in marketing

Cost per lead is the most widely used metric in performance marketing, and it is a trap. Not because it is inaccurate — it is usually computed correctly — but because it is easy to improve in ways that make the business worse, and because the people improving it are rewarded before anyone finds out.

Cost per lead can be halved by a competent marketer in a fortnight, without a single additional customer being acquired. That is not a criticism of the marketer. It is a criticism of the metric.

This is an unpopular position, mostly because cost per lead is the number in everybody's dashboard, everybody's contract, and quite a lot of people's bonus. So let us be specific about how it fails, how much it costs, and what to do about it without a two-year data programme.

The mechanism of the failure

Cost per lead measures the price of an enquiry. Nothing in it refers to whether the enquiry becomes a customer, so any change that produces more enquiries at lower quality improves the metric while reducing revenue.

The important word is any. This is not a matter of bad actors gaming a metric. Every ordinary optimisation lever — broader targeting, a shorter form, a more generic offer, cheaper placements, a competition entry — pushes in exactly this direction. The metric rewards them, so they get done.

The feedback that would correct this arrives months later, in a different system, filtered through a sales team's opinion about lead quality. By then the campaign structure has changed twice and nobody can reconstruct which decision caused what.

So the failure is not that someone chose the wrong number. It is that the loop is open. The metric is fast and wrong; the correction is slow and vague; the fast one wins every time.

What it looks like in real numbers

The clearest way to see the problem is to rank channels by cost per lead and then by cost per acquired revenue. The two orderings frequently disagree, and when they do, the second one is the one connected to your bank account.

Ranked by cost per lead, the order is display, paid social, paid search, trade press. Ranked by return, it is exactly reversed. A team optimising the first column would move budget out of the two channels producing three quarters of the revenue, and would be able to show a beautiful trend line while doing it.

This is not a contrived example. The pattern — cheap leads from broad channels, expensive leads from narrow ones, revenue concentrated in the expensive ones — is the normal shape of a considered-purchase business.

Why it survives despite being widely known

Cost per lead persists because it is available immediately, comparable across channels, computable without touching the sales system, and defensible in a meeting. Its replacement has none of those properties without deliberate work.

That is a real advantage and it should be acknowledged rather than dismissed. A metric you can see today beats a better metric you can see next quarter, especially when budget decisions are weekly. Anyone arguing to abandon cost per lead has to answer what people are supposed to look at on Tuesday.

There is also an organisational reason. Cost per lead sits entirely inside marketing's control. Cost per acquired revenue requires the sales system, which belongs to somebody else, whose data quality is not marketing's to fix and whose priorities are different.

Metrics that require another department's cooperation are structurally disadvantaged. This is worth naming, because it explains why the fix is usually political before it is technical.

The honest defence of cost per lead

Within a single channel, over a short period, with the offer and targeting unchanged, cost per lead is a reasonable early-warning signal. It tells you something moved. It just cannot tell you whether the movement was good.

Used that way it is genuinely useful and nobody should throw it away. A twenty percent jump in cost per lead on a stable campaign is worth investigating on the day it happens, and waiting ninety days for revenue data to confirm it would be negligent.

The line to hold is between monitoring and allocation. Monitor with fast metrics. Allocate with slow ones. The mistake almost everybody makes is allocating with the monitoring metric because it is the one already on the screen.

Fast metrics are for noticing. Slow metrics are for deciding. Using a fast metric to decide is how a marketing team optimises itself into a smaller business.

What to measure instead

Cost per acquired revenue is spend on a channel divided by the closed revenue traceable to it, over a window at least as long as the sales cycle. It is the same arithmetic as cost per lead with a different denominator, and the denominator is the entire point.

Computing it requires one thing that cost per lead does not: a join between the sales system's record of what closed and the source records of where those customers came from. That join is the work. Everything else is division.

1. Export closed sales for a period, by close date, with amounts.
2. Export lead and call records reaching back further than your longest cycle.
3. Match them on phone and email, normalised on both sides.
4. Attribute spend to channels over the corresponding period.
5. Divide, and report the unmatched revenue as its own line rather than distributing it.

Note what is not in that list: a tag manager project, a data warehouse, or a year of implementation. Two exports and a join will get most businesses a defensible answer within a week, and the answer is usually startling enough to change a budget immediately.

The lag problem, and how to live with it

Revenue-based metrics arrive late by construction, because the revenue arrives late. The workable compromise is a leading indicator chosen because it predicts revenue, monitored weekly, with the revenue measure reviewed monthly.

The leading indicator should be validated rather than assumed. If you decide qualified opportunities predict revenue, check that the channels producing the most qualified opportunities really do produce the most revenue. Sometimes they do not, and discovering that is worth more than the indicator.

What makes this work is holding both in view at once. A team that watches only the leading indicator has rebuilt cost per lead with extra steps; a team that watches only revenue cannot act quickly enough to matter.

Lead quality is not a sales team opinion

Every business has a running argument about lead quality conducted entirely through anecdote. Matched revenue by source ends it, because it replaces impressions of quality with the closed value each source actually produced.

The argument follows a familiar script. Sales says the leads are poor. Marketing says sales is not following up. Both cite the same three memorable examples. This can run for years and consumes an extraordinary amount of senior attention.

What resolves it is not a better argument but a shared table: source, leads, closed deals, closed revenue, average value. It is frequently uncomfortable for both sides, which is a good sign that it is measuring something real.

How to change the metric without losing the room

Introduce the revenue measure alongside cost per lead rather than in place of it, for at least two quarters, and let the divergence make the argument. Removing a metric people are judged on before the replacement is trusted produces resistance rather than adoption.

The reason to run both is evidential. When the two rankings disagree and the revenue ranking turns out to be right, that is a demonstration rather than an assertion, and it converts sceptics far more efficiently than any presentation about metric design.

It also protects you against the possibility that your new measurement is wrong. Attribution has its own failure modes, and running it in parallel with the incumbent for a while is how you find them before betting a budget on them.

CloseRev is built to make that parallel period cheap: two exports, a match, and a channel table showing closed revenue with the unmatched portion visible. No tagging project, and nothing to rip out if you decide the answer is not useful.

The agency version of this problem

Agencies are frequently contracted on cost per lead, which means their commercial incentive is aligned to the metric rather than to the client's revenue. This is a structural problem, not a question of agency integrity.

Put plainly: if you pay an agency to reduce cost per lead, a good agency will reduce cost per lead. If that makes your revenue fall, you asked for the wrong thing and got it delivered competently. Blaming the supplier for hitting the target you set is not a strong position.

The better arrangement is to contract on revenue-linked outcomes, share the sales data required to compute them, and accept that this means the agency can see numbers many clients instinctively withhold. That withholding is usually what forces the relationship back onto lead volume in the first place.

The three cheap tricks that improve it

Cost per lead can be lowered on demand by broadening targeting, shortening the form, or redefining what counts as a lead. All three are reversible in an afternoon, all three are standard practice, and all three usually reduce revenue.

Broadening targeting is the most defensible of the three, because it genuinely does find people the narrow targeting missed. It also finds a much larger number of people who were never in the market, and since the metric counts enquiries rather than customers, the dilution registers as success. The revenue effect appears one sales cycle later, by which time the change looks like ancient history.

Shortening the form is the most seductive, because every conversion rate optimisation guide recommends it and the immediate effect is real and large. What the guides tend not to say is that the fields you removed were doing qualification work. A form asking for company size and budget produces fewer, better enquiries; removing those fields produces more, worse ones, and the metric cannot tell the difference because it does not look at what happened next.

Redefining what counts as a lead is the one that should worry you, and it is remarkably common in the weeks before a quarterly review. Counting a newsletter signup, a chat session, or a whitepaper download as a lead is not fraudulent — someone can always construct a rationale — and it can transform a metric overnight without anything happening in the real world at all.

What good looks like in practice

A team measuring this well can state, for each channel, the spend, the closed revenue traced to it, the share of revenue that could not be traced, and the sales cycle length that determines their measurement window. Four facts per channel, reviewed monthly.

That is a deliberately modest standard, and most businesses do not meet it. It requires no attribution model, no data science, and no platform beyond the ability to join two exports. What it does require is that somebody owns the join and runs it on a schedule, which is an organisational commitment rather than a technical one, and organisational commitments are harder to obtain than software.

The reason to keep the standard modest is that ambitious measurement programmes fail at a much higher rate than simple ones. A monthly channel table that everybody trusts will change more decisions in a year than a sophisticated model that arrives eighteen months late and is immediately disputed by whoever it makes look bad.

If you want a single test of whether your measurement is working, it is this: when the numbers arrive, does anybody change what they are doing? Measurement that produces agreement and no action is a reporting habit rather than a management tool, and reporting habits are expensive.

The uncomfortable summary

Most marketing teams are measured on a number that can be improved while the business shrinks, by people acting entirely in good faith, using techniques taught as best practice. The fix is not sophistication; it is connecting two files that already exist.

None of this requires new infrastructure or a new discipline. The sales system knows what closed. The marketing systems know where enquiries came from. Nothing joins them, so the fast metric wins by default, and the default has been running for years.

You do not have a measurement problem. You have two files that have never been in the same room.

Questions people actually ask

What is wrong with cost per lead as a metric?

It measures the price of an enquiry, not the value of a customer. Because lead quality varies enormously by channel, a campaign can cut its cost per lead in half while producing less revenue, and the metric will report that as an improvement.

What should I measure instead of cost per lead?

Cost per acquired revenue: what you spent on a channel divided by the closed revenue that channel produced, over a window long enough to cover your sales cycle. It requires joining spend to closed sales, which is more work than reading a dashboard and is the entire difference.

Is cost per lead ever useful?

Yes, as a diagnostic within a single channel over a short period, where lead quality is roughly constant. It is a reasonable way to notice that something changed. It is a poor way to decide where budget should go.

How much does lead quality actually vary between channels?

Enough to invert a ranking. It is entirely normal for one channel's leads to close at two or three times the rate of another's, and at higher average values, which means a channel with double the cost per lead can be the cheaper source of revenue by a wide margin.

How long a window do I need to measure cost per acquired revenue?

At least as long as your sales cycle, and preferably twice it. Measuring a ninety-day cycle over a thirty-day window will make every channel look terrible and the slowest-closing ones look worst, which is precisely backwards.

What if my sales team will not report closed-won properly?

Then that is the project, and no measurement change will substitute for it. Attribution consumes the sales system's record of what closed; if that record is incomplete, every downstream number inherits the gap, and the fix is organisational rather than technical.

Found this useful?

• Share on X
• Share on LinkedIn
• Share on Reddit
• Share on Email