Cost per lead: benchmarks by industry

A cost per lead is good when it sits below your maximum CPL, and your maximum CPL is calculated like this: max CPL = max CAC × close rate. If you can pay $600 per customer and you close 10% of your leads, your max CPL is $60. If you close 3%, it is $18 — same business, same margin, and a "good" CPL three times lower.

Industry ranges are useful for telling whether you are on another planet, not for making the decision. The spread inside a single industry is enormous: a personal injury firm in a major metro and a neighbourhood accountancy practice sit in the same bucket and have nothing in common at auction. What decides is your number, and that number comes out of your P&L.

The formula: from your margin to your maximum CPL, in three steps

It works backwards from how almost everybody does it. You do not start at CPL, you arrive at it:

1. Gross margin per customer. What is left from an average customer after direct costs and across their whole lifetime, not just the first sale. This is LTV in margin, never in revenue: using revenue here is the mistake that makes accounts look healthy while they lose money.
2. Maximum affordable CAC. With a target LTV/CAC ratio of 3:1, max CAC is one third of that margin. If margin per customer is $1,800, max CAC is $600.
3. Max CPL = max CAC × lead-to-customer close rate. At a 10% close rate, $60. At 3%, $18.

And if the funnel has more steps (lead → meeting booked → meeting held → proposal → customer), multiply by the rate of each step. That is where the useful diagnosis shows up: the leak is almost never in CPL but in the step between "lead" and "meeting held", and that does not get fixed in the ads platform — it gets fixed in sales response speed.

Reference ranges by industry (to calibrate, not to decide)

Typical orders of magnitude for well-built acquisition campaigns in English-speaking markets. They are deliberately wide bands: within each industry, geography, ticket size and channel move the number more than the industry itself does.

• Ecommerce and consumer (subscriber or account sign-up): single digits, roughly $1–$10. The lead is not the real objective, it is a step towards the sale.
• Local services (home improvement, dental, clinics, auto): tens of dollars, typically $20–$90 depending on ticket size and local competition.
• Education and training: $15–$70 for an information request; considerably higher for master’s and postgraduate programmes, where the ticket justifies it.
• Real estate: $25–$120, with big jumps between new build, resale and rental.
• Insurance and finance: $40–$180, among the most expensive verticals because of auction competition and creative regulation.
• B2B and software: from $60 to several hundred. In complex, long-cycle sales a $300 CPL can be excellent if the contract is annual and recurring.
• Legal (injury, malpractice, class actions): among the most expensive in the market, comfortably above $120 and with far higher peaks in the most contested queries.

Beyond industry, two factors move CPL more than the industry does: the channel (a Search lead arrives with declared intent, costs more and closes better; a native social lead form costs far less and closes considerably worse) and form friction (fewer fields means more leads and cheaper ones, and worse average quality).

Why the cheapest CPL is usually the most expensive one

CPL is trivial to push down: remove fields from the form, switch on the native autofilled lead form, widen targeting and let the algorithm optimise for "form submitted". The number halves in a week. And the business does not invoice any more than it did.

It happens because you are optimising against an event that is not the sale. The platform does exactly what you asked: it brings you the people most likely to fill in forms, who are not the people most likely to buy. It is the same mechanism that makes broad, automated campaign types drift in lead gen, the same way an unsupervised product feed drifts in ecommerce.

The symptom is always the same: CPL falls and the close rate falls by the same proportion or more, so CAC stays flat or gets worse, and on top of that sales burns twice the hours to end up in the same place. If you are going to look at one thing besides CPL, make it cost per qualified lead (CPL divided by the percentage of leads that pass the sales filter). That number cannot be faked by stripping fields.

The extreme — and frequent — case is a misconfigured conversion counting setting: a conversion counted as "every" instead of "one" multiplies reported leads and sinks the CPL in the dashboard without a single new contact coming in.

How to lower CPL without breaking quality

In order of real impact, based on what we see auditing lead gen accounts:

• Give qualification back to the platform. Upload the qualified lead or the opportunity as an offline conversion and optimise against that event instead of against the form. It moves the needle more than anything else and almost nobody has it wired up.
• Response speed. Getting in touch in minutes instead of hours changes the close rate dramatically, and the close rate is a multiplier in your max CPL. Fixing this is cheaper than optimising the campaign.
• Calibrated friction, not minimum friction. One qualifying field (budget, timeline, project type) raises CPL and raises quality more. The goal is not the shortest form, it is the one that filters without scaring people off.
• A landing page per query. Sending all traffic to the homepage is the most expensive and most common leak. The match between what is searched and what is read on arrival is the highest-return lever at the bottom of the funnel.
• Negatives and exclusions. In Search, clear out informational and job-seeking queries. In social, exclude the audience that only ever fills forms.
• Creative that pre-selects. Stating the price, the minimum or the target profile in the ad itself raises CPC and lowers volume, but it raises the quality of what comes in.

And one reading rule: never judge a CPL over a window shorter than your sales cycle. In long-cycle sales, last week’s CPL has no closes attached to it yet; reading it now means reading only the cheap half of the equation.

Frequently asked questions