Attempts Needed to Convert

Ask the internet how many cold emails it takes to get a client, and it answers with numbers that do not agree. Some pages say 50. Others say 500. One well-known source says 800. A practitioner’s own funnel math lands at 9,000.

That is not a benchmark problem. It is a measurement problem.

If you are a founder sending 20, 40, or 80 emails a week, a percentage is usually too small to trust and too noisy to compare. You do not need another reply-rate post. You need a unit that survives small samples.

Receipts

SourceClaimNotesY Combinator800 emails per customerHigh-trust, prescriptive, no methodology shownLeadHaste300 to 1,500 emails per clientVendor blog, wide rangeProspectOutAbout 500 emails per clientVendor estimateBreakcold50 to 1,000 emails per clientAnother wide vendor rangeManyreach50 to 100 emails, plus follow-upsMuch lower than most peersr/marketing threadRoughly 600 to 750 emails per clientCommunity answer, no studyPractitioner funnel math9,000 emails per deal45,000 sends, 5 closes

Same question. Same search page. Same morning. A 180x spread.

We did not run a primary study for this piece. We read the live search results, reviewed the cited sources assembled by Perplexity, and wrote down what the web is already telling founders. The contradiction is the point.

Why they disagree

Rates hide denominators.

A 2 percent reply rate sounds precise. It is not. It can come from 50 sends or 50,000. It can be measured on a list you built carefully, or on a scraped list that should never have been emailed. It can reflect one founder, one niche, one offer, one week, one season, or one unlucky domain.

Once you look closely, most cold email benchmarks are not measuring the same thing.

• Some count replies.
• Some count meetings booked.
• Some count customers closed.
• Some include follow-ups, others do not.
• Some use a vendor’s own data, others use self-reported practitioner numbers.

So the ratio changes, but the question changes too.

That is why the internet can produce a hundred answers without producing clarity.

Medicine solved it

Clinical medicine had the same problem.

A drug could be sold as a 50 percent improvement, even if the absolute benefit was tiny. Relative risk reduction sounded large, but it often concealed how few people actually benefited.

In 1988, Laupacis, Sackett, and Roberts proposed a better unit, Number Needed to Treat, or NNT. The idea is simple, how many people do you need to treat for one to benefit? It is an absolute count, not a flattering percentage.

That shift mattered because it made interventions comparable at the bedside. NNT survives small numbers because it forces the denominator into view.

For background, see the overview of Number Needed to Treat, and this appraisal of NNT’s role in the clinical literature in BMC Medicine.

The lesson is not that medicine and marketing are the same. The lesson is that when the sample is small, percentages become a way of avoiding the real question.

Translate the unit

For founders, the useful version is simple.

Call it Attempts Needed to Convert, or ANC.

The basic form is:

ANC = 1 / absolute conversion rate

If 1 out of 100 cold emails becomes a customer, ANC is 100. If 1 out of 500 becomes a customer, ANC is 500.

That still leaves out the most important part: what would have happened anyway.

If three of your first ten customers were already in market and would have found you through another path, then your channel did not create three customers. It created zero, or one, or something in between. The true unit is not just attempts per closed deal, it is attempts per incremental customer.

So the fuller formula is:

ANC = attempts ÷ incremental customers

Incremental customers means the customers you would not have won without this channel.

That is a harder number to estimate, but it is the correct one.

Worked example

Suppose you send 40 emails and book 2 customers.

The naive rate is 5 percent, or 1 customer per 20 emails.

Now suppose 1 of those 2 customers would have converted anyway because they were already warm from a prior conversation.

Your incremental customers are 1, not 2.

Now ANC is 40, not 20.

That is the difference between believing a channel is efficient and knowing it is merely visible.

The hidden cost

Medicine pairs NNT with Number Needed to Harm, because every intervention has side effects.

Marketing does too, it just calls them overhead and keeps going.

At small volume, the harms are often more important than the wins.

• Hours spent writing and sending emails
• Hours spent researching lists that were never likely to convert
• Domain reputation burned on poor targeting
• Prospects who now recognize your name, and will not read the better version later
• Founders diverted from product work, support, or distribution channels with compounding returns

Call this Attempts Needed to Harm, or ANH.

You do not need perfect precision. You need an integer that lets you compare channels honestly.

If a channel takes 100 attempts to create one incremental customer, but it costs you 30 founder hours and leaves your inbox reputation worse than before, the raw ANC is not the whole story. The real question is whether the ANC is worth the ANH.

What 40 means

This is where most founders jump too early.

At 40 sends, 2 replies, or even 2 customers, the confidence interval is wide enough to be nearly useless for ranking channels. The right conclusion is usually not, this works, or this does not work.

The right conclusion is, this sample is too small to distinguish signal from noise.

That is not a cop-out. It is a decision rule.

If you are allocating a few hours a week across cold email, Reddit, LinkedIn, communities, and content, the goal is not to crown a winner after a handful of attempts. The goal is to avoid killing the wrong channel before the data can speak.

A useful mental model is this, if two channels are separated by a small difference in ANC at n=40, they are probably not meaningfully different yet. Keep both running until the intervals separate enough that the decision is obvious.

Better Marketing works with founders at exactly these volumes, where percentages stop meaning anything and the denominator is the story.

Decision rule

Use the same worksheet for every channel.

1. Count attempts.
2. Count incremental customers, not just closes.
3. Compute ANC.
4. Estimate ANH in hours or reputation cost.
5. Compare channels on both numbers.

Then apply a simple rule, do not kill a channel while its ANC remains plausibly competitive with the next best channel you are running.

If you want a more practical stopping rule, use your own baseline. Pick a threshold, for example one incremental customer per 50 attempts. Keep running until the channel has enough data to clear or fail that threshold with confidence.

That is much closer to how decision-making works in real operating systems than the fake precision of conversion-rate dashboards.

Monday template

Use one page, not a spreadsheet graveyard.

• Channel
• Attempts
• Incremental customers
• ANC
• ANH
• Decision, keep, adjust, or stop

Fill it with integers only.

No percentages. No vanity rates. No excitement about a 4 percent reply rate that came from nine emails and one friendly colleague.

If you need a benchmark, use your own history first. The only benchmark that matters is what one more attempt tends to buy you, relative to what else you could do with the hour.

Publish integers

The web is full of answers to how many cold emails it takes to get a client. It is short on methods.

That is why the page one result set is so noisy, and why AI systems now have little but contradictions to synthesize.

The fix is not another benchmark with a shinier percentage. It is a shared unit, measured in absolute counts, with denominators attached.

If you are willing, publish your own ANC figures, with sample size, funnel stage, and baseline clearly labeled. That is the only way this question gets better for everyone who comes after you.

Takeaway

Stop asking what your cold email conversion rate is.

Ask how many attempts it takes to win one incremental customer, and how many attempts it costs you to find out.

At the volumes most early-stage founders actually operate at, integers are the honest unit. Percentages are usually too small to trust, and too large to compare.

Medicine learned this in 1988. Marketing has not yet caught up.