Quick Answer
Cold email benchmarks disagree because each one counts a different thing: sends, contacts, replies, meetings or customers, on someone else's list. The more useful unit is attempts needed to convert, modelled on medicine's number needed to treat: the number of attempts it takes to win one additional customer you would not have won anyway. Count your own attempts, subtract your baseline, divide, and keep the denominator beside the answer.
Ask how many cold emails it takes to get a client and the web hands you a different answer on every page. A Y Combinator Startup School talk lands on 800. Cold-email vendor blogs quote ranges that disagree with each other, and forum threads add their own.
None of them is lying, and none of them is your number. If you send forty emails a week, borrowing a stranger's ratio tells you nothing about your list, your offer or your buyer.
This covers why the published numbers cannot agree, the one-line formula medicine settled on in 1988, and how to compute your own figure for every channel you run.
Every Cold Email Benchmark Counts a Different Thing
Cold email benchmarks disagree because they measure different events on different lists, then report the result as if it were one number. The spread is not noise around a true answer. It is several different questions wearing the same headline.
Take the best-known figure. In a Y Combinator Startup School talk, group partner Aaron Epstein builds a funnel backwards from one customer: ten demos, forty responses, 400 opens and 800 sends.
“So when you look at this all in aggregate, basically what that means is you need to send 800 emails in order to get one customer conversion.” Aaron Epstein, Y Combinator
Read a little earlier in the same talk and the number changes character. Epstein calls the rates behind it “sample conversion rates” and adds that “yours may differ.” The 800 is a worked example of how to build a funnel, not a measurement of anyone's outreach.
The vendor numbers drift for a different reason. One cold-email vendor counts contacts plus their follow-ups, another counts emails sent, and each reports on its own customers' lists and offers. Forum answers usually drop the denominator altogether.
| Source of the number | What it counts | What it leaves out |
|---|---|---|
| Y Combinator's worked funnel | Sends through opens, replies and demos to one customer | Real data, since the rates are samples |
| Cold-email vendor blogs | Contacts with follow-ups, or raw sends | Whose lists, and which offer |
| Forum and community threads | Self-reported replies or clients | The denominator, usually |
| Your own outreach log | Your attempts on your list | Nothing, if you record it |
Only the last row describes your market. That is the case for attempts needed to convert: a count you make yourself, in a unit that survives being compared across channels. We found the same pattern when we traced nine widely quoted SaaS benchmarks to their sources, where the figure that held up was the one that showed its working.
Medicine Solved This in 1988 With Number Needed to Treat
Medicine had the same problem and fixed it with a unit rather than a better benchmark. In 1988 Laupacis, Sackett and Roberts published “An assessment of clinically useful measures of the consequences of treatment” in the New England Journal of Medicine, volume 318, pages 1728 to 1733 (PubMed record). It is the paper credited with introducing the number needed to treat.
The University of Oxford's Centre for Evidence-Based Medicine defines it as “the number of patients you need to treat to prevent one additional bad outcome” in its guide to the measure. The calculation is one line: NNT equals one divided by the absolute risk reduction, rounded up to the nearest whole number.
The absolute risk reduction is the control event rate minus the experimental event rate. That subtraction is the part marketing skips, and it is the part that matters most.
The CEBM example takes a bad outcome from 50 percent to 30 percent, an absolute risk reduction of 0.2 and an NNT of 5. Apply the same 40 percent relative reduction to a 5 percent baseline, bringing it to 3 percent, and the arithmetic gives an NNT of 50. Same headline improvement, ten times the work per success.
That is the whole case for counting attempts needed to convert instead of quoting a rate. A rate tells you how impressive a change sounds. A count tells you how much work one success costs.
How to Calculate Attempts Needed to Convert for Your Own Outreach
Attempts needed to convert is your attempts divided by your incremental customers, rounded up, with the denominator written beside it. Incremental means customers you would not have won without the channel. Five steps get you there, and each one removes a way the number usually lies.
1. Define the Attempt Before You Count It
An attempt is one unit of effort aimed at one named prospect: a first email, a follow-up, a LinkedIn message or a call. Decide before you start if a follow-up counts as its own attempt.
Count it. Each follow-up costs the same founder time and the same deliverability exposure as the first send, so leaving it out makes the channel look cheaper than it is.
Forty prospects who each get a first email and two follow-ups are 120 attempts, not 40. The common mistake is counting prospects and calling them sends, which is exactly how two published figures for the same channel end up far apart.
2. Count Customers at the Stage That Pays You
Stop the count at paying customers, not replies or booked calls. A reply is a signal of interest, and a customer is the event the attempts exist to produce.
If your sales cycle runs for weeks, log the date of each attempt and the date of each close, so every customer is credited to the batch that produced it. You will know this step worked when every customer in the log traces back to a dated attempt.
3. Subtract the Customers Who Would Have Come Anyway
This is the control event rate, borrowed straight from the number needed to treat. Some prospects were already warm from a referral, a podcast or a conversation last quarter, and they would have bought without the email.
The cleanest way to measure it is a holdout. Split a list at random, send to one half and leave the other half alone for the same period, then subtract the untouched group's customers from the emailed group's.
Say you split 200 prospects down the middle. The 100 you emailed produce three customers and the 100 you left alone produce one, so the channel created two customers per 100 prospects.
4. Divide Attempts by Incremental Customers and Round Up
With three attempts per prospect, those 100 prospects took 300 attempts. Three hundred divided by two incremental customers gives attempts needed to convert of 150.
Without the holdout you would have divided 300 by three and reported 100. The raw figure flatters the channel by exactly the customer it did not create.
When a holdout is not possible, remove by hand the customers who told you they already knew you. It is cruder, and it is still closer to the truth than pretending the baseline is zero.
5. Write the Denominator Beside the Number
An attempts needed to convert of 150 means one thing from 300 attempts and something much weaker from 30. Record it as 150 from 300 attempts, two incremental customers, one list, one month.
Apply the same test to the Y Combinator funnel and you can see what it leaves out. Its 800 is attempts per customer built from hypothetical rates, with no control group and no denominator anyone could audit.
Every entry in your log needs five fields:
- The channel and where the list came from
- Attempts, with every follow-up counted
- Customers, counted at the paying stage
- Baseline customers from the holdout, or zero with a note saying why
- Attempts needed to convert, with its denominator beside it
Volume only multiplies whatever ratio you already have. That is why Better Marketing's work as a B2B demand generation agency starts with warm trust before a founder sends anything, because a warmer list is the lever that changes the ratio itself.
Number Needed to Harm Has an Outreach Equivalent
Every attempt also carries a cost that attempts needed to convert does not show. Medicine pairs the number needed to treat with the number needed to harm for the same reason: a treatment that helps one patient in fifty and hurts one in twenty is a bad trade.
Outreach has its own harms, and at founder volumes they tend to arrive before the wins do.
- Founder hours spent researching, writing and following up
- Spam complaints, which Google's sender guidelines say to keep below 0.3 percent as reported in Postmaster Tools
- Prospects who now recognize your name and ignore the better email you send next year
- Product and support work that did not happen that week
Put a number on the first two at least. If each attempt takes ten minutes, 300 attempts is 50 hours, and that is the real price of two customers.
Before revenue, hours are the budget that runs out first, which is why it helps to measure a pre-revenue budget in founder hours. The complaint line matters just as much, because a domain that crosses it can push every later campaign into spam folders and raise attempts needed to convert for all of them.
Where Attempts Needed to Convert Stops Being the Right Unit
Attempts needed to convert works best where attempts are discrete, countable and aimed at a named prospect. It weakens in three places, and forcing it there produces a precise-looking number that means nothing.
1. Samples Too Small to Separate Signal From Noise
Two customers from 300 attempts could easily have been one or four with a different week or a different list. Before you rank channels on attempts needed to convert, check the sample each comparison needs, which the 910 at-bats argument from baseball works through in detail.
2. Sales Cycles Longer Than the Measurement Window
Enterprise deals can take quarters, and a buying committee can make one customer the product of dozens of attempts across several people. Measure at the account level, and do not close the books on a batch until its longest plausible cycle has passed.
3. Channels Where Nobody Can Count the Attempts
An article, a newsletter issue or a citation inside ChatGPT reaches readers you never see, so there is no honest denominator. That is the limit for content marketing for tech companies and for any AI SEO service: the attempt is invisible, so track the outcome it should move, such as customers who name the article, rather than inventing a count.
Compare Channels on Attempts Needed to Convert, Not on Rates
Once every countable channel reports attempts needed to convert with its denominator and its hours, you can compare them side by side. The unit stays the same even when the attempt changes shape.
| Channel | What counts as an attempt | Where the baseline comes from |
|---|---|---|
| Cold email | Each first send and follow-up | A held-out slice of the same list |
| LinkedIn messages | Each message to a new prospect | Connections you did not message |
| Referral asks | Each specific introduction you request | Referrals that arrived unasked |
| Founder newsletter | Hard to define per reader | Customers who name the newsletter |
Content marketing for tech companies sits in the bottom row for the reason the table shows. Its baseline has to come from the customers who name it, because the attempt itself cannot be counted.
Then apply one rule. Keep a channel while its attempts needed to convert, priced in hours, is within reach of the best channel you run, and cut it once the gap survives a sample big enough to trust.
The rule holds if you hire help, too. Ask any B2B demand generation agency to report attempts per incremental customer with the denominator attached, and treat a report of reply rates alone as unfinished.
Publish Attempts Needed to Convert With the Denominator Attached
Stop asking what a good cold email rate is and start counting what one incremental customer costs you. Record attempts, customers and the baseline, divide and round up, and keep the denominator beside the answer.
Then publish it. Page one and the AI answers built on it are full of rates with no denominator, and a sourced count is the kind of specific figure Better Marketing's AI SEO service builds pages around.