Two vendors published cold email reply rates for the same calendar year, using the same stated denominator. One said 0.45 per cent. The other said 3.43 per cent.
Neither is lying. Both are internally consistent. They are measuring different self-selected populations of their own customers, and because no standards body defines a reply rate for prospecting, both can be right about their own data and useless to you.
That is the whole problem with channel benchmarks in one example, and this piece is about how deep it goes. We set out to test a claim rather than assert it: that no neutral United States source publishes comparable performance data across prospecting channels. The claim survives, with three qualifications that turn out to be more interesting than the claim.
What the government actually measures
The federal statistical system is good at some things and completely silent on others, and the boundary is sharp.
We read the annual business survey questionnaire rather than a summary of it. The closest it comes to a channel question is asking which types of customer accounted for ten per cent or more of sales: federal government, state and local government, other businesses, other organisations, individuals. It asks who you sell to. It never asks how you found them. There is no question on marketing spend, advertising, prospecting or customer acquisition anywhere in the instrument.
The economic census does collect a purchased advertising and promotional services expense line, which is a genuine positive. It is one of eleven detailed operating expense items, collected only for construction, manufacturing and mining, as a single aggregate dollar figure. It tells you what an industry spent, never what it bought.
We grepped the small business advocacy office’s flagship statistical publication for “marketing”, “advertising”, “referral”, “word of mouth” and “customer acquisition”. Zero hits across all five. The federal office whose statutory job is researching small business concerns publishes nothing on how small businesses get customers.
The central bank’s small business survey has an entire chapter titled “Customers”. It contains two questions: which customer types account for ten per cent or more of sales, and how far customers are from headquarters.
Institutions measure the cost of a sales function with real precision. None of them measures what any channel returns.
Note what that leaves you with. The only figure in this entire field that comes from a probability-based federal survey is the cost of a salesperson: a median wage of $76,460 across roughly 1.57 million jobs, with projected employment change of zero per cent over the decade. You can cost the input precisely. You cannot source the output from anyone disinterested.
The honest counter-example, and why it matters
Before taking the vendors apart, it is worth naming what good disclosure looks like, because one institution in this area does it and the contrast is the real argument.
The central bank’s small business survey states in print that it “is not a random sample; results should be analyzed with awareness of potential biases that are associated with convenience samples”. It names the commercial databases it drew its sample from. It publishes its weighting method and names the academic organisation that developed it.
Even the direct mail trade association, whose benchmark we are about to criticise, prints twice in its own report that the data “should not be considered benchmark data”.
Institutions and honest trade bodies tell you their limits. The software vendors do not. That is the distinction worth carrying, and it is sharper than a blanket accusation of bias.
Tracing the numbers
For each channel, the most quoted figure and what the organisation publishing it sells. The measurement and the commercial interest are the same party throughout.
The social selling case is the purest, so it is worth stating fully. The platform defines the index, computes each user’s score, publishes the study validating the index, and sells the subscription that raises the score. The primary page for that study now returns a not-found error, and across the pages citing it, not one gives a sample size, a date, or a definition of “opportunity”.
Four figures to refuse outright, because we tried to trace them and could not:
The claim that the great majority of small businesses get customers by word of mouth comes from a vendor-commissioned survey of small business owners in one city, conducted in 2014.
The trust-in-recommendations figure everyone quotes is from fieldwork conducted in 2011, in a 56-country global survey of consumers with home internet access, about advertising formats. It is not US, not B2B, and measures a stated attitude rather than any behaviour. The same publisher’s later edition reports a materially lower number, and the two circulate interchangeably.
The most cited cold email response rate outside the vendor reports comes from a study of link building, guest posting and PR outreach to bloggers and editors, a population that expects to be pitched. It states no time period. It is routinely quoted as a sales benchmark.
And the trade show cost-per-lead comparison against a field sales call has the shape of a legacy advertising research number from decades ago, recirculated as current.
The channel everyone names first, and nobody measures
Referrals are the channel business owners cite before any other, and they are the one with no institutional data at all. We checked the small business advocacy office, the annual business survey, the central bank survey and the business trends survey. The word “referral” appears once across all of them, and it is about how a borrower chose a lender, not how a company found a customer.
That absence is itself worth stating. The most-cited channel in the country is measured by no federal statistical programme.
What fills the gap is worse than nothing, because it looks like evidence.
A single-city vendor survey from 2014, and consumer fieldwork from 2011 about advertising formats. Neither is B2B, and neither is current.
There is one piece of serious work here, and it is worth knowing precisely because it is so much narrower than the claims it gets used to support. A peer-reviewed study tracked roughly ten thousand customers of a large bank for nearly three years and found that referred customers were worth at least 16 per cent more than non-referred customers matched on demographics and acquisition timing, with higher margin and higher retention.
Note the caveats, because they are the point: it is German, it is retail banking, it is B2C, it is a single firm, and it measures the value of a formal referral programme rather than the share of revenue produced by organic word of mouth. It is also, by a distance, the best-evidenced statement anyone can make about referrals.
The channel nobody sells software for, and its free outcome data
There is one exception to everything above, and it is the channel with the least marketing around it.
Federal award data is published openly, and we queried it rather than reading about it. Per award, it returns the recipient, the amount, the awarding agency and sub-agency, start and end dates, the classification code for what was bought and a plain description. Ask for one award in detail and it also returns the solicitation identifier, the number of offers received, whether the work was competed at all and on what authority, the pricing type, the parent company, and named executives with compensation.
For one channel, the outcome data is public, free, granular and audited. It is the one nobody builds prospecting software for.
We ran a filtered query on small business set-aside awards within a single marketing services classification code and got back named companies with contract values in the millions, by agency. A firm in that sector can list, by name and dollar amount, every competitor that won a set-aside in its own code. There is no equivalent for any private channel, at any price.
Two further things make this channel structurally different rather than merely transparent.
Payment terms are set in regulation. Payment is due within a defined period after a proper invoice or after acceptance, interest runs automatically from the day after the due date without any claim being filed, and the rate is set by the treasury and published every six months. A further penalty applies if the interest itself is paid late.
Small business access is a statutory target, and it is exceeded. The government-wide goal is 23 per cent of prime contracting dollars. The most recent reported achievement is nearly 28 per cent, representing $179 billion in prime contracts.
None of which makes public bidding easy. It demands administrative competence and a long horizon, and it is the least improvisable channel in this article. But it is the only one where the targeting data is audited, the outcomes are public and the buyer’s payment behaviour is regulated.
The finding that reframes the whole argument
Here is where the bias framing stops being the point.
A peer-reviewed economics study examined twenty-five large randomised field experiments run with major retailers and brokerages, most reaching millions of customers, together representing millions of dollars of advertising spend. Its conclusions are uncomfortable for everyone in this discussion.
The median confidence interval on return on investment was over 100 percentage points wide. Individual sales are so volatile that, in the authors’ framing, an informative advertising experiment “can easily require more than 10 million person-weeks”, making such experiments costly and potentially infeasible for many firms. And because advertising is targeted, selection bias is described as a crippling concern for the observational methods everyone actually uses.
The bias framing suggests a better source would fix it. This finding says the number is not cheaply producible, by a vendor or by you.
Read that twice, because it changes what you should do. The bias framing invites you to think a better source would fix the problem. This says a better source is not economically producible, and that the same constraint applies to you measuring your own channels. The vendors are not withholding a number they have. Nobody has it.
There is genuine cross-channel causal work, and it looks nothing like a benchmark table: a peer-reviewed study of two retailers, using hierarchical models plus a randomised field experiment for causal support, concluding that direct mail drove acquisition while email drove sales to existing customers, with a modest and conditional reallocation gain. Two firms, two channels, a named funder, a published method, a careful result. Set that beside a 15.6 per cent response rate from twenty-six survey respondents.
What to do with this
Stop shopping for a benchmark. Whatever number you find will come from a company selling the channel, and the study above says the number they would need to produce honestly costs more than it is worth to them. Ask instead who published it and what they sell, and if you cannot answer in one sentence, do not put it in a plan.
Manage on what you can observe. Not modelled return, but counts a human recorded: contacts worked, conversations held, meetings booked, opportunities opened. They are crude and they are yours.
Cost the input precisely, because that part is knowable. A salesperson’s fully loaded cost is the one figure in this field with a federal probability survey behind it. Put your own version of it against the meetings actually produced, and you have a ratio that means something in your business even though it means nothing in anyone else’s.
Ask one question at the start of every discovery call: how did you come across us. Recorded by the person in the room rather than reconstructed from a model afterwards, it is the only attribution that survives a long sales cycle.
And if you sell anything a public agency buys, look at the award data before you buy a list. It is free, it names your competitors and their prices, and it tells you when the incumbent’s contract runs out. No commercial channel will give you that, because no commercial channel has it.
Does any US institution publish comparable performance data across prospecting channels?
No. We checked the Census Bureau, the Bureau of Labor Statistics, the Small Business Administration's advocacy office and the Federal Reserve's small business survey. Several ask which types of customer account for a share of sales. None asks how those customers were found. The advocacy office's flagship statistical publication contains no occurrence of marketing, advertising, referral, word of mouth or customer acquisition. What federal programmes do measure well is the cost of a sales function, not its channel performance.
Why do published cold email reply rates disagree so much?
Because they are platform populations, not samples. Two vendors published reply rates for the same calendar year using the same stated denominator, replies divided by total emails sent: one reported 0.45 per cent across 7.5 million emails, the other 3.43 per cent across an undisclosed volume described as billions of interactions. Neither is reconcilable with the other and neither publishes an auditable sampling frame. Both are internally consistent, because no standards body defines a reply rate for prospecting.
Is the direct mail response rate benchmark reliable?
It is produced by the trade association of the channel, gated behind a member login, and drawn from a self-selected survey rather than a probability sample. In an earlier edition that can be read in full, the headline response rate rests on 26 respondents, and a question added to that edition found that the share reporting actual metrics rather than estimates averaged about 21 per cent. The association writes twice in the same document that the data should not be considered benchmark data.
Which prospecting channel has genuinely neutral outcome data?
Public sector bidding, and it is free. Federal award data is published openly and shows, per award, the recipient, the amount, the awarding agency, the start and end dates, the classification of what was bought, a description, the solicitation identifier, the number of offers received and whether the work was competed at all. A small firm can list every competitor that won a set-aside contract in its own classification code, by name and dollar value. No commercial channel offers anything comparable.
Can a small business measure its own channel ROI instead?
Not reliably, and that is the deeper finding. A peer-reviewed study of 25 large randomised advertising experiments found the median confidence interval on return on investment was over 100 percentage points wide, and that because individual sales are highly volatile, an informative experiment can require more than ten million person-weeks of exposure. The problem is not that vendors are biased. It is that the number is expensive to produce for anyone, which is why observational estimates fill the gap.