The evidence

Does direct mail work for B2B?
Nobody has actually run the experiment.

We sell direct mail, so the commercially sensible thing to publish here would be a page of response-rate statistics. We went looking for the evidence behind those statistics first, and what we found was that for business-to-business direct mail it does not exist. Not thin evidence. Effectively none.

That is an awkward thing for a mail company to say and it is the most useful thing on this page, because it is the whole reason to measure your own program against a holdout rather than trusting anybody’s benchmark, including ours.

What the literature holds

Five out of sixty-one,
and none of them mail.

The most useful thing we found is a census rather than a study. Duncan Simester of MIT Sloan had a research assistant read every issue of the five leading quantitative marketing journals from 1995 to 2014 — the Journal of Marketing, JMR, Marketing Science, QME and the marketing section of Management Science — and catalogue every field experiment in them.

Sixty-one papers, reporting eighty-six field experiments. His finding, quoted exactly: “Only five of the papers report field experiments in which firms target other businesses.” None of those five tested direct mail or advertising to business prospects. He also notes that almost thirty percent of published marketing field experiments do not randomise at all.

The census stops in 2014. We checked what has appeared since. The B2B field experiments published in Marketing Science and Management Science after that date are about pricing algorithms and procurement negotiation. Still nothing on mail.

Field experiments in marketing, 1995–2014

5 of 61

papers involving firms targeting other businesses.

journals read, every issue5

field experiments catalogued86

B2B papers5

testing direct mail0

Simester, D. (2015), “Field Experiments in Marketing”, MIT Sloan. A book chapter rather than a peer-reviewed paper, and a systematic census with a stated method.

What gets cited instead

Consumer data, wearing a suit.

The statistics that circulate in this category are not fabricated. They are real numbers about a different question, repeated until the qualifier falls off.

Household mail figuresThe read and open rates everyone quotes, ours included, come from consumer household panels. A person sorting post at their kitchen table is not a procurement lead opening the mail at a distributed company. The number may still be directionally useful; it is not B2B evidence.consumer

The two direct mail field experimentsSearch the literature for randomised direct mail trials and two good papers surface: a European insurer and an international beauty retailer. Both are rigorous. Both mailed consumers.consumer

The 13.5 percent liftWidely quoted from the beauty-retailer study. It is a model-implied projection of what reallocating budget could produce, not the experimental result. The measured experimental increase was 6.5 percent, and the recipients were shoppers.misread

The B2B brand researchThe most respected work in B2B marketing science is survey-based. We downloaded the flagship report and searched all sixty-four pages for “randomised”, “control group” and “experiment”. Zero matches. Its method is 609 and 616 online panel interviews. Excellent descriptive work, and not causal evidence.observational

The industry response-rate reportThe most-cited benchmark in the category. The edition we read reports 15.6 percent for house lists and 10.8 percent for prospect lists — from twenty-six and twenty-five self-reported campaigns respectively, with no control group. The publisher’s own methodology section states the data “should not be considered benchmark data”, and that on average only about a fifth of respondents used actual metrics rather than estimates. It also states it had too few responses to report a B2B figure at all.self-reported

The 4.4 percent everyone quotesAttributed across dozens of vendor pages to that same report. It does not appear in the edition we read, and the pages citing it mix figures from different years. We could not trace it to any primary document.untraceable

“4.4 percent versus 0.12 percent for email”The pairing that launched a thousand decks. The 0.12 percent traces back to a 2012 industry report whose actual conclusion was the opposite of how it is now used: that email returned roughly four times direct mail’s ROI. The comparison survives; the finding it came from did not.inverted

The ABM benchmark reportsIf mail were a proven account-based play you would expect the annual ABM benchmark studies to measure it. Across seven consecutive waves it is never measured, appearing once as a planned budget line. The major analyst blog category on ABM has nothing on it either.absent

Vendor benchmarksResponse-rate and ROI figures published by platforms, including the ones on our own competitors’ sites, are almost always drawn from customer programs with no control group. They measure what happened, not what the mail caused.no control

The genuine evidence

Randomised mail to organisations
does exist. It is mostly government.

There is real, well-identified, large-sample randomised evidence about mailing organisations. It comes from tax authorities and government agencies, the effects are modest, and several of the best-designed trials found nothing at all. This is the honest ceiling on what a letter to a business does.

IRS employer-firm letters (Boning et al., Journal of Public Economics, 2020)12,172 US firms. A letter raised the probability a firm remitted any tax by 3 percentage points, with no measurable effect on the amount remitted, and the effect did not persist beyond one quarter. An in-person visit moved the same measure 12.9 points.randomised

Chilean VAT audit-threat letters (Pomeranz, American Economic Review, 2015)445,734 firms, 100,000+ mailed. Median declared VAT rose about 7.6 percent against baseline, then decayed back to control levels over roughly fifteen months. A placebo arm established that the content did the work, not the arrival of mail.randomised

Capabilities statements emailed to federal buyers (US Office of Evaluation Sciences, 2026)32,759 federal buyers. Professionally rewritten B2B marketing collateral produced no detectable effect. Click rate was 0.65 percent for the treatment against 0.80 percent for the control, a negative point estimate at p = 0.115.randomised, null

SBA contracting-assistance emails (US Office of Evaluation Sciences, 2024)15,811 applicants. Null on every outcome measured, including submission and completion.randomised, null

USDA microloan letters to farmers (US Office of Evaluation Sciences, 2015)randomised by zip code. A positive result, and worth stating at full precision: the successful applicant rate moved from about 0.09 percent to about 0.11 percent. Often described as a doubling of activity, which is true of the ratio and misleading about the base.randomised

Read those together. Mail from an authority with the power to audit you moves a binary behaviour by about three points for a single quarter and moves the amounts by nothing. Better-written B2B collateral, tested on thirty-two thousand professional buyers, did nothing measurable. That is the base rate a commercial letter is arguing against.

The one strong positive we found is adjacent and instructive: letters to five thousand high-prescribing physicians, telling them their prescribing was elevated and under review, cut the target prescription by 11 percent at nine months and 15.6 percent at two years. Mail changed expert professional behaviour, durably, when it carried social comparison and the implication of scrutiny. Not a sales letter, and worth knowing what it took.

When it has been measured properly

The randomised results are humbling,
and worth knowing anyway.

Consumer marketing does have randomised mail experiments, run by academics with real control groups. We are quoting them against our own interest, because a page that only surfaced the flattering findings would be doing the exact thing it is complaining about.

Advertising ROI is hard to measure at allA study of twenty-five large randomised advertising field experiments found that the median confidence interval on return exceeded a hundred percentage points. Even with true randomisation and enormous samples, the honest answer is frequently “we cannot tell.” (Lewis and Rao, Quarterly Journal of Economics, 2015, predominantly online display rather than mail.)strong

Catalogue mail can backfire on your best customersA randomised field experiment on retail catalogues found current advertising affects future sales, but not always positively: for the firm’s best customers the long-run effect may be negative, through brand switching and simply pulling purchases forward. (Simester and colleagues, Economic Inquiry, 2009.)strong

Mail outlasts email, when testedRandomised experiments across catalogue and email found catalogues had substantially longer-lasting impact on purchasing than email did. The nearest thing we found to a modern, properly controlled direct mail incrementality study — and its subjects were retail consumers. (Zantedeschi and colleagues, Management Science, 2017.)strong

The postal service’s own auditors flagged the gapAnalysis commissioned by the USPS Office of Inspector General notes that its survey’s 12.3 percent figure measures households considering a response, and states plainly that this is “much higher than actual response rates to direct mail, which are typically 3 percent or less.”moderate

The one rigorous B2B study points the other wayA peer-reviewed analysis of a B2B firm’s marketing mix found a large profit gain from shifting budget away from firm-initiated offline contact. One study is not a verdict, and it is the closest thing to rigorous B2B evidence we found on the question — and it does not favour us. (Wiesel, Pauwels and Arts, 2011.)counter-evidence

The multichannel case has a footnote nobody quotesThe best-known effectiveness study crediting mail with about a ten percent uplift ranks it fifth of ten media, draws on self-submitted award entries, and spans 1980 to 2016. The same report finds that over-weighting budget toward mail reduces effectiveness. The largest recent econometric study of media returns excluded direct mail outright for insufficient effectiveness data.thin

Nobody has measured mail under hybrid workingEvery study above predates or ignores the question of whether a piece addressed to an office reaches a person who no longer sits there. For B2B specifically this is not a footnote, it is the central deliverability risk, and there is no research on it at all.unmeasured

Everyone else runs these trialsPolitical scientists randomise get-out-the-vote mail routinely. Development economists randomise letters to firms. Tax authorities randomise at the scale of hundreds of thousands of businesses. Commercial B2B marketing, which has more at stake per recipient than any of them, does not.the asymmetry

Reading it correctly

Absence of evidence
is not evidence of absence.

What this does not mean

It does not mean B2B direct mail fails. Nobody has run the trial that would tell you either way. The commercial world does not publish its experiments, and academic marketing has spent thirty years studying consumers because that is where the data is.

Plenty of teams have built real pipeline with mail. We work with them. What none of us has is a published, randomised, independently-verified number.

What it does mean

Every benchmark you are shown for this channel — including any we might put in front of you — is either consumer data, an uncontrolled customer program, or a projection. None of them tell you what mail would do for your accounts, your offer, and your follow-up.

So the only number worth having is the one your own program produces against your own holdout. Not because measurement is a nice feature, but because in this category there is genuinely nothing else to go on.

That is the entire argument for how we build. Run the channel as a designed experiment, hold accounts back, report the lift with its interval, and say plainly when the sample cannot support a conclusion. The holdout testing guide covers the design, and measurement shows the report it produces.

How we searched

And where our own search was weak.

A page claiming an absence of evidence should show how hard it looked, and admit what it could not reach.

What we searchedThe OpenAlex, Crossref and Semantic Scholar catalogues by title and abstract, for randomised and field-experimental work on B2B advertising, direct mail, holdouts and incrementality; plus the primary texts of the studies cited above, read rather than summarised.

What came back emptyA catalogue search for randomised holdout experiments in business advertising returned zero results. Searches for B2B field experiments returned pricing, delivery, auctions and churn work, and nothing on marketing communications. The B2B direct mail literature that does exist is descriptive and dates from 1991 to 2000.

What we could not accessGoogle Scholar, SSRN full text, ScienceDirect and Springer were unreachable, and one catalogue rate-limited us. A paywalled randomised B2B mail study could exist and we would not have found it. If you know of one, we would genuinely like to read it.

What we deliberately did not claimWe found references to further randomised letter trials on firms whose numbers we could not verify at source, and we have left them out rather than cite a figure we have not read in the original.

Start free Schedule a call

Does direct mail work for B2B?Nobody has actually run the experiment.

Five out of sixty-one,and none of them mail.

Consumer data, wearing a suit.

Randomised mail to organisationsdoes exist. It is mostly government.

The randomised results are humbling,and worth knowing anyway.

Absence of evidenceis not evidence of absence.