Twenty-two companies, including the agencies behind sampling programs for three of the largest drinks companies in the world, and not one bar sampling result between them

In October 2026 we went looking for published evidence that on-premise alcohol sampling works. Not marketing claims about experiential in general, but a specific program, at specific bars, with a sales number attached and some account of how it was measured.

We read the public case studies, insights sections and capability pages of nine experiential agencies, six digital on-premise platforms, and seven beverage-alcohol research firms. Twenty-two companies in all.

Four case studies in the entire set reported any sales figure at all for an alcohol brand activation, and none of them meets the standard set out below. Two were in stores and two were at festivals. A fifth reported a sales lift in bars, from a program that wasn't sampling. None reported what a sampling program did to sales in a bar.

Here are the twenty-two, so anyone can repeat this or contradict it. Agencies: Attack! Marketing, GMR Marketing, Team Enterprises, Inspira Marketing, Enthuse Marketing, BeCore, Legacy Marketing Partners, Advantage Solutions, Mosaic. Digital on-premise platforms: Bars.com, Union, PayBev, LineLeap, Shared Spirits Marketing, Overproof. Research firms: NielsenIQ (CGA), Circana, IWSR, Technomic, Mintel, Datassential, Numerator. We read each company's public case study library, insights section and capability pages as they stood in autumn 2026, and counted a result only if it named a licensed on-premise venue type, reported a sales figure, and said something about where the number came from. Impressions, samples poured, store visits and collateral placements didn't count. Private results, gated material and anything behind a client login were out of reach and aren't part of the claim.

What we found

Attack! Marketing publishes two. Beefeater, a 19.5% increase in sales across more than 2,000 in-store sampling events in 16 markets. Josh Cellars, more than four bottles sold per hour per activation at Total Wine and BevMo. Both were in stores. Neither states a baseline, a comparison period, a control, or where the number came from. Attack! also publishes a 23% sales lift for Southern Comfort Egg Nog, which we left out: despite the name, it's a non-alcoholic dairy product.

Inspira Marketing and Team Enterprises each publish one from a festival. Inspira's, published as partner content in Event Marketer rather than on its own site, credits Barone Fini wine sampling at six hot-air balloon festivals with 4,655 additional glasses bought at the wine concession next to it. Team's Patrón Hacienda program at two music festivals reports more than 450 cases sold. Both involve drinks sold at a licensed event, which makes them the nearest thing in the set to on-premise sampling evidence. Neither is a bar, and neither describes a comparison.

Mosaic publishes a Moët Hennessy case study reporting more than 154,000 store visits and more than 400,000 pieces of POS and displays. Those are activity counts. There's no sales figure anywhere in it.

Legacy Marketing Partners publishes the only bar sales figure in the set, and it isn't from sampling. Its Corona Protect Our Beaches program brought a bottle-crushing machine to bars and restaurants nationwide and reports a 13% average sales lift. It's a recycling activation rather than sampling, so it falls outside the count, but it's worth naming: the one bar sales figure in the set comes with no account of how it was measured. BeCore publishes sample counts, and GMR, Enthuse and Advantage Solutions publish nothing quantified that we could find. None of the four publishes a sales figure.

The digital platforms report activity, not sales. Bars.com publishes participation and opt-in rates and LineLeap publishes user and venue counts; Union, PayBev, Shared Spirits Marketing and Overproof publish no sampling result at all. The research firms do track on-premise sales, but as market share and trends rather than as the result of any one program. On sampling, the nearest any of them comes is NIQ's CGA survey work, in which consumers say they're more likely to order a brand they've seen activated in a venue. That's stated intent, not purchases.

Legacy Marketing Partners was named Pernod Ricard USA's national on-premise agency of record more than a decade ago. Team Enterprises has worked with Bacardi since 1989. Mosaic has run sampling for AB InBev's Labatt business in Canada. These aren't small shops that lack the data. They're firms that have run programs for three of the largest drinks companies on earth, and none of them has published what a sampling program did to sales in a bar.

A channel is measurable when a stranger owns the data

The obvious explanation is that on-premise is simply harder to measure than retail, and that's true, but it's worth being precise about why, because the reason isn't technical.

Retail sales are measurable because a third party sells the numbers. Scanner data exists as a product: nobody had to persuade a grocery chain to share anything, because a data company bought it, cleaned it, and now sells it to everyone, including people the grocer has never met. The measurement exists because it was profitable for someone with no stake in the result to build it.

On-premise has a partial equivalent, and it matters exactly where it stops. Distributors produce depletion reports (the cases they sold into each account), and brands and their agencies routinely get access to them. VIP, which builds distributor systems, describes depletions as visible "by account, chain, premise type, and class of trade," with monthly files typically arriving a week or two after close. So a brand running a program at a bar can usually see whether that bar reordered.

That's a real instrument, and it's the one the category actually uses. It is also measuring the wrong event. VIP says it plainly: depletions show "how much product is reaching retail," while scan or POS data is a separate data set that "measures what the consumer purchased at the register." A depletion records a case crossing a bar's back door. It does not record a drink crossing the bar.

For most of what a distributor sells that distinction is tolerable, because a case that goes into a bar eventually gets poured. For sampling it's fatal. The question is whether the sample changed what people ordered next, and a reorder four weeks later can't tell that apart from a busy month, a bar that ran low before the program started, or the account manager's visit the same month.

So the measurability of a sales channel turns out to be a question about which event gets recorded. The last link, a person deciding at a bar to buy a particular drink, sits in a point-of-sale system owned by the bar. Some of those records do get pooled: CGA by NIQ builds on-premise market tracking from point-of-sale data, and suppliers use it to follow share and trends by channel and region. But a projected market read is built to say how a brand is doing across bars in general, not what happened at the twelve bars where it ran a program, in the weeks it ran. At that level the records stay scattered across tens of thousands of small businesses with no reason to hand them over.

That explains the absence of data. It doesn't explain the absence of attempts.

Silence is the only move nobody gets punished for

An agency that publishes a strong on-premise result would own the category. It would be the only firm in the business able to answer the first question any skeptical CMO asks. The commercial incentive to produce one is enormous and has been for twenty years.

Follow what happens to the agency that tries. It builds measurement into a program, runs it properly, and gets a number. There are three ways that goes. The number is weak, in which case publishing it damages every pitch the agency makes for the next decade. The number is ambiguous, which is the most likely outcome at pilot scale, and publishing it invites exactly the scrutiny the category has spent its existence avoiding. Or the number is strong, and then the client, who paid for the program and whose competitors read case studies, frequently declines to let it out.

Against that, consider the agency that measures nothing. It's never contradicted, never scrutinized, and never loses a pitch on the strength of someone else's published figure, because there isn't one. Its case studies report activity, which is both impressive-sounding and unfalsifiable. "Over 400,000 pieces of POS/displays placed" cannot be wrong.

Nobody had to conspire for this to happen. Each firm, acting sensibly in its own interest, arrives at the same place, and the place is a category where producing evidence is a risk only the producer carries while the benefit is shared by everyone.

That explains the agencies. It doesn't explain everyone, and it's worth saying so before the argument gets more credit than it has earned. A trade body, a distributor, a brand with a category-leadership agenda, or an academic with the right industry access could all publish a bar result without facing an agency's asymmetric downside.

None of them has either. That's a fact the incentive story doesn't reach, and it points back at the simpler explanation above: venue-level records sit in thirty thousand small businesses, and nobody has built the thing that would collect them program by program.

Read the incentive argument as the reason no agency has gone first, not as the reason the whole field is empty. The field is empty mostly because the data was never assembled.

The category reports constantly. It reports attestations.

A great deal gets reported, through real software that real brands pay for, and the reason none of it produces a publishable result is worth understanding, because it isn't negligence.

The field-execution platforms this category runs on are capable pieces of software. An ambassador working a shift completes a report in an app before leaving: a count of interactions, samples poured against units sold, photos of the product and display, notes on placement and pricing, expense receipts. Each interaction carries a timestamp, and in some products a manager on site signs off on the conversions.

Brands get dashboards built from all of it, filterable by date, market and product, often close to real time.

Set against a clipboard and an end-of-night phone call, that's an enormous improvement, and it's made the operational half of sampling manageable: did the person show up, did they stay, was the display right, what did they see.

But look at what the number at the center of it is. Every sample count in this category originates as a figure a person typed into a phone about work they themselves performed. The photo proves someone was there. The timestamp proves when. A manager's sign-off proves a manager was willing to sign. None of them is a record of a transaction, and none of them was produced by a party with no stake in the figure.

That's the gap, and it's structural rather than a matter of anyone cutting corners. At one end of the category's measurement chain is a self-reported pour count. At the other is a monthly depletion file showing cases crossing a back door. Nothing observes the actual event in between: a person at a bar choosing a drink and paying for it.

Which is why the published record looks the way it does. You can't write a credible case study out of two instruments that both decline to measure the thing in question. The honest version of most sampling reports is "our people say they poured this many, and the distributor says the account reordered," and nobody wants to publish that sentence.

What we're not claiming

We couldn't find a published on-premise result, which is different from proving none exists. Private results surely do. A brand that ran a program and measured it internally has no obligation to publish, and the better ones probably don't.

We're also not claiming the field-execution platforms are doing their jobs badly, or that we know what any particular brand sees privately. They're built to run programs and they run them well. The argument is about what kind of evidence an attested count can be, not about the quality of the software that collects it.

Nor are we claiming sampling doesn't work. It plainly does. Event Marketer's EventTrack research, built on surveys of more than a thousand US marketers plus attendees, reports that hands-on trial raises purchase inclination for most consumers, and that free samples remain the top motivator for engagement at live events. The peer-reviewed literature on in-store sampling is more rigorous, and more mixed: it finds that trial can move purchases, by amounts that vary widely, and that part of what a free sample appears to sell is purchases that would have happened anyway.

The claim is narrower and worse than "sampling doesn't work." It's that the category sells on-premise sampling on the strength of retail evidence, and asks you to assume the two behave the same way.

They might not. A supermarket aisle and a Thursday night aren't the same purchase occasion. The grocery shopper is already shopping, has a cart, and can act on the sample inside ninety seconds. The bar guest is out with friends, is drinking what they already ordered, and the next relevant decision is days away and in a different building. The direction of the effect probably transfers. The size of it is an open question, and nobody selling into this category is treating it as one.

Some company sites wouldn't load for us in full. Where that happened we read the same case studies through published mirrors and trade press. If there's a result on a page we couldn't reach, let us know and we'll update this.

We haven't shown our work either

The audit applies to us. We sell on-premise sampling, and we've measured sales at every bar where we've run it. Our site quotes one of those results. It doesn't say how it was measured, which is exactly what this article faults everyone else for.

The difference we'd claim is mechanical rather than moral, and it's the gap described above. A sample that the guest confirms and the venue's own point of sale records is a transaction in a system neither we nor the brand controls. It's the same category of evidence as a scanner record, produced at the moment of the event rather than reconstructed from it afterward. That's the raw material a result can be built from. An attested count isn't, however carefully it's collected. We've been building results from that record in our pilots, and our wider receipt-level sales data now covers more than 500 bars. We just haven't shown our work.

Which leaves us in the same position as everyone else in the audit, with one fewer excuse. The question that ends every pitch in this category should be the same one, and it should be put to us: can you show me a bar result, and how you got it?

Not yet. But ask again in one week…