Sales lift is the increase in purchases that can be credited to an advertising campaign, measured by comparing what people who saw the ads bought with what a comparable group who did not see them bought over the same period. The gap, stated as a percentage of the control group's sales and as incremental revenue, is the lift. The measure exists because most digital reporting counts every sale that follows an ad, including the many that would have happened anyway, and because most packaged goods are still bought in physical shops that no tracking pixel reaches.

How a study is built

Every study rests on a counterfactual. The test group is the audience exposed to the campaign; the control group stands in for what that audience would have bought without it. Purchases by both are observed through transaction records rather than clicks or survey answers, and the difference is credited to the advertising.

Three designs dominate. The strongest is the randomised holdout, in which a platform splits eligible users before any auction and withholds the campaign from one cell. Early versions filled the control cell's slots with public service announcement (PSA) ads, which cost money and skewed delivery. The ghost ads method, published in the Journal of Marketing Research in 2017 by Garrett Johnson, Randall Lewis and Elmar Nubbemeyer, instead logs each moment a control user would have seen the ad and lets the next bidder win. According to the authors' working paper, a PSA control would have added about 43% to the cost of their test, which measured a 10.8% sales lift for an online sporting goods retailer.

The second design randomises places rather than people. Google's Conversion Lift based on geography assigns regions to test and control, which lets it count offline as well as online conversions across channels, though Google's documentation describes the results as noisier and the required budgets as typically higher than user-based studies.

The third, and the workhorse for television and cross-publisher campaigns, is the matched panel. Where randomisation is impossible, a measurement firm identifies exposed households, then builds a control from unexposed households with similar demographics and, above all, similar purchase histories before the campaign. The sales data come from loyalty cards, receipt-scanning panels or payment card records. Catalina alone links 130 million US households and more than 400 million shopper IDs across 70 retail banners. The design is quasi-experimental: it controls for whatever the matching variables capture and nothing else.

A worked example shows the arithmetic. Suppose a hypothetical snack campaign reaches 2 million households, which spend an average of $3.12 on the brand over 12 weeks, while matched controls spend $3.00. Lift is 4%, or 12 cents per household and $240,000 across the exposed base. Against $120,000 of media, incremental return on ad spend (iROAS) is $2.00 per dollar. An attributed return on ad spend (ROAS), crediting every purchase by an exposed household, would divide $6.24 million by the same $120,000 and report 52.

Packaged goods studies usually split the gain further into more buyers (penetration), more trips per buyer and more spend per trip, which shows whether a campaign recruited households or sold more to existing ones. Circana's NCS unit says its reports arrive within five weeks of a flight's end.

Who runs it

Advertisers and agencies commission studies and increasingly configure them inside buying platforms. Amazon DSP began letting advertisers self-serve third-party studies on May 15, 2026, with NCS Offline Sales Lift among them.

Sellers run the rest. Platforms holding first-party purchase data measure their own inventory: DoorDash introduced a ghost ads-based Sales Lift Measurement product for Sponsored Products on March 24, 2026, and Walmart Connect is testing closed-loop measurement of YouTube campaigns against Walmart store and online transactions. Independent suppliers, among them Circana, ABCS Insights, Affinity Solutions and Samba, supply the purchase data where a seller has none.

Origin and evolution

The method predates the internet. In January 1980 Information Resources Inc (IRI) launched BehaviorScan in Marion, Indiana and Pittsfield, Massachusetts, according to the International Directory of Company Histories. About 2,000 households in each town shopped with identification cards at scanner-equipped supermarkets, while devices on their televisions let IRI substitute different commercials into cable broadcasts. Split-cable testing allowed a brand to show heavier advertising to half a town and read the difference at the checkout.

The results were sobering. A meta-analysis of 389 BehaviorScan experiments run between June 1982 and December 1988, published by Leonard Lodish and colleagues in 1995, found that heavier advertising produced a significant sales gain in only 33% of tests for established brands and 55% for new products, even at a lenient 80% one-tailed confidence threshold.

Scanner data met media panels on December 14, 2009, when Nielsen and Catalina Marketing formed a 50-50 joint venture, Nielsen Catalina Solutions, combining Nielsen's television and internet panels with purchase data from 50 million shoppers. Its Sales Effect product launched in 2010 and was rebuilt around machine learning, with reads while campaigns are still running, in June 2022, according to NCS.

Digital platforms followed. Facebook disclosed in September 2012 that it was matching ad exposure against Datalogix loyalty-card records from roughly 70 million US households, and said that in 70% of the 45 campaigns measured each dollar returned $3 in incremental sales. Oracle agreed to buy Datalogix on December 22, 2014, when it covered 110 million households. In July 2021 Google announced NCS sales lift reporting for connected TV (CTV) campaigns in Display & Video 360, covering percentage lift, incremental sales and ROAS without pixels.

Consolidation came next. Circana agreed to buy NCS in August 2024, weeks before Oracle shut its advertising business on September 30, 2024. The NCS deal completed on June 2, 2025.

Why it matters

Sales lift is the nearest thing advertising has to a receipt, which is why consumer packaged goods (CPG) marketers treat it as a currency. In a 2022 survey by Brand Innovators and NCS, more than 70% of CPG marketers said they relied on incremental sales or sales lift to judge outcomes, and 41% called sales lift reports very important when deciding whether to advertise with a publisher. That second figure helps explain why publishers pay for studies of their own inventory.

Retail media has sharpened the argument. An IAB paper co-written with Instacart argued in April 2026 that marketing mix modelling undercounts retail media and that causal, closed-loop iROAS should lead budget decisions. The IAB's Guidelines for Incremental Measurement in Commerce Media, published on November 3, 2025, rank randomised tests, holdouts and matched markets as the strongest evidence and place platform-reported incrementality in the weakest tier.

Limitations and disputes

Noise is the fundamental problem. Randall Lewis and Justin Rao analysed 25 field experiments for the Quarterly Journal of Economics in 2015 and found that the standard deviation of individual sales is typically ten times the mean. The median confidence interval on return on investment exceeded 100 percentage points; telling a 0% return from a 10% return would have required campaigns 62 times larger, according to the paper.

Matching does not fix selection. Ad systems deliver to people already likely to buy, the endogeneity that inflates observational estimates. Across 663 Facebook experiments, Brett Gordon, Robert Moakler and Florian Zettelmeyer found a median experimental lift of 5% for lower-funnel outcomes, against 24% from a machine learning model applied to the same data. Karan Dhir, a product manager at Genentech, argued in September 2026 that platforms still present experimental and modelled results side by side.

Coverage is partial. Loyalty panels see only participating retailers, and card data records the merchant and amount rather than the items in a basket, which suits restaurants and telecoms better than a cereal maker.

Much published evidence comes from sellers. A TikTok-commissioned study with Samba reported a median 172% rise in ticket purchase rates across 38 film campaigns; Grocery TV and ABCS Insights cited 11.7 times iROAS for a candy brandwithout disclosing how controls were built.

Windows are short. TransUnion and MMA Global case studies found long-term effects 1.8 to 6 times larger than short-term ones, which a 12-week read cannot see.

Not the same as

Brand lift uses the same exposed-versus-control logic but measures survey answers such as awareness, not transactions.

Conversion lift applies the design to any tracked event, including sign-ups and app installs. Sales lift is the subset concerned with purchases, often offline.

Attributed ROAS credits all sales following an ad within a window, including view-through conversions after unclicked impressions. It measures association, not cause.

Promotional lift, in retail analytics, is the volume a price cut or display adds over baseline. It concerns trade promotions, not media.

Recent developments

Purchase data is consolidating. Infillion bought Catalina in February 2026 and made its US data available only through Infillion's platform. Nielsen began offering Predictive Sales Lift in April 2026, forecasting results mid-flight from hundreds of past campaigns; Nielsen describes the output as directional, not a replacement for full studies.

Measurement is also merging. Cint combined brand and sales lift in one dashboard on June 17, 2026, using Affinity Solutions card data, and today added the combined studies to its self-service study builder, with linear TV in beta. Laura Manning, Cint's senior vice president of measurement, said "marketers are under increasing pressure to show how changes in consumer perception ultimately connect to the bottom line".

Timeline

  • January 1980: IRI launches BehaviorScan split-cable testing in Marion, Indiana and Pittsfield, Massachusetts
  • 1995: Lodish and colleagues publish a meta-analysis of 389 BehaviorScan advertising experiments
  • December 14, 2009: Nielsen and Catalina Marketing form Nielsen Catalina Solutions
  • 2010: NCS launches its Sales Effect measurement product
  • September 2012: Facebook discloses offline sales matching with Datalogix across about 70 million US households
  • December 22, 2014: Oracle agrees to buy Datalogix
  • 2015: Lewis and Rao publish their analysis of 25 advertising experiments in the Quarterly Journal of Economics
  • 2017: Johnson, Lewis and Nubbemeyer publish the ghost ads method
  • July 28, 2021: Google announces NCS sales lift for CTV campaigns in Display & Video 360
  • June 6, 2022: NCS releases a machine learning version of Sales Effect with in-flight reads
  • August 26, 2024: Circana agrees to acquire NCS
  • September 30, 2024: Oracle's advertising products reach end of life
  • June 2, 2025: Circana completes the NCS acquisition
  • November 3, 2025: IAB publishes Guidelines for Incremental Measurement in Commerce Media
  • February 2026: Infillion acquires Catalina
  • March 24, 2026: DoorDash Ads launches ghost ads-based Sales Lift Measurement
  • April 2026: Nielsen introduces Predictive Sales Lift; Grocery TV adds ABCS Insights sales lift data
  • May 15, 2026: Amazon DSP opens self-service third-party studies including NCS Offline Sales Lift
  • June 17, 2026: Cint combines brand and sales lift in one dashboard
  • September 28, 2026: Cint makes combined brand and sales lift studies self-service

Summary

Who: Advertisers and agencies commission sales lift studies; platforms such as Meta, Google, Amazon, DoorDash and Walmart run them on their own inventory; independent firms including Circana's NCS, ABCS Insights, Affinity Solutions, Samba and Cint supply purchase data and analysis.

What: A comparison of purchases by people exposed to a campaign against a randomised or matched control group, reported as percentage lift, incremental sales and incremental return on ad spend.

When: Split-cable testing began with IRI's BehaviorScan in January 1980; Nielsen Catalina Solutions brought purchase-based measurement to television and digital from December 2009, and platform, retail media and modelled variants multiplied through 2026.

Where: Inside ad platforms and demand-side platforms, retail media networks and measurement vendors' dashboards, drawing on loyalty card, receipt panel, payment card and retailer transaction data, mostly in the United States.

Why: Attributed sales count purchases that would have happened anyway, and most packaged goods are bought offline. Sales lift estimates what advertising actually added, though noisy sales data, imperfect matching, partial retailer coverage, seller-run studies and short measurement windows limit what one study proves.