By Donald Gorassini, howinfluenceworks.com
This blog discusses the impact of commercial advertisements on product sales, focusing on three key areas: measuring advertising success, understanding how advertising influences sales, and determining if ad exposure leads to sales that surpass advertising costs.
Commercial advertising can cause additional sales, but its effects vary substantially across campaigns. The average effect is generally modest, and estimating it accurately is more difficult than the abundance of advertising and sales data might suggest. Moreover, an increase in sales or revenue does not necessarily mean that the revenue exceeds advertising expenditure.
McGuire’s Question
Do we know if advertising affects sales? How would we know?
In a frequently referenced commentary, social psychologist William McGuire challenged the assumption that mass media necessarily exert large, direct effects on attitudes and behavior. Commercial advertising provided a particularly relevant example: despite the industry’s enormous scale, establishing how much advertising actually changes purchasing behavior presents a difficult empirical problem.1McGuire, W. J. (1986). The myth of massive media impact: Savagings and salvagings. In G. Comstock (Ed.), Public communication and behavior (Vol. 1, pp. 173–257). Academic Press.
Based on his cognitive sequence model of responses to advertising, McGuire proposed that much could go wrong along the path from an advertising stimulus to a purchase response. According to the model, several responses must occur in between: Exposure → Attention → Comprehension → Acceptance → Retention → Action (Purchase). In this model, failure at an intermediate stage can prevent the process from culminating in a purchase.
McGuire’s thesis requires adequate assessment methodology. As it turns out, good methods of evaluation, especially as they have evolved in recent years, provide solid data for addressing McGuire’s concerns.
In this blog I will be concerned with describing what those methods are and what they say about the nature of advertising effect magnitude and variability. Essentially, I address the causal effect of advertising exposure on purchasing. In a future blog, I will focus on intervening events like those stipulated in McGuire’s model.
Determining if Advertising Works
Commercial advertising is ubiquitous, and companies spend enormous sums on it. It might seem straightforward to establish whether advertising increases sales: compare sales before and after a campaign. But a rise in sales following an advertising campaign does not, by itself, establish that the advertising caused the increase.
After launching the ad campaign utilizing the jingle, famous at the time, “plop plop, fizz fizz, oh, what a relief it is” for Alka-Seltzer, both its company and its advertising firm claimed a doubling in sales due to the ad.2The ad was catchy. The claim of great success was based on unsystematic observations, corporate claims, and reference to internal company data, which were not available to the public. See article here. How would they have known this, if they knew it at all?
A significant challenge in the endeavor to assess advertising effectiveness is applying a method capable of testing cause and effect. This has been historically difficult. However, advancements in testing methods are now providing clearer insights into causality, effect magnitude, and return on ad spending.
Three methods for evaluating ad performance include correlational, quasi-experimental, and experimental. The first two emerged during early assessment efforts and remain in use; both are observational. The experimental approach has gained prominence recently, drawing particular benefit from digital environments.
Correlational methods assess the relationship between ad strength, often measured by campaign spending, and product sales. A correlation between stronger ads and higher sales suggests a potential causal link, but does not demonstrate it, this due to possible confounding factors. For instance, if stronger ads target higher income consumers, the observed sales pattern may result from consumers’ spending ability rather than ad strength. Researchers try to address this by identifying and controlling for confounds using statistical methods, which can increase confidence in a causal interpretation, though the possibility of unknown confounds will always remain.
Quasi-experiments in advertising compare groups exposed to advertising with comparison groups that were not exposed, without randomly assigning participants to the groups. Conditions of a study are tested on participant samples with preexisting differences, which can introduce confounding issues similar to those in correlational studies. As with the correlational method, researchers employ statistical methods to account for known potential confounds. Researchers sometimes have the ability to select comparison groups that on the surface appear very similar.
Experiments use two essential tactics in order to test causal relationships. In one, the researcher applies a manipulation by establishing two or more conditions that represent the putative cause; in advertising research, this might involve showing ads to one group while withholding them from another, a setup called a withholding or expose-withhold design. The other tactic consists of randomly assigning participants to the conditions before the experiment.
Random assignment makes the treatment and control groups comparable in expectation, balancing both measured and unmeasured characteristics on average.3As sample size increases in an experiment, the likelihood that random assignment fails to balance groups declines. Put differently, large samples sharply lower the risk of “random assignment failure,” meaning baseline differences between conditions. The substantial samples in the studies discussed here make such failure highly unlikely in any single campaign. If repeating the study yields the same outcome, confidence increases further. Caution is still needed, though. Even when random assignment failure becomes less probable, it is never entirely removed as a possibility. A related risk in any experiment is weak design, where the procedure itself introduces confounds. Consequently, a difference in purchasing behavior between the groups can be attributed to the advertising treatment, subject to sampling uncertainty and the assumptions required for a valid experiment.
Research on advertising effectiveness has moved from observational designs toward experimental ones. These experiments occur in the field, meaning that they are conducted with actual consumers, in real settings, offering actual purchase opportunities, and measuring authentic purchase behavior. Accordingly they are termed randomized field experiments.
Randomized field experiments became increasingly practical as digital advertising platforms made it easier to assign consumers to conditions and measure subsequent behavior. Offers could be made to consumers, who had the opportunity to buy the goods or services offered. It was also due to the adaptability of the setting to conduct experimentation. Manipulations could be delivered and participants, often in large numbers, could be randomly assigned to conditions and observed. As with so many advances in technology in history (e.g., the microscope), these advances facilitated scientific progress.
Correlational studies and quasi-experiments offered suggestive evidence on the impact of advertising on sales. It was then possible using the subsequently instituted experimental work to see if the pattern of findings would hold. I will now follow the evidence on advertising’s sales impact.
Impact of Advertising on Sales
The influence of advertising on sales resolves to not only if there is an impact but also to how big the effect is in a typical ad campaign and how much it varies over campaigns. All these matters are consequential.
Summarizing Advertising Effects
Researchers often summarize advertising effectiveness using a single numerical estimate. In an experiment, for example, they may calculate the difference in average spending or purchasing between an advertising-exposed group and a control group. Across multiple campaigns, researchers may summarize the estimated effects using a mean or median. Such summaries are useful, but they can conceal substantial differences in effectiveness from one campaign to another.
Across a substantial body of research, estimates generally indicate that advertising can increase sales, but the size of the estimated effect varies considerably.4Correlational methods revealed a positive relationship between advertising strength and product sales, but the estimated effect size was small. Subsequent studies using better evaluation techniques indicated even smaller impacts, but impacts nonetheless. See, for example, Sethuraman, R., Tellis, G. J., & Briesch, R. A. (2011). How well does advertising work? Generalizations from meta-analysis of brand advertising elasticities. Journal of Marketing Research, 48(3), 457–471.,5A 1995 meta-analysis examined the impact of many television advertising campaigns on product purchases by comparing two similar groups of households, one exposed to ads and the other not, across 389 tests from 1980 to 1988. The households were matched according to demographics and prior purchase behaviour. The findings indicated that advertising generally led to increased sales. Effectiveness varied considerably over individual tests. See Lodish, L. M., Abraham, M., Kalmenson, S., Livelsberger, J., Lubetkin, B., Richardson, B., & Stevens, M. E.(1995). How T.V. advertising works: A meta-analysis of 389 real world split cable T.V. advertising experiments. Journal of Marketing Research, 32(2), 125–139.,6In a meta-analysis of 432 randomized field experiments using the “Ghost Ads” methodology. Across the hundreds of campaigns, the researchers found a median lift value of 8% for conversions (any desired action set by the advertiser after interacting with an ad, often a purchase) and 17% for website visits. It is important to specify the outcome measure in interpreting lift values. They also found high variability in lift irrespective of measure but consistently showed that modern, platform-integrated experimentation could establish a statistically solid baseline for ad increases and positive returns. See Johnson, G., Lewis, R. A., & Nubbemeyer, E. (2017). The online display ad effectiveness funnel & carryover: Lessons from 432 field experiments. Available at SSRN 2701578. Average effects are often modest, although estimates from different studies are not always directly comparable because they use different methods, outcomes, and settings.
This finding, of a modest overall effect, while informative, is far from the whole story. A look at variability of advertising impacts over campaigns provides needed context for interpreting single numerical estimates of impact.
High Variability of Impacts Over Campaigns
The same above research that reports a modest positive effect overall of advertising on sales shows substantial variability across individual campaigns. Some studies find convincing positive effects, others find effects that are small or statistically indistinguishable from zero, and some produce negative estimates. The strongest negative estimates may indicate that advertising reduced sales, although uncertainty must be considered before drawing that conclusion.
The research addresses McGuire’s fundamental questions about average effects and how to demonstrate them, while introducing a third question regarding the varying effectiveness of different ads. Why are some ads so much more successful than others? The current blog aims to address McGuire’s concerns, while a subsequent blog will explore the characteristics that differentiate more effective ads from less effective ones.
Effect of Methodology
Research methods should reflect advertising impacts. As it turns out, some methods early in the history of advertising research have introduced bias in the assessment process and distorted what the measures seem to reveal.7Korkames, J., Stanley, T. D., & Stremersch, S. (2026). Meta-analysis of advertising effectiveness: New insights from improved bias corrections. International Journal of Research in Marketing, 43(1), 131–151.
One large correlational research project from the period examined method bias.8Assmus, G., Farley, J. U., & Lehmann, D. R. (1984). How advertising affects sales: Meta-analysis of econometric results. Journal of Marketing Research, 21(1), 65–74. Several statistical models for removing confounds and thereby arriving at a pure gauge of advertising effect strength were applied. As all the models, 128 in total, were designed to reflect the same outcome, they should all have given approximately the same estimates for an advertising component, varying only by random error. It turned out that estimates across models varied considerably. This suggests caution when interpreting data generated using a correlational method.
A large investigation compared randomized field experiments with observational approaches for measuring the impact of digital advertising on sales.9Gordon, B. R., Zettelmeyer, F., Bhargava, N., & Chapsky, D. (2019). A comparison of approaches to advertising measurement: Evidence from big field experiments at Facebook. Marketing Science, 38(2), 193-225. The study addressed three questions: whether randomized experiments demonstrate a positive effect of advertising on consumer purchases, whether observational approaches also indicate positive effects, and how closely the observational estimates correspond to those obtained from randomized experiments, which were treated as the benchmark for causal inference. The researchers analyzed data from 15 advertising campaigns conducted on Facebook involving approximately 500 million user-experiment observations and 1.6 billion advertising impressions.
In the randomized experiments, Facebook users were assigned either to a treatment condition or a control condition. Users in the treatment condition were eligible to receive the advertiser’s ads during the campaign period. Users in the control condition were prevented from receiving those ads. Because assignment was random, differences in purchasing behavior between the groups could be interpreted as the causal effect of advertising.
The experimental results generally indicated that advertising increased sales, although the effects were often small, and they varied considerably across campaigns. Some campaigns produced little measurable enhancement, whereas others generated substantially larger effects. Overall, the pattern of results was consistent with previous randomized field experiments, which have typically found modest average advertising effects together with substantial variability across campaigns and brands.10The researchers’ analysis evaluates lift values from 15 experiments, measuring the sales increase in the experimental group compared to the control group. The score represents the percentage increase in sales, with a score of 3 indicating a 3% higher sales in the experimental group than the control group, while a score of 0 indicates no sales difference between the two groups. For example, if the control group sold 100 units and the experimental group sold 103, the score is 3%; if both sold 100, the score is 0%. Although, when aggregated over studies, sales lift was 3.6%, and several of the individual values tended to cluster around this value, there is also considerable variation among the 15 scores. There are values that do not differ from zero, are negative, and even highly positive. Negative estimates suggest that advertising may have reduced sales in those campaigns, although the statistical uncertainty surrounding each estimate must be considered before concluding that the effect was genuinely negative. The very high scores indicate a particularly robust impact of advertising on sales, but even here statistical uncertainty must be assessed. Clearly, something important in the individual advertising campaigns is modulating their impact.
The researchers then examined what would happen if the same campaigns were analyzed using observational methods. Instead of relying on random assignment, they compared users who happened to be exposed to an advertiser’s ads with users who were not exposed. These exposed and unexposed groups differed naturally in many ways that could also influence purchasing behavior. To address these differences, the researchers employed a variety of statistical adjustments, including advanced matching procedures and models incorporating extensive demographic and behavioral variables.
The observational analyses frequently produced positive estimates of advertising effectiveness. However, these estimates often differed substantially from the experimentally measured effects. The observational methods overestimated advertising effectiveness on average by a factor of roughly three. In several instances, the discrepancy was much larger than this. More importantly, observational estimates were generally poor predictors of the sales enhancement measured by the randomized experiments. Even sophisticated matching procedures applied to highly detailed user-level data in the observational studies failed to reliably reproduce the experimental results.
Overall, the correspondence between observational and experimental findings was weak. Had advertisers relied solely on the observational analyses, they often would have obtained a misleading picture of the true incremental impact of advertising on sales. The results suggest that rich data and advanced statistical techniques are not necessarily sufficient to eliminate the biases that arise when ad exposure is not randomly assigned.
The authors concluded that observational methods based on the types of data typically available in the digital advertising industry often fail to accurately measure advertising effects. Randomized field experiments provide particularly strong evidence of causal effects. The comparison shows that even sophisticated observational methods can produce substantially misleading estimates when the factors determining advertising exposure also influence purchasing behavior.
To this point, then, we can conclude that advertising affects sales, estimated effects are often modest, the effect varies considerably across individual campaigns, and the issue of why the variability across campaigns occurs needs to be addressed.
Return on Ad Spend
Even when an advertising campaign boosts sales, a company might not recover the costs of the advertising. The expense of the endeavor could exceed the revenue generated by it. Consequently, a metric comparing sales to investment offers a useful gauge. Accordingly, researchers have examined return on ad spend (ROAS).11ROAS should not be confused with profit. Profit is typically defined as total revenue from a product or business activity minus all associated expenses. The additional revenue generated by advertising contributes to total revenue, while advertising expenditures constitute only one component of total costs. Other expenses, such as production, inventory, labor, shipping, and overhead, must also be considered when determining profit.
The effects of advertising on sales and on ROAS using two methods were assessed in a large observational investigation.12Shapiro, B. T., Hitsch, G. J., & Tuchman, A. E. (2021). TV advertising effectiveness and profitability: Generalizable results from 288 brands. Econometrica, 89(4), 1855–1879. In a correlational approach, the investigators analyzed varying levels of advertising exposure and its association with sales and ROAS. Separately, in a quasi-experimental approach, they employed an expose-withhold design, comparing groups from adjacent geographical areas but not randomly assigned. In both approaches, the investigators used statistical methods to control for possible confounds.
In both instances, the researchers found the familiar pattern of a weak positive impact overall and high variability when it came to sales and when it came to ROAS. For both outcomes, some values resided around the single numerical estimate, some values failed to differ from zero, and a few either fell below or considerably above the single estimate. The findings should be interpreted with caution due to the observational study design.
Large studies employing the experimental approach have assessed ROAS and found the typical pattern of positive single numerical estimates of impact and much variation across advertising campaigns in these values.13This study also found a 3.6% increase in sales relative to the control group. Johnson, G. A., Lewis, R. A., & Reiley, D. H. (2017). When less is more: Data and power in advertising experiments. Marketing science, 36(1), 43-53. In a very large randomized field experiment on Facebook and Instagram, researchers assessed advertising effectiveness by randomly allowing some users to see advertisers’ ads while withholding those ads from others. The experiment encompassed more than 700,000 advertising campaigns conducted by over 200,000 advertisers. The researchers compared treatment and control groups on measures including sales revenue and ROAS.14Tadelis, S., Hooton, C., Manjeer, U., Deisenroth, D., Wernerfelt, N., Dadson, N., & Greenbaum, L. (2023). Learning, sophistication, and the returns to advertising: Implications for differences in firm performance (No. w31201). National Bureau of Economic Research.
Consistent with many previous advertising experiments, the advertising-exposed groups generally outperformed the control groups. Exposure to advertising increased revenues, purchases, and the number of purchasers. ROAS was expressed as incremental revenue generated per dollar spent on advertising. In this study, each advertising dollar generated approximately $3.31 in additional revenue overall. It is important to note that advertising can increase revenue by more than the amount spent on advertising, but that alone does not establish that a campaign is profitable. Profitability depends on the contribution made by the additional sales after accounting for all the relevant costs, not just that of advertising.
The researchers also found substantial variation in ROAS values generated, indicating that the effectiveness of advertising differed markedly from one campaign to another.15The researchers also provided evidence that a portion of this variation may reflect differences in advertiser experience, engagement, and sophistication, suggesting that some firms learn to use advertising more effectively over time. I will address why some ads are more successful than other in the next blog.
Advertising not only affects sales compared with a control group but can also justify its expense by generating revenue above the project cost. Like sales effects, advertising returns vary considerably across campaigns. Advertising can generate substantial additional revenue in some cases, but the financial return is not uniformly positive, and revenue-based measures alone do not establish a positive return on advertising investment.
Why All the Variability?
Why all the variability? This is a crucial concern spurred by the research reviewed. Consider this thought experiment: You wish to develop an ad campaign for your product and have read this blog. You know that ads can produce a positive effect and bring in additional revenue that may exceed the advertising expenditure. This creates room for optimism. An ad can work. This is good but far from adequate information on how to go about it.
Due to the high variability of ad success, you know your ad could succeed, possibly wildly, or it could fail, perhaps by a large amount. You would not know, yet, what past enquiry suggests would make your ad the type likely to succeed. Based on the present blog alone, you would not be in a good position to begin developing an ad campaign. To fill this void, the next blog will focus on factors known to impact ad success.
Conclusions
Does commercial advertising affect sales? The evidence indicates that it can: randomized field experiments have demonstrated that advertising causes additional purchases. However, the average effect is generally modest, and the impact varies considerably across campaigns. Some campaigns produce substantial increases in sales, while others have little measurable effect.
Determining the size of the effect is challenging. Sales and advertising often rise together, but observational data can give a misleading picture of how much advertising itself contributes to sales. Randomized experiments provide particularly strong evidence of causation, although their findings must still be interpreted in light of the products, consumers, and circumstances studied.
Finally, an increase in sales does not necessarily mean that advertising is a well founded investment. That depends on the additional revenue generated compared to the relevant costs.
The findings offer a broader lesson about social influence: the existence of an influence attempt, or the confidence with which it is undertaken, does not establish its effectiveness. Why do some advertisements succeed much more than others? The next blog will examine the factors that help explain this variation.
Leave a Reply