Campaign numbers can show that marketing activity coincided with sales, but they cannot, by themselves, prove which sales the campaign caused. Reliable answers require connecting campaign exposure to transaction data and comparing outcomes with a credible estimate of what would have happened without the campaign. That distinction matters: Nielsen has reported that, across the campaigns it studied, only about half of measured media investments produced positive return on investment. This article explains what campaign attribution can and cannot tell you, how attribution differs from incrementality and marketing mix modeling, and how to build a measurement approach that supports better budget decisions.
Campaign measurement defines what sold and what the campaign caused
Campaign measurement is the collection and analysis of data about marketing exposure, customer actions, and business outcomes. The American Marketing Association describes marketing analytics as the practice of managing and analyzing data to evaluate marketing performance and maximize its effectiveness. In sales measurement, the key distinction is between observed sales—transactions recorded after or during a campaign—and incremental sales—transactions that would not have occurred in the campaign’s absence.
A campaign dashboard may report impressions, clicks, attributed conversions, revenue, and return on ad spend (ROAS). Those figures describe activity under a particular measurement method; they do not automatically establish causation. For example, a shopper who was already planning to buy may click a search ad just before purchasing. The ad receives credit in a last-click report, but the click may not have created the sale.
This is commercially important because measurement affects where future budgets go. Nielsen’s analysis of campaigns across industries found that only 54% of measured media investments generated positive ROI. The finding is a warning against treating delivery metrics or attributed revenue as proof that every campaign worked. A useful measurement plan therefore states the outcome, attribution rules, comparison method, and limitations before results are interpreted.
Attribution models assign credit to campaign touchpoints
Marketing attribution is a method for assigning credit for a conversion or sale to one or more marketing interactions. Google Analytics documentation describes attribution models as rules, or sets of rules, that determine how credit for key events is assigned across touchpoints in a user’s path. Attribution answers, “Which interactions receive credit under this model?” It does not necessarily answer, “Which interactions caused additional sales?”
Common attribution hyponyms include single-touch models, such as first-click and last-click, and multi-touch models, such as linear, position-based, and data-driven attribution. Each organizes the same journey differently. A customer might see a social ad, search for a product, click a paid search result, and then purchase. Last-click may assign all credit to paid search; first-click may assign it to social; a multi-touch method may distribute credit across both.
That variation is useful for understanding customer paths, but it can also produce conflicting channel reports. Attribution can be affected by incomplete cross-device tracking, consent choices, platform-specific reporting windows, identity gaps, and offline purchases. The practical bridge is to treat attribution as a descriptive view of the journey, then validate consequential budget decisions with experiments or broader models that estimate causal lift.
Single-touch attribution is simple but narrow
Single-touch attribution assigns all conversion credit to one selected interaction. First-click credits the first recorded touchpoint; last-click credits the final one before conversion. These models are easy to explain and can be useful for quick reporting or for campaigns with short, simple paths to purchase.
Their limitation is that they omit other interactions by design. If a customer discovers a brand through a video ad but buys after a branded search, last-click can make search appear to have created the entire sale. Google Analytics notes that changing attribution models can change how credit is allocated in reports, even though the underlying conversion has not changed. Teams should therefore label the model and avoid comparing figures produced under different rules as if they were equivalent.
Multi-touch attribution distributes credit across the journey
Multi-touch attribution allocates credit to multiple recorded interactions. A linear model can divide credit evenly; a position-based model can emphasize selected positions, such as the first and final touch; and data-driven models use observed paths and model assumptions to estimate credit. These approaches can offer more journey detail than a single-touch report.
More detail does not guarantee causal accuracy. The model still depends on which interactions can be observed, how identity is matched, which conversion window is used, and what assumptions govern credit allocation. A report that assigns 30% of a sale to a channel is not evidence that removing that channel would reduce sales by 30%. Use multi-touch results to generate hypotheses about channel roles, then test those hypotheses where the decision warrants it.
Incrementality testing estimates sales caused by a campaign
Incrementality is the additional outcome produced by a marketing activity compared with an appropriate estimate of the outcome that would have happened without it. A randomized controlled experiment is a strong way to estimate this effect: eligible people, households, geographic areas, or other units are assigned to a treatment group that receives the campaign or a control group that does not. The difference in outcomes estimates lift, subject to the experiment’s design and implementation.
The U.S. National Institute of Standards and Technology’s statistical guidance explains that random assignment helps make treatment and control groups comparable, reducing bias from pre-existing differences. In practice, marketing teams may use platform conversion-lift studies, customer holdouts, matched geographic tests, or switchback tests. A well-designed test defines the outcome in advance, uses a sufficient sample, limits contamination between groups, and accounts for the campaign’s expected time to affect sales.
Incrementality is especially valuable when attribution reports disagree or when a channel reaches customers likely to buy anyway. It also makes uncertainty visible: results should include an estimate of lift and an uncertainty interval, not only a point estimate. A test may be inconclusive because the effect is small, the sample is underpowered, or external conditions changed—not necessarily because the campaign had no effect.
Holdout and randomized lift tests compare exposed and unexposed groups
A holdout test withholds campaign exposure from a randomly selected control group while the treatment group remains eligible to receive the campaign. The analyst compares outcomes such as purchase rate, units sold, or revenue per eligible customer. For instance, an advertiser could reserve a portion of eligible customers from a promotional campaign and compare their purchases with those of randomly selected customers who receive it.
The comparison is most credible when assignment occurs before exposure, groups are large enough, and the control group is not reached through another route that recreates the treatment. Teams should also check whether the test measures the intended business outcome: a lift in conversions may not mean a lift in profitable sales if discounts, returns, or customer acquisition costs are excluded.
Geo experiments test campaigns where user-level randomization is difficult
A geographic experiment assigns regions, markets, or store catchment areas to treatment and control conditions. It is useful when exposure is difficult to randomize at the person level, such as for television, out-of-home, or broad local media. Researchers compare changes in sales between treated and comparison locations, ideally matching areas on historical performance and relevant market characteristics.
Geo tests can be distorted by spillover across boundaries, uneven local promotions, weather, inventory shortages, or too few comparable regions. A design should select markets before launch, document other activity, and use historical data to assess balance. When possible, analysts can use difference-in-differences methods, comparing the treatment-control gap before and after the campaign rather than relying only on raw post-campaign sales.
Marketing mix modeling measures sales across channels and time
Marketing mix modeling (MMM) is a statistical approach that uses historical business and marketing data to estimate how channels and other factors relate to sales over time. Google’s Meridian documentation describes MMM as a method for estimating the contribution of marketing activities while accounting for factors such as seasonality and other variables. Typical models may include media spend or impressions, price, promotions, distribution, holidays, and economic conditions.
Unlike person-level attribution, MMM generally works with aggregated time-series or geographic data, so it can evaluate channels that are difficult to track at the individual level. It can also estimate diminishing returns and support budget scenario planning. Its outputs are estimates, not a direct accounting of every sale: results depend on data quality, model specification, sufficient variation in marketing activity, and validation against experiments or business knowledge.
MMM and experiments answer complementary questions. An experiment can provide a more direct causal estimate for a specific campaign or audience, while MMM can offer a wider view across channels and longer periods. The strongest programs use experiments to calibrate or challenge model estimates and use MMM to guide planning between tests.
Sales data and control variables shape model reliability
A useful MMM dataset aligns sales with marketing activity over consistent time periods and includes variables that plausibly influence demand. These can include product price, discounts, distribution, stock availability, seasonality, competitor actions, and major events. If a product was out of stock, for example, low sales may reflect availability rather than weak advertising.
Analysts should distinguish correlation from causal evidence and check whether the data contain enough independent movement to separate channel effects. If television and digital spending always rise together, a model may struggle to identify each channel’s separate contribution. Documentation of assumptions, diagnostics, and validation results is essential for interpreting the estimates responsibly.
ROAS and profit metrics translate estimates into decisions
Return on ad spend is commonly calculated as revenue attributed to advertising divided by advertising spend. Incremental ROAS instead uses revenue estimated to be caused by advertising divided by spend. Neither metric necessarily equals profit: margins, discounts, fulfillment, returns, and other variable costs can materially change the business result.
For a decision-ready report, show the sales or revenue definition, whether the figure is attributed or incremental, the time window, the cost included, and uncertainty where available. A campaign can report high attributed ROAS yet generate little incremental demand, while another may appear modest in last-click reporting but contribute meaningfully to discovery or future purchases. The relevant metric depends on the decision and the economics of the business.
A measurement plan connects campaign data to actual sales
A practical measurement plan combines transaction records, campaign delivery data, and a method appropriate to the business question. It should distinguish revenue associated with a campaign from revenue caused by it, and preserve enough context to reproduce the analysis later.
- Define the outcome: specify whether “sold” means orders, units, net revenue, gross profit, or another measure, and account for cancellations and returns.
- Set the causal question: state what would be compared if the campaign had not run, and choose attribution, an experiment, MMM, or a combination based on the decision.
- Join data carefully: align campaign dates, exposures, transactions, products, markets, and costs while respecting consent and privacy requirements.
- Check alternative explanations: review seasonality, promotions, pricing, stockouts, competitor activity, and other campaigns.
- Report uncertainty and limitations: include sample size, measurement window, model rules, confidence or credible intervals where appropriate, and known data gaps.
- Use findings to act: shift budgets only when the evidence is strong enough for the decision, and retest after material changes in audience, creative, offer, or channel.
A simple comparison chart can clarify the roles of the main methods: attribution describes how credit is allocated across observed touchpoints; incrementality testing estimates lift against a control; MMM estimates channel contributions from aggregated historical data. A dashboard should display these as different kinds of evidence rather than blending them into a single “sales caused” number.
Conclusion: campaign numbers need causal context
Campaign numbers can tell you what was recorded and how a chosen model distributed credit, but they cannot alone establish what actually sold because of the campaign. Attribution is useful for understanding customer paths; incrementality tests estimate additional sales against a counterfactual; and marketing mix modeling helps assess broader channel contributions over time. Revenue, ROAS, and conversion metrics become decision-worthy only when their definitions, costs, assumptions, and uncertainty are clear.
Before making the next major budget decision, label attributed and incremental results separately, inspect the underlying transaction and exposure data, and test high-impact assumptions with a holdout or geographic experiment where feasible. Further reading on attribution, experimental design, and marketing mix modeling can help teams build evidence that is both more transparent and more useful.
Sources: Nielsen, “Only 54% of Media Investments Drive Positive ROI, Nielsen Analysis Finds,” https://www.nielsen.com/news-center/2023/only-54-of-media-investments-drive-positive-roi-nielsen-analysis-finds/; American Marketing Association, “What Is Marketing Analytics?,” https://www.ama.org/what-is-marketing-analytics/; Google Analytics Help, “About Attribution and Attribution Modeling,” https://support.google.com/analytics/answer/10596866; National Institute of Standards and Technology, “NIST/SEMATECH e-Handbook of Statistical Methods: Randomized Block Designs,” https://www.itl.nist.gov/div898/handbook/pri/section3/pri332.htm; Google, “Meridian: Marketing Mix Modeling,” https://developers.google.com/meridian/docs/intro; Google Analytics Help, “Attribution Models,” https://support.google.com/analytics/answer/10596866
