Written and reviewed by Bc. Petr Kučera, MSc., MBA
Performance marketing: what it is and how to evaluate it
Understand performance marketing, why reports differ, and how to connect conversions to sales: an overview of attribution, experiments, and measurement for Czech businesses.
performance-marketing
At PPC ONE, we use a working definition: performance marketing is marketing tied to measurable customer actions and business results, with campaigns adjusted continuously as those results come in. The IAB supports this framing, which is our working definition.
Consider two hypothetical situations: clicks keep rising, but nobody can connect them to orders; form submissions arrive, but it is unclear whether they turn into actual jobs or deals. In both cases, we recommend first asking which business outcome you need to record, and only then mapping advertising results to that outcome. That recommendation draws on IAB Europe and Google’s guidance on connecting lead outcomes to advertising.
Figures and platform settings in this article reflect the state as of November 10, 2026.
In our working definition, a campaign tracks a defined action, ties it to a business goal, and gets adjusted based on what happens. Measurement tracks results; an incrementality test estimates the ad’s causal contribution to purchases (IAB). Attribution assigns credit for recorded conversions. Incrementality asks which results happened only because of the ad (Gordon et al., Google Ads Help).
For a related explanation, read how PPC campaigns work.
Binet and Field describe long-term brand building and short-term activation as complementary (Binet and Field). Their data come from selected campaigns entered into IPA competitions, and the findings are observed associations (IPA).
We therefore recommend reading these frequently cited splits alongside their origin.
| Source | Brand to activation split | Context and scope |
|---|---|---|
| Secondary summary of earlier Binet and Field work (Vexpower) | Roughly 60:40 | Vexpower presents it as a rule of thumb |
| Effectiveness in Context, 2018 (Binet and Field) | 62% brand, 38% activation | Based on cases entered into the IPA Effectiveness Awards |
| Selected IPA cases for B2B, 2019 (LinkedIn B2B Institute) | Roughly 46% brand, 54% activation | Funded by the LinkedIn B2B Institute, which has an interest in marketing investment. The report gives no exact sample size, no uncertainty interval, and no Czech validation |
The IAB Europe framework separates media effectiveness, brand effectiveness, and sales effectiveness. Impressions, reach, and cost per click do not express revenue, profit, or relevant inquiries on their own. These supporting metrics can still help you look for the causes of changes. Pick the metrics according to what your company is trying to achieve.
The following order is our recommendation at PPC ONE. It draws on the IAB Europe framework.
Layer — What it answers
What the company actually records
Which customer or job the result belongs to
Which actions passed a business review
Which partial steps visitors took
How the ad itself performed in the report
Drawing on IAB Europe, we recommend separating financial and order records from reports used to optimize campaigns. The business view should cover paid orders, cancellations, and returns, plus customer lifetime value where repeat purchases matter. Depending on the company’s goals, we also recommend revenue, margin, new customers, customer acquisition cost, and overall marketing efficiency. Those choices extend beyond IAB Europe’s categories. Meta’s glossary provides related terminology.
Here, marketing efficiency ratio (MER) means total revenue divided by paid advertising spend. That is the working definition we use throughout this article, consistent with, for example, the variant used by Adtribute.
However, HubSpot and Shopify also use MER for a different ratio: total revenue divided by total marketing costs. Always state the formula, which revenue and costs it includes, and that both cover the same period. A change in the ratio describes the trend in overall efficiency; incrementality measurement estimates advertising’s causal contribution (Google Meridian).
Google offers enhanced conversions for leads. Customer data and imported events connect what happens to an inquiry outside the website back to the ad interaction (Google Ads Help). One example of an imported event is a lead converting into a customer (Google Ads Help).
When you pass a deeper-funnel outcome, Google Ads can work with an event other than the form submission itself (Google Ads Help, Google Ads Help). The number of form submissions alone does not determine lead quality.
According to an internal PPC ONE record, in September 2024 we worked on a service-sector client’s tracking of email clicks, phone clicks, form clicks, and “generate lead,” the event marking a new inquiry, as conversions. Our recommendation was to assign different values based on actual service orders and distinguish existing customers. This example illustrates the recommended way to separate outcomes with different business value.
Advertising cost per customer equals cost per click divided by the share of people who click and then become customers. That cost can rise even as clicks get cheaper if the customer conversion share falls by a greater relative amount. Related concepts appear in Meta’s glossary and the IAB Europe framework.
An attribution window defines how far back a report considers interactions before an outcome. Reports can differ in their windows, eligible interactions, inclusion of modeled conversions, and reporting dates. As Sklik, Seznam’s advertising platform, and Google Analytics explain, a discrepancy does not necessarily indicate a technical error, but it can be a reason to check the measurement implementation.
Several platforms can claim the same order. Add up Google Ads, Meta, and Sklik conversions, and you do not get the number of unique orders (Sklik Help, Shopify). Some specific systems do remove duplicates.
The checklist below is ours, derived from documented methodologies (Sklik Help, Google Analytics Help).
| What to check | Question to answer |
|---|---|
| Conversion definition | Do both reports count the same event? |
| Attribution | Which model and window does each report use? |
| Date | Is the conversion reported on the click date or the conversion date? |
| Modeling | Does the report include modeled conversions? |
| Deal identity | Is there a unique order or deal ID, or only platform credit? |
For these comparisons, we distinguish the product, its conversion tracking implementation, and the report’s time basis.
| Tool | What applies / What to watch for |
|---|---|
| Sklik | Modeled conversions since April 19, 2022 (Sklik, Seznam) A modeled conversion added to the report is not an extra real purchase caused by the ad |
| Sklik and Zboží.cz | Since May 6, 2024, a unified conversion code gives credit to the last source within this pair. Announced April 24, 2024 (Seznam) Deduplication does not cover Google Ads or Meta |
| Sklik, time-limited campaigns | Sklik recommends the “Konverze (k datu kliku)” metric, which reports conversions under the click date (Sklik) The day of the click is not the day of the order |
| New Sklik accounts | Must set up conversion tracking through Seznam Event Measurement (SEM), Seznam’s event tracking system (Sklik) Applies to new accounts |
| Shopping Ads | The documentation names SEM as the only supported system for new event tracking implementations (Shopping Ads) The scope differs from new Sklik accounts, and no date of the change is given |
| Heureka | 30 days after the click, or an optional shorter window of 7 days. Conversions are reported under the click date (Heureka) When comparing with purchase dates, align the date basis first |
| Original Sklik code | The documentation lists a 30-day cookie (Sklik) This cannot be carried over to SEM automatically |
Table source: Sklik.
An older technical account from Unikum describes the original Sklik system as using the last click without a choice of attribution model.
GA4 includes reports on traffic sources and key events; attribution settings affect only some reports. By default, it uses data-driven attribution for key events in reports with event-scoped traffic dimensions (Google Analytics Help, Google Analytics Help). You can change this setting.
The documentation lists data-driven, paid-and-organic last click, and Google-paid-channels last click (Google Analytics Help). The names show whether a model works from data or uses the last click within a set of channels. First click, linear, time decay, and position-based were removed in November 2023 (Google Analytics Help), so we skip a detailed explanation of them.
A few more differences matter when you compare reports.
| Setting or source | What it means for comparison |
|---|---|
| GA4 general key event lookback windows | For the acquisition events first_open and first_visit, the choices are 7 or 30 days, with 30 days as the default. Other events have 30-, 60-, or 90-day options, with 90 days as the default. These are attribution settings, rather than a guaranteed business cycle. See Google Analytics documentation. |
| Beta web conversion management | Google announced beta web conversion management for Google Ads customers on January 16, 2026. It supports individual attribution settings for each conversion and applies only to some GA4 properties. See Google’s update history. |
| Click windows within that management feature | From August 11, 2026, this feature supports click windows in whole days from 1 to 90 days. It is a distinct feature from the general GA4 lookback windows above. Consult the update history and attribution settings documentation. |
| Google Ads data-driven attribution (DDA) | Google recommends at least 200 conversions and 2,000 ad interactions in supported networks over 30 days. That is a recommendation, not an eligibility minimum: the documentation states that DDA is available regardless of data volume. |
| New Google Ads conversion actions | Unless a different setting has been selected, the default view-through conversion window for a new conversion action is 1 day. The account’s actual settings determine the applicable window. Check those settings against Google’s guidance. |
| Meta windows described by vendors | Technical vendor accounts describe the common 7-day click and 1-day view attribution setting. The campaign’s settings decide. See Initiators Media and Dataslayer. |
| Meta Ads Insights API | According to PPC Land, the removal of 7-day and 28-day view-through windows from Ads Insights API was announced with an effective date of January 12, 2026. This concerns the API used to retrieve reporting data. It does not establish a blanket change to every campaign. |
| Meta click definitions and engage-through | Meta’s March 3, 2026 announcement describes changes to click definitions and an expansion of engage-through, a category of counted interactions. Dataslayer’s technical account associates engage-through with a 1-day window. |
Table source: Google’s update history.
Table source: Dataslayer.
For more on Meta tracking, read Meta Pixel and Conversions API.
Drawing on Google and the methodological literature, we choose an approach according to the question, available data, and acceptable uncertainty. Business records establish recorded transactions; attribution assigns credit; a controlled experiment estimates incremental effects. Marketing mix modeling (MMM) examines marketing and other influences in aggregate data. See Google’s guidance on lead outcomes and experiments, Meridian’s methodology, and Gordon and coauthors.
| Question | Evidence layer | Measurement context |
|---|---|---|
| Which business transactions actually occurred? | Order records and CRM (customer relationship management) data. See Google. | These records show actual transactions; an experiment measures how many would have occurred without advertising. Gordon and coauthors discuss the causal distinction. |
| Which interactions receive credit? | Attribution distributes credit for recorded conversions. See Gordon and coauthors. | Assigned credit describes the system’s rules; causal effects are measured by an experiment, against which Gordon and coauthors compare observational methods. |
| What additional outcomes did advertising exposure produce? | A controlled experiment estimates incremental effects. See Google. | How conclusive the result is depends on the study’s design, duration, budget, and historical data. Lewis and Rao examine that uncertainty. |
| How do marketing and other influences relate to aggregate outcomes? | MMM models those influences using aggregate data. See Meridian. | Conclusions depend on the data, model specification, and assumptions. |
Table source: Gordon and coauthors.
An attribution report tracks how credit for recorded conversions is split between interactions. An incrementality test estimates how many additional deals advertising generated (Gordon et al.). A data-driven attribution model assigns conversions to interactions; estimating causal effects is a separate task (Google Analytics Help).
To read the research below correctly, you need to know its scope and units. ROI is the return on investment in advertising. DML stands for double machine learning, a method that estimates an effect from observational data without running an experiment.
| Study | What it found | Scope and context |
|---|---|---|
| Gordon et al., 2019 (Gordon et al., RePEc) | 15 US experiments on Facebook, 500 million user-experiment pair observations, roughly 1.6 billion ad impressions | The unit of observation is a user-experiment pair. Two of the four authors, Neha Bhargava and Dan Chapsky, were affiliated with Facebook Inc. at publication |
| DML, upper funnel (Kellogg) | Median absolute difference from the experimental lift of 115 percentage points | Published in 2023, based on 663 experiments on Facebook. The figure relates to the DML method in those 663 Facebook experiments and is expressed in percentage points |
| DML, mid funnel (Kellogg) | Median absolute difference of 107 percentage points | A 2023 publication based on 663 experiments on Facebook |
| DML, lower funnel (Kellogg) | Median absolute difference of 62 percentage points | A difference from the experiment within this research |
| Lewis and Rao, 2015 (Lewis and Rao, The Quarterly Journal of Economics) | 25 experiments by large US retailers and brokerage firms. In the retail experiments, the median standard error of the ROI estimate was 26.1 percentage points | This is the uncertainty of the estimate, not the ROI achieved, not a relative error, and not the universal precision of tests |
Table source: Kellogg.
A test group and a control group let you estimate the effect against having no exposure to that advertising. According to Google, how conclusive a test is depends on its design, duration, budget, and historical data, among other things (Google Ads Help). A result that is not statistically significant means the test did not detect the effect with sufficient precision; on its own, it does not determine whether advertising had zero effect (Lewis and Rao).
Branded search ad experiments outside the Czech market have produced different results.
| Experiment | Result | Context and scope of the result |
|---|---|---|
| eBay on Yahoo and MSN, summarized by Chicago Booth on February 26, 2015 (Chicago Booth) | After brand ads were switched off, roughly 99.5% of the traffic those ads would otherwise have brought arrived through unpaid results | The result applies to eBay; decide whether to switch off your own brand campaigns based on your own incrementality test |
| Edmunds.com, summarized by the authors for CEPR on February 18, 2018 (CEPR) | Unpaid results replaced roughly half of the traffic from the switched-off brand ads, and the other half was lost | According to the authors’ CEPR summary, the share refers to traffic from the switched-off Edmunds.com brand ads |
Google Conversion Lift is available depending on whether your account is eligible. Google Ads Help on user-based studies lists USD 5,000 and 1,000 observed conversions among the requirements (Google Ads Help). This is not an agency fee.
Google had already announced lower budget requirements for incrementality testing on May 21, 2025. Its November 11, 2025 announcement recaps the earlier change.
Meridian’s methodology uses aggregate data on marketing and other influences. Results depend on the data, model, and assumptions. Drawing also on the Meta Robyn methodological guide, we recommend comparing estimates with experiments and applying conclusions only to the market covered by the data.
Meridian’s methodology explains that models with similarly good fit can produce different ROI estimates. A causal interpretation requires consideration of other causes of the outcome and appropriate variable selection.
The overview below summarizes example tools and their limits.
| Tool | What the source says / Limit |
|---|---|
| Meridian and Robyn (Meridian, Robyn) | Google provides the open-source MMM tool Meridian. Meta provides Robyn |
| Meridian 2.0 (overview, changelog) | Google’s overview lists availability from September 3, 2026. The changelog dates release 2.0.0 to September 2, 2026 |
| Meridian data requirements (documentation) | Documentation updated October 28, 2026: as a guide, at least two years of weekly data for geo models and three years for national models. With monthly data only, at least three years According to Meridian’s documentation, data sufficiency also depends on variability in the data and the ability to separate influences |
| GeoLift (GeoLift) | According to Meta, an open-source tool for geo experiments using synthetic control Czech regions or cities need a specific study design |
| PyMC-Marketing (PyMC-Marketing) | According to its documentation, an open-source library for Bayesian MMM with calibration from lift tests |
Google describes geographic experiments with GeoX, automated experimental calibration, and a full-funnel framework (Google Meridian).
In MMM forum threads, contributors describe preparing clean, stable data as a practical obstacle in their experience with these tools. Other contributors report low statistical power in geographic tests and difficulty achieving convergence with PyMC-Marketing.
To connect advertising to business results, first define the business outcome and where it is recorded, then track the business and operational reporting layers side by side.
Drawing on IAB Europe and Google, we recommend naming the outcome first and identifying its system of record. For inquiries, distinguish a submitted form, a lead qualified by the sales team, and a closed deal.
Track both the business layer and the operational layer, and write down each report’s attribution rules (Shopify). Assess operational reports alongside your records of real deals (IAB Europe).
If you are setting up a new offline conversion integration, take Google’s changes into account. Since June 15, 2026, Google has restricted uploads of offline conversions through the Google Ads API based on each integration’s prior use, and it points new integrations to the Data Manager API (Google Ads Developer Blog). A developer overview updated October 30, 2026, describes blocking developer tokens that made no uploads between December 17, 2025, and June 15, 2026 (Google Developers). This does not block all older integrations across the board.
When you migrate to a new enhanced conversions for leads action, note that Google’s guides differ. One says three conversion cycles, another one to two. We take the conservative route: wait for the longer of three conversion cycles or four weeks. Only then should the new action become the primary conversion action and the original become the secondary conversion action.
This recommendation applies only to that migration. A conversion cycle is the usual time from an ad interaction to a conversion (Google Ads Help, UK version, another guide).
For B2B companies with long sales cycles, evaluating too early can understate results, according to Google Ads documentation. We recommend allowing for the time from advertising contact to a closed deal. See also the Google Analytics documentation.
Google Ads does not import offline conversions uploaded more than 90 days after the associated last click. For enhanced conversions for leads, the limit is 63 days. Both limits run from the last click to the upload, regardless of when the conversion occurred. These technical import deadlines are separate from how long a company retains its customer records.
If you cannot connect reported inquiries to business results or explain differing definitions, we consider that, based on these sources, a reason to seek expert help (Google Ads Help, Sklik Help).
For further orientation, read our article on questions about a Google Ads account.
Technical setup or modeling alone does not ensure legal compliance.
Consent Mode adjusts how Google tags behave based on the consent state passed to them (Google Developers). The feature itself does not prove legal compliance.
According to their documentation, Google and Sklik model some of the conversions they cannot observe (Google Ads Help, Sklik Help).
Figures published by the platforms themselves illustrate why additional reported conversions should be distinguished from additional sales.
Source: Seznam.
In the Czech Republic, since January 1, 2022, § 89(3) of Act No. 127/2005 Sb. requires prior, demonstrable consent to store information on a device or access information there. Exceptions cover technical operations necessary to transmit a communication or provide an explicitly requested service. Closing the cookie banner or continuing to browse does not, by itself, constitute consent (the Czech Office for Personal Data Protection (ÚOOÚ), ÚOOÚ cookie guidance). We rely on the version of the provision in force from November 1, 2026.
A server-side implementation does not replace a legal basis for processing personal data or consent where the cited provision requires it. Whether tracking is lawful cannot be determined merely from the fact that it runs on a server. The relevant Czech context remains § 89(3) of Act No. 127/2005 Sb. and ÚOOÚ’s cookie guidance.
According to the technical and legal sources, server-side tracking can reduce dependence on third-party cookies. However, server-side Sklik also uses cookies created after consent to match activity.
For related context, read about Consent Mode v2 and ad measurement and server-side conversion tracking.
Our services cover Google Ads, Facebook Ads, Instagram Ads, and Seznam Sklik. The PPC ONE service offering also includes server-side conversion tracking and cookies. On the same page, we use the motto “We turn clicks into business.”
According to our services page, we offer an independent audit by an experienced senior specialist, along with Google Ads and Sklik management and server-side conversion tracking. PPC ONE was founded by Petr Kučera.
According to an internal PPC ONE record, in October 2024 we launched Meta test campaigns for one client using the “1-day click” attribution setting, a one-day click-through window. The budget was set in advance, with evaluation planned after two weeks.
Attribution assigns credit for recorded conversions. Incrementality examines how many results would not have happened without advertising. Comparing a test group with a control group estimates the incremental effect (Google Ads Help). The link between advertising and a purchase alone shows only correlation (Gordon et al.).
No. The sum is not the number of unique orders, because one deal can collect credit from several platforms (Sklik Help). One exception is deduplication between Sklik and Zboží.cz with a unified conversion code (Seznam). That exception does not cover Google Ads or Meta.
They may use different windows, interactions, modeling, or reporting dates. A discrepancy does not necessarily mean a technical error, but it can warrant checking the implementation. See Sklik and Google Analytics for the methodological differences.
Customer data and imported events connect a result that happens outside the website with the ad interaction (Google Ads Help). Instead of a form submission, Google Ads can then work with, for example, a lead converting into a customer (Google Ads Help).
By default, GA4 uses data-driven attribution for key events in reports with event-scoped traffic dimensions (Google Analytics Help). You can change the setting. The attribution setting affects only some GA4 reports (Google Analytics Help).
Google Ads does not import offline conversions uploaded more than 90 days after the related last click. For enhanced conversions for leads, the limit is 63 days. The period runs from the last click to the upload, regardless of when the conversion happened (Google Ads Help).
Only with a unified conversion code. Since May 6, 2024, credit goes to the last source within the Sklik and Zboží.cz pair (Seznam). This is not deduplication against other systems.
No. A server-side implementation replaces neither a legal basis nor required consent. In the Czech Republic, § 89(3) of Act No. 127/2005 Sb. governs storing information on devices or accessing it. Exceptions cover technical operations necessary to transmit a communication or provide an explicitly requested service. ÚOOÚ’s guidance explains the context. Server-side Sklik also uses cookies created after consent for matching. Moving the implementation to a server does not itself change these requirements.
Have questions about how you evaluate your advertising? Send us an inquiry. Describe what you evaluate today and where you need clarity.
Contact us
About the author
Bc. Petr Kučera, MSc., MBA
AI Systems & Marketing Strategist, founder of PPC ONE
Petr discovered the power of performance campaigns back in 2011, and they have been an inseparable part of his professional life ever since. His edge comes from working with more than a hundred companies.
Data verified as of 10/11/2026.
80 client reviews