One View of Marketing

Executive summary

Most marketing teams do not lack data. They lack agreement. On a typical Monday, the paid team reports platform ROAS from three ad managers, the social team reports engagement rates calculated four different ways, the SEO lead reports clicks from Search Console that never match sessions in analytics, the email team reports open rates that a privacy feature quietly inflates, and the PR agency reports share of voice from a monitoring tool nobody else can log into. Every number is defensible on its own terms. Together they do not add up to an answer to the only question leadership asks: what is marketing producing for the money and time we put in?

This paper lays out a practical approach to unified marketing measurement for teams of five to fifty marketers. It is built on four ideas. First, a metric taxonomy that separates delivery, response and outcome metrics, so you never add Instagram views to LinkedIn impressions and call it reach. Second, normalization rules that translate each platform’s definitions into yours and document what cannot be translated. Third, a layered view of efficiency and causality: platform ROAS for steering inside a channel, MER for the business, and incrementality experiments plus media mix modeling for decisions that move budget between channels. Fourth, an operating cadence that turns one dashboard into weekly, monthly and quarterly decisions.

We also cover two forces that make this urgent. Privacy changes in browsers, inboxes and consent frameworks mean user-level tracking captures a shrinking and biased share of journeys⁹Source 9 · Google for Developers (Tag Platform)Consent mode overviewdevelopers.google.com¹⁰Source 10 · WebKitTracking Prevention in WebKitwebkit.org⁸Source 8 · AppleMail Privacy Protection & Privacyapple.com. And AI answer engines change what search visibility means: a Pew Research Center study of 900 U.S. adults’ browsing data found that Google users clicked a traditional result on 8% of visits to pages with an AI summary, versus 15% without one²Source 2 · Pew Research Center, 2025Google users are less likely to click on links when an AI summary appears in the resultspewresearch.org. Measuring only clicks now undercounts how often your brand is seen and cited.

Why do channel reports never add up?

Channel reports fragment for structural reasons, not because anyone is careless. Understanding the causes is the first step to fixing them, because each cause needs a different remedy.

1. Every platform defines the same word differently

Take engagement. LinkedIn’s organization share statistics API defines its engagement figure as organic clicks, likes, comments and shares divided by impressions⁵Source 5 · Microsoft Learn (LinkedIn Marketing API)Organization Share Statisticslearn.microsoft.com. Instagram’s media insights expose total_interactions as likes, saves, comments and shares minus unlikes, unsaves and deleted comments, and they do not include link clicks⁴Source 4 · Meta for DevelopersInstagram Media Insights (API reference)developers.facebook.com. Google Analytics 4 uses engagement rate for something else entirely: the share of sessions that lasted longer than 10 seconds, had a key event, or had at least two page or screen views⁶Source 6 · Google Analytics Help[GA4] Engagement rate and bounce ratesupport.google.com. Three tools, three different quantities, one label. A dashboard that averages them is not wrong in an obvious way; it is wrong in a way nobody will notice for months.

2. Platforms count their own conversions

Ad platforms attribute conversions using their own windows and models, and they see only their own touchpoints. Google Ads made data-driven attribution the default model and retired first click, linear, time decay and position-based models during 2023¹¹Source 11 · Google Ads Help, 2023First click, linear, time decay, and position-based attribution models are going awaysupport.google.com; Meta, LinkedIn and TikTok each apply their own click and view windows. When two platforms touched the same buyer, both can claim the sale. Summing platform-reported conversions across channels therefore routinely produces more conversions than the business recorded.

3. Different tools see different slices of the same visit

Search Console counts a click when someone clicks your result on Google. GA4 records a session only if the page loads, the tag fires and the visitor’s consent state allows it. Google’s own documentation notes that some tools, such as Google Analytics, only track users who have JavaScript enabled, and that Search Console reports daily data in Pacific time while Analytics uses the property’s time zone⁷Source 7 · Google Search Console HelpAbout Search Console datasupport.google.com. The two will never match exactly, and they should not be forced to.

4. Ownership is split by channel, not by question

Each team picks the tool, metric and cadence that flatter its work, usually in good faith. Social teams optimize for engagement because it moves fastest; paid teams optimize for platform ROAS because it is in the bidding algorithm; SEO teams report clicks because revenue attribution is hard. Nobody owns the cross-channel question, so nobody owns the definitions that would answer it.

A unified metric taxonomy

A taxonomy is a short, agreed list of metric families, each with one definition, one owner and an explicit rule for whether it can be summed across channels. We recommend three layers. The layering matters more than the exact metric names, because it prevents the most common mistake in cross-channel reporting: adding up things that are not the same kind of thing.

Layer 1: Delivery — did people see it?

Impressions or views count how many times content was displayed or played. Reach counts distinct people or accounts. Both are platform-specific and mostly not additive across platforms: the same person sees you on Instagram and LinkedIn, and no platform can de-duplicate the other. Report delivery per channel, and if you need a cross-channel figure, label it "gross impressions" and never call it reach.

Layer 2: Response — did they act?

Engagements (reactions, comments, shares, saves), clicks and sessions measure response. Clicks are platform-counted; sessions are counted by your analytics. Choose one canonical response metric per question: engagements for content resonance, clicks for traffic generation, sessions for what actually reached your site. The click-through rate should always state its denominator.

Layer 3: Outcome — did it matter to the business?

Key events (sign-ups, demo requests, add-to-carts), pipeline, orders and revenue are outcomes. Outcomes are the only layer you should sum across channels, and only when they come from a single system of record (your CRM, billing or shop backend), with channel credit assigned by one documented rule. Note the GA4 vocabulary here: events that matter to the business are now called key events, and a Google Ads conversion is created from a key event¹⁷Source 17 · Google Analytics Help[GA4] About key eventssupport.google.com.

How do you normalize metrics across platforms?

Normalization is the translation step between each platform’s raw export and your taxonomy. It has three parts: mapping fields to canonical names, recomputing derived metrics from components rather than importing the platform’s ratio, and documenting the gaps you cannot close.

Map fields, then recompute ratios yourself

Never import a platform’s engagement rate. Import its components (impressions, reactions, comments, shares, saves, clicks) and compute your own ratio with your own denominator. LinkedIn’s engagement figure includes clicks in its numerator⁵Source 5 · Microsoft Learn (LinkedIn Marketing API)Organization Share Statisticslearn.microsoft.com; Instagram’s total interactions exclude them⁴Source 4 · Meta for DevelopersInstagram Media Insights (API reference)developers.facebook.com. If your canonical engagement excludes clicks, you subtract LinkedIn clicks before dividing. That one rule makes a "LinkedIn vs Instagram" comparison meaningful.

Handle definitional drift as a data event

Platforms change definitions. Instagram’s API marks the impressions metric as deprecated for media created after July 2, 2024, and exposes views instead⁴Source 4 · Meta for DevelopersInstagram Media Insights (API reference)developers.facebook.com. Treat such a change like a schema migration: record the date, keep both series where possible, annotate the chart, and never splice the old and new metric into one line without a visible break.

Email opens are a special case

Apple’s Mail Privacy Protection downloads remote content, including tracking pixels, in the background by default, whether or not the recipient interacts with the message⁸Source 8 · AppleMail Privacy Protection & Privacyapple.com. For any list with a meaningful share of Apple Mail users, opens are no longer a reliable engagement signal. Keep them in the dataset for deliverability diagnostics, but use clicks and downstream key events as the email response and outcome metrics.

Earned media needs its own normalization

Press and social listening tools count mentions, estimated reach and sentiment using proprietary methods. Normalize to a small set you can defend: mentions (count), share of voice (your mentions ÷ mentions of you and a fixed competitor set, over the same sources and period) and, where the tool supports it, mentions with a link to your domain. Keep the competitor set and source list fixed for at least a quarter, otherwise share of voice moves because the denominator moved.

ROAS, MER and incrementality: which efficiency metric should you trust?

Use all three, for different decisions. ROAS tells you how a campaign performs according to the platform that ran it; it is a steering wheel inside one channel. MER tells you how all marketing performs against all revenue; it is the speedometer for the business. Incrementality tells you how much of the outcome would not have happened without the spend; it is the only one of the three that measures causation.

ROAS is useful because it is fast and granular, and because bidding algorithms optimize toward it. It is dangerous because it is self-reported, overlaps across platforms, and credits conversions that would have happened anyway (people who were already searching for your brand, for example). MER is robust because it cannot double count: the numerator comes from your books. Its weakness is that it cannot tell you which channel moved it. That is why the third layer exists.

What incrementality testing adds

Incrementality is measured with controlled experiments. Google describes Conversion Lift as splitting an audience into a treatment group that sees ads and a control group that does not, then measuring lift as the increase in conversions caused by the presence of the ad; studies can be user-based or geography-based and report incremental conversions and incremental ROAS¹²Source 12 · Google Ads HelpAbout Conversion Liftsupport.google.com. You can run the same logic yourself with geo holdouts: pause or reduce spend in comparable regions for a few weeks and compare against regions where spend continued.

Media mix modeling, without the mystique

Media mix modeling (MMM) estimates how much each channel contributes to an outcome by regressing that outcome, usually weekly revenue or conversions, on spend or exposure by channel, while controlling for seasonality, pricing, promotions and other drivers. Because it uses aggregate time series rather than user-level journeys, it is unaffected by cookie loss and consent gaps, which is why it has returned to prominence.

Two open-source projects lowered the barrier considerably. Google’s Meridian applies a Bayesian model that is "designed to estimate the true causal impact of your marketing", supports geo-level data, can incorporate reach and frequency, lets you encode prior knowledge as priors and is meant to be calibrated with geo experiments, and includes budget optimization and scenario planning¹³Source 13 · Google for DevelopersAbout Meridiandevelopers.google.com. Meta’s Robyn models the lagged effect of advertising with geometric or Weibull adstock, models diminishing returns with a Hill saturation function, decomposes trend and seasonality with Prophet, treats calibration against experiments as an objective in its multi-objective optimization, and ships a budget allocator¹⁴Source 14 · Meta Marketing Science (Robyn documentation)Robyn featuresfacebookexperimental.github.io.

The three concepts you need to read an MMM

• Adstock (carryover): advertising keeps working after the week it ran, decaying over time. A channel with long carryover looks worse in weekly ROAS than it really is.
• Saturation (diminishing returns): each additional unit of spend buys less outcome than the previous one. The response curve tells you where the next dollar does the most.
• Calibration: experiments anchor the model to causal ground truth. An uncalibrated MMM fits the past; a calibrated one is more likely to describe cause.

When MMM is worth it

MMM needs history and variation. As a practical rule, you want two or more years of weekly data, meaningful spend in several channels, and periods where spend changed independently across channels. If every channel rises and falls together, the model cannot separate them; that is where planned geo tests and deliberate budget variation earn their keep. For a team spending modestly in two channels, a disciplined MER view plus periodic holdouts usually delivers more decision value than an MMM.

What do consent and browser privacy do to attribution?

They make user-level attribution structurally incomplete, and incomplete in a biased way. Three mechanisms matter most for a marketing dashboard.

Consent choices change what tags can collect

Google’s consent mode lets tags adapt to four consent types: ad_storage, analytics_storage, ad_user_data and ad_personalization. In basic mode, Google tags stay blocked until the visitor responds to the banner, and nothing is sent if they decline. In advanced mode, tags load with consent denied by default and send cookieless pings that Google uses to model conversions, subject to data thresholds⁹Source 9 · Google for Developers (Tag Platform)Consent mode overviewdevelopers.google.com. Either way, a share of your conversions in Google Ads and GA4 becomes modeled rather than observed. That is fine, as long as your dashboard says so.

Browsers cap identifiers

WebKit’s Intelligent Tracking Prevention deletes cookies created in JavaScript after seven days without user interaction with the site, and caps them at 24 hours on landing pages when it detects link decoration used for cross-site tracking¹⁰Source 10 · WebKitTracking Prevention in WebKitwebkit.org. A buyer who clicks an ad in Safari and returns to purchase ten days later can look like a brand-new, direct visitor. Long consideration cycles, typical in B2B, are hit hardest.

Server-side signals help, but do not restore certainty

Server-side event APIs such as Meta’s Conversions API connect an advertiser’s own marketing data to the platform, with server events processed like pixel events¹⁵Source 15 · Meta for DevelopersConversions APIdevelopers.facebook.com. They improve match rates and resilience against browser restrictions. They do not make attribution causal, and they do not remove overlap between platforms.

How do you measure visibility in AI search (GEO)?

Generative engines such as Google’s AI Overviews and AI Mode, ChatGPT, Perplexity and Copilot answer questions directly and cite a handful of sources. Being cited is a new form of visibility, and it is not fully captured by clicks. The term generative engine optimization was introduced by Aggarwal et al., whose experiments found that content optimizations could increase visibility in generative engine responses by up to 40%¹Source 1 · arXiv (Aggarwal et al., KDD 2024), 2023GEO: Generative Engine Optimizationarxiv.org.

For Google specifically, traffic from AI Overviews and AI Mode is reported in Search Console’s Performance report under the Web search type, and Google states that no special optimization is required to appear there³Source 3 · Google Search CentralAI features and your websitedevelopers.google.com. That means classic Search Console reporting includes AI-feature traffic but does not isolate it. Meanwhile, Pew’s analysis of 68,879 searches by 900 U.S. adults found that 18% produced an AI summary, and that links inside those summaries were clicked on about 1% of visits²Source 2 · Pew Research Center, 2025Google users are less likely to click on links when an AI summary appears in the resultspewresearch.org. Visibility and clicks are decoupling.

A practical GEO measurement model

1. 1
   Step 1:

   Define a prompt set

   List 50 to 200 questions your buyers actually ask, grouped by topic and funnel stage. Source them from sales calls, support tickets, Search Console queries and community threads. Keep the set stable for a quarter.

2. 2
   Step 2:

   Sample answers on a schedule

   Run each prompt weekly across the engines that matter to your audience, in a clean session and a fixed locale. Answers vary run to run, so sample several times and record rates, not single outcomes.

3. 3
   Step 3:

   Score presence, citation and position

   For each answer, record whether your brand is mentioned, whether your domain is cited as a source, which competitors appear, and whether the description of your product is accurate.

4. 4
   Step 4:

   Connect to outcomes

   Track referral sessions from AI engines in analytics (they usually arrive with a recognizable referrer), and watch branded search impressions in Search Console as a lagging signal of awareness.

Search Console’s branded queries filter, rolled out from November 2025 to eligible top-level properties, splits Performance data into branded and non-branded views using an AI-assisted classifier rather than regular expressions¹⁶Source 16 · Google Search Central Blog, 2025Introducing the branded queries filter in Search Consoledevelopers.google.com. It is a useful companion signal: if AI answers are building awareness, branded demand should rise even as non-branded clicks flatten. For the full playbook on optimizing content for answer engines, read our post on generative engine optimization.

The one-dashboard blueprint

With a taxonomy, normalization rules and an efficiency framework in place, the dashboard itself becomes straightforward. The architecture has four stages: connect raw sources, normalize them into canonical models, define measures once in a semantic layer, and publish one overview with drill-downs.

The overview page: six blocks, top to bottom

1. 1Business outcome strip: revenue or pipeline, new customers, total spend, MER and blended CAC for the period, each versus the prior period and target.
2. 2Channel contribution table: one row per channel family with spend, sessions, key events, attributed revenue, and an evidence column (observed, modeled, experimental).
3. 3Paid efficiency: spend and CPA or ROAS by platform, with the latest lift-test result pinned next to each major channel.
4. 4Organic reach and resonance: delivery and normalized engagement for each social network, plus Search Console clicks and impressions split branded and non-branded.
5. 5Earned and AI visibility: mentions, share of voice versus a fixed competitor set, and AI share of voice for the prompt set.
6. 6Notes and annotations: launches, definition changes, tracking incidents and test windows, shown on every time-series chart.

Resist adding more. Each block answers one question and links to a channel-specific page for depth. The Marketing command center template implements this layout; the SEO funnel and social media overview templates provide the drill-downs. In the app, the command center shows the overview on simulated data, and the step-by-step build is in our guide to setting up a unified marketing dashboard.

The joining key that makes it work

Cross-channel reporting lives or dies on campaign naming. Spend lives in ad platforms; sessions and key events live in analytics; revenue lives in your CRM or shop. The only thing they reliably share is the campaign identifier you put in UTM parameters and platform campaign names. A governed taxonomy, enforced at link creation, is the cheapest data-quality investment in marketing. Our UTM taxonomy guide gives a complete convention.

Operating cadence: turning one view into decisions

A dashboard nobody reviews on a schedule becomes a museum. Tie each layer of the measurement stack to a meeting with a clear decision owner.

Two habits make the cadence stick. First, every weekly review ends with written decisions, logged next to the dashboard so the next review starts from them. Second, alerts replace polling: configure thresholds on spend pacing, a sudden drop in sessions from a channel, or a spike in "(not set)" campaigns, and let alerts bring problems to the owner instead of waiting for Monday.

Methodology and limits

This framework synthesizes public platform documentation, open-source modeling documentation and published research cited in the sources list, together with patterns we see when teams set up cross-channel reporting. Platform field names and definitions were checked against public documentation at the time of writing and change often; treat the normalization tables as a starting point to verify, not a specification.

• All charts labeled "Illustrative data" use simulated figures to demonstrate a pattern. They are not benchmarks.
• Third-party statistics are cited to their original publisher and should be read in their own context; for example, Pew’s figures describe a U.S. panel in March 2025.
• The GEO measurement model is a sampling method. AI answers vary by session, locale and time, so results are estimates with uncertainty and should be compared as trends.
• No measurement approach here removes the need for judgment. MMM and lift tests reduce uncertainty; they do not eliminate it.

Conclusion and checklist

One view of marketing is not a tool you buy; it is a set of agreements you encode. Agree on three metric layers and add up only outcomes. Translate each platform into your definitions and document the gaps. Steer channels with ROAS, judge the business with MER, and settle causal questions with experiments and, when you have the history, a calibrated media mix model. Accept that consent, browsers and inboxes will hide part of every journey, and say so on the dashboard. Add AI search visibility to the picture before it becomes the main way buyers meet your brand. Then review it on a fixed rhythm and write the decisions down.