Which measurement and ROI platforms report honestly

Every measurement and ROI platform is honest about something and quiet about something else. The quiet part usually decides whether the campaign made money. The ten below are real products, and each one carries an interest in the answer.

What to do with the numbers once you have them is in measurement and ROI.

What to take away

The platforms worth knowing

These ten are sorted by what they actually measure. None is neutral, and none substitutes for the others.

Google Analytics 4 is free and runs on properties you own. It reports sessions, conversions and channel attribution, and it is the first place most brands look. It replaced Universal Analytics, whose standard properties stopped processing hits on July 1, 2023.

Its blind spot is everything before the click: a podcast mention, a creator's spoken code, a billboard. Consent choices also remove a slice of measurement, and not a random slice.

Northbeam is a paid multi-touch attribution platform built for ecommerce and DTC brands. It stitches ad spend, platform data and site behavior into one model, then reports channel-level contribution. Pricing is quote-based.

It suits teams spending enough on paid media to justify the fee. Its blind spot is the model itself: it fits the data it was given.

Triple Whale is paid ecommerce analytics, common among Shopify merchants. It pulls platform spend and store revenue into one dashboard, and it is cheaper and simpler than Northbeam. Pricing is tiered by store revenue and published on its site.

It suits smaller DTC teams. It reports what the platforms tell it, so platform over-attribution flows straight through.

Measured runs incrementality tests. It holds spend back from a channel to see what actually changes, which suits brands whose media budget makes a holdout affordable. Pricing is quote-based.

Its blind spot is scale: a clean test on one channel says less about the whole mix.

Nielsen sells panel and survey research, including brand lift and cross-platform reach. The panel is where its reach numbers come from. It suits brands that need to know whether people remember a campaign, rather than whether they clicked. Pricing is quote-based.

Its blind spot is the sample: panels measure what people are willing to say.

Kantar runs brand tracking and campaign evaluation across many markets. It suits brands with a long measurement horizon and a budget to match. Pricing is quote-based.

Its method is proprietary, and its respondents are not your buyers.

AppsFlyer and Adjust are mobile measurement partners. They attribute installs and in-app events, and they are the standard for app-based campaigns. Both are quote-based at scale, and both let you set the attribution window, which changes the install count without changing the campaign.

Their blind spot is everything outside the app.

HypeAuditor and CreatorIQ sit on the creator side. HypeAuditor scores audience quality and flags suspicious follower patterns. CreatorIQ manages creator relationships and campaign reporting. Both suit brands running influencer programs at volume, and both are quote-based.

Their scores are their own. No one outside the vendor can verify them.

The four sources underneath

These do not fail in the same direction. Platform data over-attributes to the platform. Click-based instrumentation under-attributes to anything that creates demand rather than capturing it. Two flavors of one source do not help at all.

Where native analytics are strong

For diagnostics, nothing else works as well. Whether an audience is real, where it is, when it watches, how far viewers get, and whether comments look like people: none of that is visible anywhere else.

That is why analytics viewed live on the creator's own account is the standard check, not a screenshot. The screening question belongs with fraud and brand safety.

Where native data is weak is as evidence of effect. The definitions belong to the operator, and they change. A count of views is not a claim about anything except views.

Where your own tracking breaks

Link-based tracking is the most common source of self-inflicted measurement error, and the failures are boring and specific.

Parameters get stripped by apps that open links in their own browsers. Codes get typed wrong, shared, and used by people who were never exposed to the campaign. Redirects lose the parameters. Three people naming campaigns three ways produce a report that cannot be aggregated.

The tagging convention in general use is described under UTM parameters, and the wider practice under web analytics. Neither is complicated. Both reward a written convention and a test before the campaign rather than a diagnosis afterward.

Where modeled analysis earns its place

Across a long history and many channels, a model can estimate contributions no click can show. That matters for channels that create demand rather than capture it.

A model fits the data it was given, and it will fit noise happily if allowed to, which is the ordinary problem of overfitting. A model also cannot validate itself. The only real check is an experiment the model did not see.

Treat a modeled result as a hypothesis worth testing rather than as a finding.

Choosing, in practice

Start from the decision, not the tooling.

If the question is whether an audience is real, the answer is native analytics viewed live. No dashboard replaces it.

If the question is whether the campaign moved sales, the answer is an experiment. No combination of reports substitutes for one.

If the question is whether people remember it, the answer is a survey, with all the caveats a survey carries.

Three checks apply to any source before you rely on it.

The third one catches most people. What has to be recorded against the campaign anyway is set out in campaign briefs, and the terms under which a creator shares their own analytics belong in contracts and disclosure.

Bottom line

Name the source before you argue about the number. Native analytics for diagnostics, your own instrumentation for behavior you can see, surveys for attitudes, models as hypotheses. For anything causal, run an experiment.

Common questions

Is one source ever enough?

For a small campaign whose purpose is not measurement, yes, provided you say what it can and cannot show. For a budget decision, no.

How do you handle disagreement between two sources?

Ask what each one is blind to. The disagreement usually sits exactly there, and the reconciliation is a description rather than an average.

Should creators be required to share analytics?

Ask for it, in the agreement, with a stated scope and duration. An open-ended right to a partner's data is a term worth being reasonable about.

Does more instrumentation produce better answers?

Only up to the point where the convention stops being followed. Beyond that it produces more rows and less clarity.