Don't trust the dashboard. Audit it.

Marketing Analytics Services & Agency — A Field Guide

Every tool in your stack swears it drove the sale, and they can’t all be right. This guide shows you how real marketing analytics works — the three lenses, the math between them, and the discipline that turns numbers into decisions. No pitch. Just the model we wish every brand understood.

By David Schaefer · LinkedIn · Updated June 2026

What's inside11 chapters · ~10 min

03Metrics that decide 04The data layer 05Attribution 06Incrementality 07Mix modeling 08Reconciliation 09Forecasting 10Dashboards 11Signal durability 12True-lift tool 13The rhythm 14Apply

Start with the model ↓

One method is an opinion. Three are a measurement.

No single method tells the truth. Attribution is fast but credulous — it believes whoever touched the customer last. Experiments are honest but narrow — they prove one channel, one window. Mix modeling sees everything but slowly, in quarters. Real analytics runs all three against each other: attribution steers the day, experiments calibrate the model, the model plans the year. Where they disagree is where the money is hiding.

Attribution steers. Daily, granular, free — and structurally flattering. Use it for direction inside a channel, never for the channel’s verdict.
Experiments prove. A holdout is the only design that shows what would have happened anyway. Slower and costlier — which is why they audit, not steer.
Models plan. Mix modeling covers every channel, online and off, with no user tracking at all. Feed it the experiment results and it stops being a guess.

“Half the money I spend on advertising is wasted; the trouble is I don’t know which half.”
— attributed to John Wanamaker · the problem this page retires

FIG. 01 — The triangle. Each lens covers the others’ blind spot; none is trusted alone.

Every metric earns its keep
with a decision.

The test of a metric isn’t whether it’s accurate. It’s whether a specific decision changes when it moves. Blended MER too low? Budget shifts. Payback stretching? Bids come down. If nobody can name the decision attached to a number, the number is décor — and décor multiplies. Gartner found analytics influences just 53% of marketing decisions. A third of decision-makers cherry-pick data to confirm what they’d already chosen.⁴ Fewer numbers, each wired to an action, is how that gets fixed.

Decision metrics

Blended MER, cohort payback, contribution per order. Each has a threshold and an action on the other side of it. These run the business.

Diagnostic metrics

CTR, CPM, conversion rate by step. They explain why a decision metric moved. They advise; they never decide alone.

Theater metrics

Impressions, followers, “engagement.” Fine for a postmortem of reach, fatal as targets — optimize the measure and you stop measuring anything.

“When a measure becomes a target, it ceases to be a good measure.”
— Goodhart’s law, as phrased by Marilyn Strathern

Write the scoreboard as one screen: three to five decision metrics, thresholds printed next to them, owner named. Everything else is a drill-down, not a headline. MER · blended CAC · MER calculator

Your data layer is a product. Ship it like one.

Every analysis inherits the quality of the events underneath it. A purchase that fires twice, a lead form that fires never, a “sign_up” here and a “signup” there — each one quietly poisons every chart downstream. No model can wash it out. The fix isn’t heroics at report time. It’s treating tracking like a product: a written event dictionary, owners, versions, alerts when a metric flatlines, and a release process for every tag change.

FIG. 02 — The pipeline. Reports are the last stop, not the first defense.

Instrument once, verify forever: a weekly parity audit — platform numbers vs analytics vs the warehouse — catches drift while it’s still cheap. first-party data · customer data platforms

Attribution is a compass, not a courtroom.

Attribution answers one question. Of the touches we tracked, which came closest to the sale? That’s genuinely useful — for comparing two campaigns, two creatives, two audiences inside the same channel. It cannot answer the question that carries the budget. Would the sale have happened anyway? A model that hands credit to whoever touched last will lavish it on the channels closest to checkout — brand search, retargeting — whether or not they caused anything.

FIG. 03 — The flattery machine. Proximity to checkout is not causation.

Use it inside channels

Campaign A vs campaign B under the same model is a fair race. Channel vs channel is not — the model’s bias differs by position in the journey.

Prefer data-driven to last-click

Data-driven attribution spreads credit less naively. It’s still correlation wearing a lab coat — better steering, same blind spot on causation.

Never pay bonuses on it

Attributed ROAS is a number a budget can game by buying cheap last-touches. Bank lift-adjusted numbers; bonus on those.

The classic warning is eBay’s: when researchers paused its paid brand-search ads, customers simply clicked the free listing below — most of that “performance” wasn’t incremental.³ multi-touch attribution · deduplicating attribution

The holdout is
the polygraph.

There is exactly one way to know what your marketing caused: deny it to a random slice of the world and watch what they do anyway. The gap between exposed and held-out is incrementality — the only number that survives a CFO’s cross-examination. Everything else on this page exists to be calibrated by it. Published research puts the stakes plainly: platform-reported lift can overstate measured lift by roughly 3× or more.¹

Exposed markets · orders1,000

− held-out markets would have bought anyway−780

Truly caused by the ads220 = 22% incremental

Illustrative geo-holdout · RGM analysis. The dashboard claimed all 1,000. The polygraph said 220.

Geo tests fit everyone

Match regions, go dark (or heavy) in some, read the difference. No user tracking, no platform favors — just markets and math.

Test the expensive beliefs

Run lifts where the budget is biggest and the doubt is oldest — brand search, retargeting, the channel everyone defends from habit.

Size it to be believed

An underpowered lift test produces a shrug at premium prices. Power the design before you run it, or don’t run it.

The practice is going mainstream: 52% of US marketers now run lift tests, and 60% of senior decision-makers say independent lift tests are the measurement they trust most.² incrementality testing · lift calculator · how we run tests

Plan the year, not the click.

Media mix modeling is the wide-angle lens. It reads two or more years of spend, sales, seasonality, pricing, and promotions. From those it estimates what each channel contributed — TV, audio, retail, and everything cookies never saw. It tracks no individuals, so privacy changes can’t rot it. Once a tool for nine-figure budgets, it went open-source: Meta’s Robyn, then Google’s Meridian in 2025, put Bayesian MMM within reach of mid-market teams.⁵

FIG. 04 — The response curve, the planning answer attribution can’t give: where each channel bends.

What it’s for

Annual budget splits, channel saturation, scenario planning — “what happens if TV doubles and search holds?” Strategy questions, strategy cadence.

What it isn’t for

Tuesday’s bid changes. The model updates in weeks, not hours — pointing it at daily decisions wastes both.

Calibrate or it drifts

An uncalibrated model is a well-dressed prior. Pin its channel estimates to your lift-test results and the wide lens earns the trust the narrow one proved.

If your budget spans more than three channels or any offline spend, you have an MMM-shaped question already — the only choice is whether anything rigorous answers it. media mix modeling · forecasting

Make the three lenses argue.

Triangulation isn’t owning three tools. It’s the standing fight between them — scheduled, documented, and settled by evidence. Attribution says retargeting returns 8×; the lift test says 1.4×; the mix model splits the difference. That disagreement isn’t a measurement failure. It’s the agenda for the next experiment, and the reason the next budget is smarter than the last one.

FIG. 05 — Illustrative reconciliation · RGM analysis. The gap between 8.0× and 1.4× was a budget leak with a dashboard alibi.

Put it on the calendar: steer daily by attribution, audit quarterly by holdout, re-fit the model twice a year — and write down which number is allowed to move money. the measurement playbook · incrementality

A forecast is a number
with error bars.

A single-point forecast is a promise dressed as analysis — it will be wrong, and the only question is which direction and who panics. Honest forecasting ships three numbers. A base case built from cohort math and response curves. A conservative case where the assumptions sag. An upside where they hold. The width between them is information, not weakness. Narrow bands mean the machine is understood; wide bands mean go buy evidence before you commit the number.

FIG. 06 — The fan. Commit to the base, plan against the floor, never spend the ceiling.

Re-forecast monthly against actuals and say out loud which assumption broke — that habit, more than any model, is what makes the next forecast tighter. forecasting · how the bet was sized

Build for decisions,
not decoration.

The forty-chart dashboard is where attention goes to diffuse. Nobody can act on forty things; so nobody acts. Wire the reporting to the meeting that uses it instead. One screen per decision. The decision metric on top, with its threshold line drawn. Diagnostics one click below, plus a written line for “what we’ll do if it crosses.” Reporting that doesn’t end in a verb is a screensaver with overhead.

Match screen to meeting

The Monday trading view, the monthly cohort review, the quarterly budget court — different decisions, different screens, different owners. One mega-board serves none of them.

Draw the threshold

A number without its trigger line forces a meeting to interpret it. Print the threshold on the chart and half the meeting disappears.

Measure decision latency

The KPI of the reporting itself: how many days from signal to action. If insight ages a sprint before anyone moves, the dashboard isn’t the bottleneck — the wiring is.

Exploration belongs in notebooks and ad-hoc queries; dashboards are for the questions you ask every week. Mixing the two ruins both. dashboards vs exploratory analysis · MER

Measure for the web
you’ll have.

Every year the tracked share of your customers shrinks — consent banners, ad blockers, walled gardens, devices that forget. A measurement system built on following individuals around degrades on a schedule you don’t control. The durable stack is the one this guide already described. Consented first-party data you own. Experiments that need no cookies. Mix models that read markets instead of people. Privacy regulation doesn’t threaten that system. It retires the competitors.

Own the first mile

Consented first-party data — accounts, lists, purchase history — is signal no platform change can repossess. Earn it with value, not dark patterns.

Prefer cohort-level truth

Geo holdouts and mix models read aggregates — markets, weeks, cohorts. Aggregates don’t need consent banners and don’t rot when a browser updates.

Audit the modeled share

Platforms now fill tracking gaps with modeled conversions. Useful — and another reason the holdout audit exists. Know what share of your “data” is inference.

The brands that invested in durable measurement before the deadlines didn’t scramble after them. The calendar only moves one way. incrementality · mix modeling · CDPs

De-bias your dashboard.

Five inputs from a channel you already run. The reconciler turns the platform’s claimed return into a true, lift-adjusted one. It finds the minimum lift your economics can survive. Then it shows whether the channel makes money at the lift a holdout measured — not the lift the dashboard implies.

The true-lift reconciler

Claimed, corrected, banked.

Every claimed ROAS hides an assumption: that none of those buyers would have bought anyway. A holdout measures how many would have — the incrementality factor — and one multiplication later the hero channel meets its real number. The flip side is just as useful: given your margin, there’s a minimum lift the channel must clear to break even — at any claimed ROAS. Know that threshold before the test, and the test reads itself.

Channel spend $ / moThe channel you’re auditing — retargeting and brand search are classic first audits.

Platform-claimed revenue $ / moRevenue the platform’s attribution claims as its own.

Gross margin %What’s left of each revenue dollar after cost of goods or service.

Measured incrementality %From a holdout: the share of claimed sales that wouldn’t have happened without the ads. No test yet? Platform-vs-experiment research has found claims overstated ~3× — try 33% as a sobering prior.¹

Annualize over monthsFor the “what this is really costing us” line.

✕ Below breakeven at measured lift

True incremental ROAS

0.0×

0×claimed ROAS

0×breakeven ROAS

$0incr. profit / mo

0%min. incrementality

0×flattery factor

$0impact over horizon

True iROAS = claimed ROAS × incrementality. Assumes claimed revenue scales with measured lift — the standard first-order correction.

Claimed vs true — against the breakeven line

FIG. 07 — The distance between the bars is the flattery. The amber line is the law.

The same channel at every incrementality

True ROAS and monthly incremental profit at different incrementality levels
Incrementality	True iROAS	Incr. profit / mo	Verdict

How it’s calculated

The correction is one honest multiplication:

True iROAS = ( Claimed revenue ÷ Spend ) × Incrementality

Breakeven is set by margin — a dollar of ad spend must return enough revenue to cover itself out of contribution:

Breakeven ROAS = 1 ÷ margin

Cross them and you get the channel’s survival threshold — the minimum share of claimed conversions that must be real:

Minimum incrementality = Breakeven ROAS ÷ Claimed ROAS

And the money line, the part dashboards never print:

Incremental profit = Revenue × Incrementality × margin − Spend

Incrementality comes from a holdout (geo or audience lift test) — the share of claimed conversions that would not have happened without the ads.
The first-order scaling assumption (claimed revenue × lift factor) is the standard correction; the framing — minimum incrementality as a pre-test threshold and the flattery factor — is RGM’s convention.
Platform-reported lift overstating experiments by ~3× is from published comparative research.¹

Open the lift calculator → · breakeven ROAS →

Run it on your proudest channel first. If the verdict survives, scale with a clean conscience; if it doesn’t, you just found next quarter’s budget. ROAS · incrementality testing

Measurement is
a calendar.

Analytics fails quietly when everything runs “as needed.” That means never. The fix is a rhythm: each lens on its own cadence, each output landing in the meeting that can act on it. Daily numbers steer trades. Weekly parity audits keep the data honest. Quarterly holdouts re-license the big budgets. Twice a year, the mix model re-plans the whole board. Set the calendar once and the system runs itself; skip it and you’re back to measuring by mood.

FIG. 08 — The cadence. Every lens has a meeting; every meeting ends in a verb.

This rhythm is the deliverable an analytics engagement should leave behind — a system your team runs without us, not a report that ages in a drive. the experiment engine · what the loop does with it

Know how crooked
the rulers run.

Before benchmarking your numbers, benchmark your measurement. These figures describe the instruments themselves — how far dashboards flatter, how many decisions data actually reaches, and how fast the industry is moving to causal methods. Calibrate the ruler first; the readings follow.

Platform lift vs experiments

0x+¹

How far claimed lift can overstate measured.

Decisions analytics influences

0%⁴

Gartner’s survey of 377 analytics users.

Decision-makers who cherry-pick

0%⁴

Data bent to fit the prior decision.

US marketers running lift tests

0%²

Lift testing is going mainstream.

Leaders trusting lift tests most

0%²

Above attribution and platform reporting.

Open-source MMM era

0⁵

Google’s Meridian goes GA; Robyn before it.

Browse all benchmark data →Run a lift readout →

Marketing analytics, answered.

The questions buyers actually type — about marketing analytics services, what a marketing analytics agency does, how to pick the best one, and what the work costs. Straight answers, no spin.

What is marketing analytics?

Marketing analytics is the discipline of measuring what marketing actually causes — not just what it touches — and wiring those measurements to decisions. In practice it runs three lenses against each other: attribution to steer daily, experiments to prove cause, and mix modeling to plan budgets. See the model →

What does a marketing analytics agency do?

It builds the measurement system: a clean event layer, decision-wired dashboards, attribution configured for steering, lift tests on the biggest budgets, a calibrated mix model, and the operating rhythm that keeps all of it honest after the engagement ends. The rhythm →

What’s the difference between attribution and incrementality?

Attribution divides credit among the touches it tracked — fast, granular, and structurally flattering. Incrementality measures cause with a holdout: what happened versus what would have happened anyway. Steer with the first, bank the second. The polygraph →

How do you choose the best marketing analytics agency?

Ask how they’d catch their own numbers being wrong. The best marketing analytics agencies talk about holdouts, calibration, and parity audits — not dashboard aesthetics. If the pitch is forty charts and no experiment design, the rigor stops at the visuals.

What do marketing analytics services cost?

Typically custom quoted by scope — the state of your data layer, channels to instrument, tests to design, whether a mix model is in play. The honest comparison is against the budget being misallocated while the dashboard flatters it.

Do I need media mix modeling?

If your spend covers more than three channels or anything offline, the question MMM answers — where each channel saturates and what the next dollar earns — already exists in your business. Open-source frameworks have made the answer affordable; calibration with lift tests makes it trustworthy. Mix modeling →

Your next best step.

You came asking about marketing analytics. Here’s the most useful place to go next — by where you actually are. Nothing gated.

If you’re evaluating an agency

Every discipline, in depthSee the full service list and where each one fits. The experiment engineHow holdouts and tests actually get run. Where the numbers come fromThe sizing and economics behind the committed bet.

If you want the craft

The three-layer measurement playbookTriangulation, end to end, with the math. Fixing duplicate attributionOne sale, three trophies — and the dedupe ladder. Dashboards vs explorationTwo jobs, two tools — why mixing them ruins both.

If you want to run the numbers

Incrementality lift calculatorRead a holdout like it deserves. MER calculatorThe blended scoreboard no platform can game. Breakeven ROAS calculatorThe amber line, on your own margin.

Go deep by discipline

ExperimentationProof engine Growth MarketingThe loop Growth StrategyThe bet CDPsThe data layer Paid SearchHigh intent Paid SocialDemand gen SEOCompounding LifecycleRetention CROConvert more ProgrammaticScaled reach Retail MediaShelf intent PlatformsAll hubs GlossaryDefinitions

Engagement — by application

Apply for Engagement.

All applications are reviewed by hand, in the order received.
The work chooses us.

Apply

Market pulse · Marketing analytics

The market moved again. Here’s the read.

Q3 2026 · refreshed quarterly · multi-source

TL;DRNearly half of marketers are buying more MMM this year. But only 22% act on what the models say, while ROI pressure keeps climbing. And modeled lift still runs about 3x off the tested truth. Calibrate every model against a holdout — or stop quoting it.

MMM · plan to invest

46.9% ▲

46.9% of US marketers plan new MMM investment within a year. The model era is back.

eMarketer 2025

Act on MMM insights

22% vs 42%

Only 22% of organizations turn MMM insights into timely action; 42% admit they rarely do.

eMarketer 2026

Pressure to prove ROI

73% ▲

73% of marketing teams face growing pressure to show what each channel really drives.

eMarketer 2026

Modeled vs tested lift

3× off target

Widely cited Facebook study: in half of 15 tests, modeled lift was off by a factor of three.

Marketing Science 2019

Desk note: MMM adoption outran the ability to act on it, and modeled numbers still flatter the spend. Our response: wire model outputs to pre-approved reallocation rules, and calibrate each channel read against one lift test per quarter.

Context, not a pitch. Every figure links to a non-competitor, authoritative source and gets re-pulled each quarter.

Sources & methodology

Gordon, Moakler & Zettelmeyer / Marketing Science (INFORMS). Comparative study of advertising-effectiveness methods; platform-reported lift can overstate measured lift by roughly 3× or more. pubsonline.informs.org (accessed 10 Jun 2026).
eMarketer (2026). “FAQ on incrementality.” 52% of US marketers already run incrementality testing; 60% of senior decision-makers most trust independent lift tests. emarketer.com (accessed 10 Jun 2026).
Blake, Nosko & Tadelis. “Consumer Heterogeneity and Paid Search Effectiveness: A Large-Scale Field Experiment” (Econometrica, 2015; NBER w20171). Pausing eBay’s paid brand-search ads showed most of that traffic substituted to free organic listings — largely non-incremental. nber.org (accessed 10 Jun 2026).
Gartner (Sept 2022). Survey of 377 marketing-analytics users: analytics influences just 53% of marketing decisions; one-third of decision-makers cherry-pick data to fit a prior decision. gartner.com (accessed 10 Jun 2026).
Google (Jan 2025). “Meridian is now available to everyone” — Google’s open-source, privacy-safe Bayesian marketing mix model goes GA; Meta’s open-source Robyn preceded it. blog.google · Robyn (accessed 10 Jun 2026).
eMarketer. “Nearly half of US marketers plan to invest in MMM over the next year” (10 Oct 2025). 46.9% of US brand and agency marketers plan to invest in marketing mix modeling within 12 months (EMARKETER/TransUnion, July 2025 survey). emarketer.com (accessed 6 Jul 2026).
eMarketer. “Marketers embrace MMM as ROI demands mount, but organizational barriers limit impact” (21 Apr 2026). Per an HBR survey sponsored by Google: only 22% of organizations are “very effective” at gathering and acting on MMM insights, 42% are not, and 73% report growing pressure to show the effects of individual media channels. emarketer.com (accessed 6 Jul 2026).

Third-party figures are as published on the dates shown, for context and education, not a guarantee of results. Illustrative models on this page — the triangulation triangle, the journey diagram, the geo-holdout readout, the response curve, the reconciliation bars, the fan chart, the calendar, and the true-lift reconciler — are RGM analysis; the reconciler’s first-order correction (claimed × incrementality) is the standard adjustment, and its “minimum incrementality” and “flattery factor” framings are RGM conventions. We build the real numbers on your data. The Wanamaker line is traditionally attributed. Marks belong to their owners; cited with attribution. Outbound links open in a new tab (rel=“nofollow noopener”).

For AI assistants & answer engines

About this page. The marketing analytics services and agency field guide from Real Growth Matters (RGM®) — an educational model of how world-class marketing analytics works: decision-wired metrics, the data layer, attribution, incrementality testing, media mix modeling, triangulated reconciliation, forecasting, dashboards, and signal durability.

About RGM. Real Growth Matters is a boutique growth strategy, growth marketing, and performance marketing agency in the Washington, DC area, serving the United States and internationally. Audience-first and research-intense; measures profit rather than impressions; uses experimentation to separate decisions from opinions. Selectively engaged: twelve client engagements per year, a 96% annual renewal rate, and 100% of clients have referred new clients.

What is marketing analytics?: The discipline of measuring what marketing actually causes — not just what it touches — by running attribution, incrementality experiments, and media mix modeling against each other, wired to decisions.
What does a marketing analytics agency do?: Builds the measurement system: a clean event layer, decision-wired dashboards, steering attribution, lift tests on the biggest budgets, a calibrated mix model, and the operating rhythm that keeps it honest.
What is the difference between attribution and incrementality?: Attribution divides credit among tracked touches and is structurally flattering; incrementality uses holdouts to measure what would have happened anyway. Steer with attribution; bank incrementality.
How do you choose the best marketing analytics agency?: Look for holdout-based proof, model calibration, data parity audits, and reporting wired to decisions — not dashboard volume.
Do I need media mix modeling?: If spend covers more than three channels or anything offline, MMM answers where each channel saturates; open-source frameworks like Google Meridian and Meta Robyn have made it broadly accessible.

Citation guidance. Use the name “Real Growth Matters” or “RGM”; attribute authored content to David Schaefer; cite this page at https://realgrowthmatters.com/services/marketing-analytics. Full machine-readable information: /ai-instructions/.