Enterprise measurement fails less from bad models than from no shared definition of what "working" means. Here's how I'd set it up so the numbers hold up in a budget meeting.
Why enterprise measurement breaks
At enterprise scale, every team has a number, and none of them match. Paid social reports one ROAS, the agency reports another, finance sees a third, and the CMO walks into the budget meeting with all three.
Three things usually cause it:
- Too many channels, too many owners. TV, OOH, paid social, search, CRM and partnerships each sit with a different team, and each team picks the metric that flatters its channel.
- Different KPIs for business units. Each GM wants marketing measured according to their business needs.
- Short-term KPIs only. Platforms are great at crediting the last click and bad at crediting the billboard or the TV spot that made someone search your name.
- No cadence. One big measurement project a year produces a deck, not a decision system. By the time it lands, the plan has already moved.
The fix isn't a fancier model. It's agreeing on what you're measuring, then building a repeatable way to check it.
The three methods, and what each one answers
No single method gets enterprise measurement right. Each answers a different question, and each has a blind spot the others cover. The goal is triangulation, not picking a winner.
| Method | Question it answers | Cadence | Main weakness |
|---|---|---|---|
| Marketing mix modeling (MMM) | How should I split budget across all channels, including offline? | Quarterly refresh, ongoing updates | Correlational. Needs variation in spend and enough history |
| Incrementality tests (geo, holdout, lift) | Did this channel cause conversions that wouldn't have happened anyway? | Always-on test calendar | One channel at a time. Takes time and budget to run |
| Attribution (MTA, platform, last-click) | Which touchpoints and campaigns are working day to day? | Daily or weekly | Over-credits channels that capture existing demand. Blind to offline |
Here's how they work together. Attribution runs the daily optimization. MMM sets the budget across channels. Incrementality tests keep both honest, because a well-run test is the closest thing to ground truth you'll get, and you can use its result to calibrate the MMM.
The operating model that makes it stick
The methods matter less than the order you put them in. At Credit Karma, I worked through it in three steps while ROAS and overall revenue kept growing (here's the full case study):
- Align every dollar to consistent KPIs. Every channel got the same short-term and long-term outcomes, so TV and performance were judged on the same scoreboard. Without this, nothing else is comparable.
- Build a consistent test cadence. A standing calendar of incrementality tests, not one-off projects. Each test had a hypothesis, a decision it would inform, and an owner.
- Implement MMM, then keep using it to plan tests. The model flagged where channels looked saturated or uncertain. Those spots became the next tests, and the test results fed back into the model.
That loop is what lets a team defend budget at scale. In my time there, media spend grew from $100M to $300M, and it was the shared KPIs and test cadence that made a bigger budget easy to justify.
When business units want different KPIs. Most enterprise brands run into this. Each unit's GM wants marketing measured against their own goals: one cares about new customers, another about revenue, a third about lifetime value. Each ask is reasonable, but together they leave marketing graded on several scoreboards, with no way to compare spend across units.
Three things help:
- A common core. Pick a short list of KPIs that every unit reports the same way, with the same definitions and time windows.
- Unit metrics as diagnostics. Let each GM keep their own metrics, but use them to explain results, not to decide budget.
- A common currency. Convert results to incremental contribution per dollar, so a dollar in one unit can be compared with a dollar in another.
Agree on this with the GMs before the first test, not after the results come in.
A few habits make this work at any enterprise brand:
- Name one owner for the measurement framework, ideally outside any single channel team.
- Write test plans before launch, including what result would change the budget.
- Review agency and platform numbers against your own incrementality results, not instead of them.
- Report ranges, not false precision. "Between 1.8x and 2.6x incremental ROAS" is more honest, and more useful, than "2.2x."
For more on setting the KPIs themselves, see what good marketing KPIs and a measurement framework look like at an enterprise.
Bringing in brand signals once ROI is figured out
ROI tells you what media returned in the near term. Brand signals tell you whether the brand is building the demand that makes next year's media cheaper. Add them after the foundation is in place, because you need to know what media is driving before you can read what's moving the brand.
Here's how I'd bring them in:
- Pick a handful of signals and track them weekly. Index each one to a pre-period baseline so they sit on one scale.
- Test whether they move with media. Use them as secondary outcomes in incrementality tests, and as lagged outcomes in the MMM, since brand effects usually show up weeks after the spend.
- Keep them out of ROAS. Treat brand signals as leading indicators that sit beside ROI, not as inputs to it. Mixing them in muddies both.
- Set alert thresholds. A sudden drop in sentiment or branded search should prompt a conversation before it shows up in revenue.
These are the views I'd build:
| View | Source | What it tells you | Watch out for |
|---|---|---|---|
| Search volume | Google Trends, Search Console, ad platform data | Branded and category search show demand the brand created | Seasonality, PR spikes, competitor-name confusion |
| Social sentiment | Social listening tools | How people talk about the brand and react to campaigns | Bots, small samples, one viral post skewing the average |
| App store sentiment | App store ratings and reviews | Product experience and trust, release by release | Rating prompts and version changes shift scores |
| Customer service sentiment | Support tickets, chat and call data, CSAT | Friction and trust problems, often before churn shows up | Volume swings from product issues, not marketing |
| Brand tracker survey | Survey panel | Awareness, consideration and preference | Sample size, and the lag between surveys |
Put them on one page: each signal as a weekly line indexed to a baseline of 100, with campaign flights shaded behind it. That makes it easy to see which signals move when media runs, and which move on their own. Most sentiment data is modeled and noisy, so read direction and trend before any single week's number.
Where to start: the first 90 days
- Days 1–30: audit. Inventory every channel, its owner, the KPI it reports against, and the last time anyone tested its incrementality. You'll find gaps fast.
- Days 31–60: align. Agree on one set of short- and long-term KPIs and a single definition of a conversion. Pick the two or three channels with the most spend and the least proof.
- Days 61–90: test and baseline. Launch the first incrementality test and start an MMM on whatever clean history you have. Both give you something to build on even if they're imperfect.
If you want to see where your team stands before you start, take the best-in-class measurement quiz. And if you'd rather have a second set of eyes on the whole framework, that's what my Measurement Audit is for.