Right-Sizing Marketing Measurement

The full marketing measurement stack, as described in conference talks and vendor pitches, includes multi-touch attribution, media mix modeling, incrementality testing, and unified data warehouses feeding real-time dashboards. Some organizations run all of it simultaneously. Most don’t, and the ones who don’t are often making a perfectly rational decision given their actual resources and business context.

The problem I’ve seen more often than under-measurement is companies either investing in measurement infrastructure they lack the capacity to act on, or measuring nothing meaningful at all. Both are waste. An MTA or ML-based attribution model that nobody in the organization has time to interrogate is not an asset.

What follows is a rough framework for thinking about measurement expectations by budget and resource availability. These tiers reflect total marketing budget (including headcount, agency fees, and media spend combined), not media spend alone. The categories aren’t precise, and budget isn’t the only variable that matters (a $3M marketing budget spread across a five-person team is a fundamentally different situation than the same budget where one person runs everything). But the tiers are a useful approximation, and the headcount ranges refer to marketing team size, not company headcount. One scope note: this framework assumes a relatively direct path from marketing activity to a measurable conversion event. B2B organizations with long sales cycles, where the journey from first touch to closed revenue runs through a sales team over months, face additional attribution complexity that warrants separate treatment.

Small Businesses and Early-Stage Companies

At this level, the honest answer is: get the basics right and don’t try to do more. The basics are not glamorous. You need to know where your leads or customers are coming from in broad strokes, you need to know what your cost to acquire a customer looks like (even as a rough estimate), and you need to know whether your core conversion actions are actually being tracked. That last one is more often broken than you’d expect. GA4’s default configuration tracks very little of what matters without deliberate setup, and most small businesses are running on defaults.

For most small businesses, this means GA4 configured properly, conversion tracking set up in whatever ad platforms you’re running, and a CRM kept clean enough to report from. If you have those three things working, you can answer the questions that actually matter at this stage: which channels are generating leads, which are not, and roughly what each one is costing. Adding “how did you hear about us?” to your intake form is worth more than most people assume. Survey data is imprecise and self-reported, but it surfaces patterns that click-based tracking systematically misses, particularly for word-of-mouth and offline touchpoints.

What you can’t do at this level, and shouldn’t try to do, is distinguish between channels with statistical confidence, build attribution models, or isolate the incremental effect of any specific investment. You don’t have the data volume. You don’t have the analyst time. And in most cases, the decisions you need to make don’t require that level of precision. Trust your directional read and move. Refinement comes with scale.

Mid-Market Companies with a Small Dedicated Team

This is where the gap between what’s possible and what most companies actually do is widest, and where better measurement practice creates the most disproportionate advantage.

At this tier, you likely have enough data volume and enough channel diversity to start measuring with real discipline, but you almost certainly don’t have a dedicated analytics person, which means someone on the marketing team is doing measurement work on top of everything else. That constraint should shape your approach: build for sustainability, not sophistication.

The priority at this level is clean, consistent channel-level data tied to actual revenue outcomes rather than proxy metrics. This means every significant traffic and lead source has proper UTM tagging, your CRM is connected closely enough to your website analytics that you can trace revenue (or at least opportunities) back to channel, and whoever is presenting results to leadership is presenting the same two or three metrics consistently rather than shifting the frame based on what looked good that month.

The most common failure mode I’ve seen here is performative reporting: weekly decks full of impressions, reach, and engagement rate presented as evidence of marketing effectiveness, with no clear line to revenue. This isn’t always cynical. It happens because connecting activity to outcomes requires cross-system data work that nobody has time to set up properly, so the default is reporting what’s easy. But impressions are not outcomes, and eventually leadership figures that out.

The specific investments worth making at this tier, in rough priority order: close the loop between marketing activity and CRM-sourced revenue (even imperfectly), establish a consistent set of metrics that won’t change quarter to quarter, build a simple but honest cost-per-acquisition view by channel, and do at least basic cohort analysis on customer lifetime value if your business has any meaningful retention component. If you have an email list, open and click rates are not the metrics that matter. Revenue from email is.

What you’re not trying to do: precise multi-touch attribution, statistically rigorous incrementality testing, or media mix modeling. At the upper end of this tier (and only with sufficient data history and media volume), lightweight MMM options may become worth investigating. But the conditions for MMM to produce reliable output are more demanding than they appear: two or more years of consistent data, meaningful variation in spending across channels and time periods, and enough weekly conversion volume for the model to detect channel-level effects against the noise. Doing it incorrectly produces confident-sounding wrong answers. A simple last-touch model that everyone understands and trusts is more valuable than a sophisticated multi-touch model that nobody believes.¹

Large Companies

Once you have someone whose job is to think about data full time, the calculus changes. The basics should be in order by now (if they’re not, start there regardless of company size), which means you can start asking harder questions.

The distinguishing capability at this tier is the ability to do some form of incrementality measurement. The relevant question is whether channels are actually generating business that wouldn’t have happened otherwise, rather than simply which channel gets credited for the conversion in a last-touch model. A channel can show strong performance in attribution while contributing very little incrementally, because it’s good at capturing demand that already existed rather than creating new demand. Branded search is the canonical example: it tends to look excellent in any attribution model, but a well-designed holdout test often reveals that a substantial portion of that traffic would have converted anyway. That doesn’t mean branded search has no value (it almost certainly does), but understanding the difference between captured and created demand is what makes budget allocation defensible.

Getting to incrementality doesn’t require enterprise infrastructure. Geo-based holdout tests, time-based holdout tests for specific channel investments, and careful use of platform-native lift measurement tools are all accessible at this scale if you have someone with the analytical ability to design and interpret them. The honest caveat is that each of these approaches has methodological limitations, and a single test should inform your thinking rather than definitively settle a question. But directionally correct and imperfect is still ahead of where most companies are operating.

At this tier you should also have enough data history to build a reasonably useful media mix model, through a vendor or through open-source frameworks like Google’s Meridian or Meta’s Robyn. These models are not oracles. They’re another input. But for companies spending meaningfully across multiple channels, having a model that estimates long-term and short-term effects separately and can scenario-plan budget allocation is genuinely useful as a complement to channel-level attribution. The realistic timeline expectation: most organizations at this scale run MMM annually, and the first run functions as a diagnostic rather than a deliverable. Consistent iteration over two or three years is where the value accumulates.

The investment priority shift at this level is from getting the data right to building the analysis capability to ask better questions. That means hiring or developing someone who can do experimental design, not just reporting, and establishing a cadence where marketing investments are evaluated against expected returns with accountability for that assessment, rather than presented as proof of activity.

One measurement workstream that tends to get deferred until too late at this tier: brand health. Attribution, MMM, and incrementality testing all measure performance against existing demand. None of them tell you whether your brand is gaining or losing ground in the consideration set of people who haven’t converted yet. For companies with meaningful investment in brand-building channels (CTV, streaming audio, out of home, sponsorships), some form of ongoing brand tracking becomes necessary to evaluate whether that investment is doing anything. Share of search is a lightweight proxy worth monitoring. Periodic tracking studies on aided and unaided awareness, brand association, and purchase intent provide a more complete picture. These don’t need to be expensive wave studies at this scale, but ignoring brand health measurement while spending material budget on brand-building channels is an accountability gap that will eventually surface as a budget justification problem.

Enterprise Organizations

Enterprise marketing organizations have the resources to address their measurement challenges. The reason most don’t isn’t budget or tooling. It’s that the organizational dynamics that emerge at scale work against measurement accuracy in specific, predictable ways.

Incentives diverge from accuracy. At mid-market scale, the person running the channel is usually the person reporting on it, and leadership is close enough to see through bad numbers. At enterprise scale, the person presenting results to a senior leader is three layers removed from the data, every channel team has a budget to defend, and the incentive to present favorable numbers compounds through each layer of removal. The performance of data-driven culture ends up substituting for the practice of it, which is a measurement problem before it’s a management one.

Cross-channel cannibalization becomes invisible. When five people run eight channels, everyone roughly knows what everyone else is doing. When forty people run fifteen channels across business units, paid search is claiming conversions that brand TV created, retargeting is taking credit for organic intent, and nobody has the cross-functional visibility to see it. The measurement infrastructure that was adequate at mid-market scale actively misleads at enterprise scale, because the interactions between channels are where the real story lives.

The data infrastructure problem inverts. At mid-market, the challenge is not having enough data. At enterprise, the challenge is too much data in too many systems that don’t reconcile. Three different sources of truth for the same conversion event is common, and reconciling them becomes a full-time job before anyone has asked a strategic question.

Testing becomes politically difficult. A geo-holdout test that suppresses paid social in four markets is a straightforward experiment at $10M in spend. At $80M, that test has a VP whose bonus is tied to those markets, a regional sales team that will blame the experiment for any quarterly shortfall, and a CFO asking why you’re deliberately leaving revenue on the table. The measurement capability exists. The organizational willingness to run clean experiments erodes as the financial and political stakes get larger.

Enterprise organizations don’t need a framework from a blog post. They need will, not direction.

The Common Thread

Regardless of where you fall in this spectrum, directional accuracy and consistency over time is worth more than sophisticated precision. A company that tracks the same five metrics cleanly for three years, with consistent methodology, can answer questions about what’s working that no amount of expensive tooling will answer if the underlying data is unreliable or the framework keeps changing. Measurement is a discipline, not a technology purchase.

The corollary: better measurement at the wrong scale wastes more money than imprecise measurement at the right one. Before adding infrastructure, the right question is whether you have the analytical capacity to act on what it would tell you.

References

1. For a detailed look at data and organizational requirements for Media Mix Modeling, see Making the Invisible Visible: A Practical Guide to Media Mix Modeling for Mid-Sized Companies. ↩