Companies often justify investments in first-party data by promising better attribution. They want to know which marketing dollars are working and which aren’t. It’s a fair question, but there are often more immediate problems to solve.

I have built and commissioned more marketing measurement work than I can count: last-touch and multi-touch attribution, media mix models, incrementality tests, and setups that reconciled their findings. I’ve even had my team build causal impact studies to support incrementality testing. Years of trying to answer executives’ questions about marketing returns have taught me that a more sophisticated model cannot compensate for unreliable inputs.

The investment sequence I recommend is efficiency first, effectiveness second, and advanced attribution after that. Basic measurement belongs in every stage. You need to establish what is happening and whether a change helps. The mistake is making a complex attribution program the first priority while the data needed to support everyday campaign decisions remains incomplete. This is a priority for investment, not a reason to postpone a useful experiment while waiting for perfect data.

Efficiency: Stop paying for the wrong audiences

There are several kinds of wasted media spend. The ANA’s 2023 programmatic study identified potential efficiency gains equivalent to roughly a quarter of open-web spending. It also found that only 36 cents of each dollar entering a demand-side platform effectively reached consumers. The difference included intermediary fees and low-quality impressions. Identity resolution alone cannot fix those supply-chain problems.

It can help address a more specific problem: paying to reach people who should have been excluded from a campaign.

A global hospitality company I worked with used hashed first-party data to suppress already-converted customers, employees, and top-tier loyalty members from relevant acquisition campaigns. Matching those records to platform audiences helped the team exclude people it had little reason to pay to acquire again and reduce unnecessary spend. The company reported significant savings from suppression, although its published case study does not quantify them.

Audience suppression is one of the first use cases I often recommend. You can check whether exclusions are working and compare spend and delivery against a baseline without building a multi-touch attribution model. If exclusions reduce total spend, you can report that saving. If the budget is reallocated, you still need to measure what the new allocation achieves.

Effectiveness: Give platforms better signals

The next opportunity is to improve the information platforms use to optimize campaigns. A global travel company moved its conversion reporting to a server-side integration, reducing its dependence on browser-based tracking and giving ad platforms more complete booking data. Server-side delivery still depends on what data can be collected and shared in line with customers’ consent choices.

After the change, the company reported a six percent reduction in cost per booking and a 29 percent increase in the conversion rate for social traffic.

Those results need careful interpretation. Recovering previously unobserved bookings can improve reported conversion rates and cost per booking even if the number of actual bookings stays the same. Better signals can also help platforms optimize toward completed bookings. The reported figures do not isolate how much of the improvement came from better visibility and how much came from additional business.

The hospitality company reported a four percent lift in paid-search conversion rate, as measured by Google, after implementing conversion APIs. That is a useful campaign result, but establishing incremental growth requires evidence beyond the change in the platform’s conversion count.

Attribution: Use the methods together

This is where a broader measurement program becomes useful. Attribution assigns credit for observed outcomes. Incrementality asks how much additional business marketing generated. Reliable conversion data supports both questions, but answering one does not automatically answer the other.

Multi-touch attribution relies heavily on accurate identity matching and event-level signals, and its findings are constrained by the journeys it can observe. Last-click can also overcredit the final interaction while missing the contribution of earlier activity. Marketing mix modeling generally works with aggregate data on spend, outcomes, and other business factors. Incrementality tests can operate at a geographic level without resolving each individual. These approaches have different data requirements; they do not all depend on the same identity system.

I use triangulation to compare what those methods tell us: marketing mix modeling for broader budget decisions, well-designed incrementality tests to estimate causal effects, and platform reporting for ongoing campaign management. When results disagree, the definitions, coverage, and assumptions behind them need examination.

The hospitality company projected roughly $13 million in annual revenue impact associated with improved attribution and optimization. The travel company reported an 18 percent increase in return on its directly attributable ad spend. These are different measures: one forecasts financial impact, while the other reports a change in attributed return. Neither establishes how much additional revenue the data changes caused.

Connect the work across martech and adtech

The practical priority is to establish reliable inputs, measure the effect of audience and signal improvements, and expand the measurement program as the questions become more demanding. Early gains can help justify further investment, but each stage still needs evidence.

The hospitality example also points to the forthcoming final article in this series. Its customer and loyalty information came from martech systems, yet it informed audience decisions in adtech platforms. Connecting those environments gives teams a way to put customer knowledge to work across campaigns. That connection, and what it takes to make it useful, is where we go next.