Most marketing measurement work is structural — what's driving the business across channels, geographies, time. Sometimes the question is narrower and harder. A specific campaign, a specific shift in spend, a specific product line: did that work? The structural model often can't answer cleanly because the campaign is too short, too localized, or too entangled with other things changing at the same time.
For a luxury travel operator, the question was concrete. The brand had meaningfully increased media weight behind a newer product line over the prior year, both in absolute spend and as a share of the total marketing budget. Leadership wanted a defensible read on whether that shift was producing a return — and whether to push further.
Three Independent Reads
No single data source was going to settle this convincingly. Bookings happen on a long lag. Leads are noisier. Web behavior is immediate but doesn't directly tie to revenue. So we built three converging reads on the same campaign, knowing that any one of them could be misleading on its own and that agreement across them would be far more credible than any single number.
When the question is "did this specific intervention work?", a single regression rarely settles it. Three different methods that point to the same answer settle it. Three different methods that disagree are also informative — they tell you the answer isn't yet there.
The first read was web behavior: when the campaign was in-market, did site visitors actually browse more of the targeted product's content? Did related brochure downloads rise? The data here is high-frequency and tightly tied to media timing, so it's a fast indicator that the upper-funnel is doing its job — assuming the rest of the funnel follows.
The second read was a regression of leads on spend, separating campaign-attributable spend from the rest of the marketing budget. The model controlled for time-of-year, prior-year baseline, and total marketing intensity. The coefficient on campaign spend has a useful interpretation: dollars in, leads out, holding everything else constant.
The third read was a Monte Carlo simulation built on the regression output. The point estimate from a single regression is fragile — the true coefficient could be higher or lower depending on data noise. By drawing thousands of plausible coefficients from the model's confidence interval and propagating each through the lead-volume math, we got a distribution of likely outcomes rather than one number. That tells the team not just "what's the best estimate" but "how confident should we be."
What the Reads Showed
All three pointed in the same direction. Site visitors browsed materially more of the campaign-related content during the in-market periods. Brochure downloads tracked the spend pattern with a tight visual fit. The regression returned a coefficient implying roughly five additional leads per $1,000 of campaign-attributable spend. The Monte Carlo simulation, drawing 10,000 plausible coefficients, put a tight distribution around that point estimate — meaning the answer wasn't just numerically positive, it was robust to the kinds of noise that would have flipped a less stable result.
of campaign-attributable spend
used to test the estimate
reads — all aligned
The 19% Gap Worth Noticing
One useful finding sat one layer beneath the headline. While the campaign was producing leads efficiently, the share of media spend going to the new product line had risen ahead of the share of bookings it was generating. The gap — roughly 19 percentage points between spend share and booking share in the prior year — wasn't necessarily a problem (acquisition spend often runs ahead of revenue when a category is growing), but it was the right tension to surface. The answer isn't to cut the spend; it's to monitor whether the booking share catches up over the next planning cycle, and to recalibrate if it doesn't.
Why "Did It Work?" Is Different from "What's Driving the Business?"
Marketing mix modeling is the right tool for the structural question — what's driving total demand, where's efficiency strongest, how should the budget split. It's not always the right tool for the campaign question. A specific campaign, a specific reallocation, a specific test in a specific window often calls for a more targeted measurement design: web-behavior tracking close to the activity, a focused regression on the relevant outcome, and an honest treatment of uncertainty around the result.
The pattern generalizes. Subscription brands testing a new audience segment, retailers piloting a category investment, B2B teams trialing a new outreach motion — anywhere a discrete bet is being placed and the team wants a defensible read on whether it's working, layered measurement beats any single number. The hard part isn't running the regression. It's having the discipline to triangulate before declaring victory.
Trying to evaluate whether a specific campaign or investment is paying off? Say hello.