Why Attribution Methodology Matters More Than Attribution Numbers
Attribution reporting is everywhere in location-based marketing. Most platforms that run campaigns against geographic audiences now offer some version of "store visit" or "foot traffic" reporting. The numbers appear authoritative. They are used to justify budget decisions, calculate ROAS, and evaluate campaign performance.
What is rarely discussed — and rarely disclosed prominently — is how those numbers are produced. And the methodology matters enormously, because two attribution systems can report very different numbers for the same campaign, and both can claim to be accurate.
Understanding the difference between deterministic and modeled attribution is not a technical exercise. It is a prerequisite for trusting your marketing data.
Modeled Attribution: What It Is and How It Works
Modeled attribution is the dominant approach in most ad platforms. Here is how it typically works:
- A campaign serves ads to a target audience.
- The platform tracks which devices received the ads.
- A sample of those devices is observed — through GPS signals, app data, or panel data — to see if they visited the relevant locations.
- A statistical model is applied to extrapolate from the observed sample to the full exposed audience.
- A visit estimate is produced: "Your campaign drove an estimated 4,200 store visits."
The key word is estimated. Modeled attribution is built on a sample and extrapolated through a model. The sample coverage varies widely by platform and is rarely disclosed — which makes it difficult to know how much of the audience the reported numbers actually reflect. The model assumptions determine the output as much as the underlying data does.
Modeled attribution answers: 'How many visits do we estimate this campaign probably drove?' Deterministic attribution answers: 'How many confirmed visits did this campaign actually drive?'
Deterministic Attribution: What It Is and How It Works
Deterministic attribution does not extrapolate. It observes. Here is how NXTeck's deterministic approach works:
- A campaign serves ads to identified target devices.
- NXTeck observes the physical location of those specific devices over time.
- When an exposed device is observed at a physical location that matches a campaign target, the visit is confirmed.
- The report shows confirmed visits — not estimates, not extrapolations, not modeled projections.
The result is a different kind of number. Instead of "estimated 4,200 visits," the report shows "confirmed 412 visits." The number is smaller. But it is real.
Why the Numbers Look So Different
A question that comes up frequently when brands switch from modeled to deterministic attribution: why is the number so much lower?
The answer is that modeled attribution systematically overstates visits. The reasons are structural:
- Panel bias. The sample used to model visits is rarely representative of the full audience. Panels tend to over-represent certain demographics and device types.
- Model inflation. The statistical models used to extrapolate from sample to full audience often produce upward bias — because they are calibrated to produce plausible-looking results.
- Attribution window inflation. Many platforms attribute any visit within a 30-day window to the campaign — even for consumers who had no meaningful exposure to it.
- Visit definition variation. Some modeled systems count a device passing near a location as a "visit." Others require a dwell time threshold. The definition choice dramatically affects the count.
| Factor | Modeled Attribution | Deterministic Attribution |
|---|---|---|
| Visit confirmation | Estimated from sample | Directly observed |
| Typical coverage | Partial sample, extrapolated | Observed across full exposed audience |
| Visit definition | Varies by platform | Physical location match with dwell |
| Reported number | Inflated estimate | Conservative confirmed count |
| Actionability | Directional only | Decision-grade |
What to Ask Your Attribution Provider
If you are receiving visit attribution data from a media partner, these questions will quickly clarify what you are actually looking at:
- What percentage of exposed devices are directly observed? If the answer is less than 50%, the rest is modeled.
- What is your visit definition? Specifically: is dwell time required, and if so, how much?
- What is the attribution window? A 30-day window attributes visits that may have nothing to do with the campaign.
- Do you provide a control group comparison? Without a holdout, there is no way to know how many of the reported visits would have happened regardless of the campaign.
- Are these estimates or confirmed observations? The word 'estimated' is the tell.
The Practical Implications
Operators who switch from modeled to deterministic attribution typically see their reported visit numbers decrease significantly. This is initially uncomfortable. But it is accompanied by a significant increase in confidence in the numbers that remain.
The downstream effect on decision-making is important. With modeled attribution, ROAS calculations are based on inflated visit counts — producing ROAS figures that look strong but are based on estimates. With deterministic attribution, ROAS is calculated on confirmed visits — producing a conservative number that can be trusted as a lower bound on actual performance.
A confirmed visit count of 400 is more valuable than a modeled estimate of 4,000. One is a real number. The other is a story about what probably happened.
The Bottom Line
Attribution methodology is not a technical detail. It determines whether your marketing data is telling you what actually happened or what a model estimates probably happened.
For location-based businesses making real budget decisions based on campaign performance, the distinction between deterministic and modeled attribution is the difference between decision-grade data and plausible-sounding estimates. Ask your partners which one they are providing. The answer will tell you a lot.
