Marketing Attribution
Marketing attribution is the practice of assigning credit for a sale to the marketing touchpoints that preceded it. A customer might see a social ad, search the brand, read a review, and return a month later to buy; attribution decides how much of that sale each step earned. Teams allocate budget by that credit, so an error in the credit becomes an error in the budget.
How it actually works
Standard attribution systems are off-the-shelf tools that split the credit for a sale by one of several rules: last-click, first-touch, or a multi-touch model. All of them run on the same tracking, the pixels and tags the platforms provide, and all of them assume the entire customer journey can be tracked.
That assumption fails in two places. The journey does not start online in many cases: a CTO who needs more bandwidth and scale in a data platform often begins by calling a peer who already solved the problem, checking Reddit, or asking an AI assistant, and may never visit the vendor's website. The journey does not end online in most cases: the US Census Bureau's quarterly e-commerce report measures e-commerce as a share of total US retail sales, and the complement of that share, over 83%, closes in physical stores, where no tag records the purchase. Phone calls, store visits, and point-of-sale terminals produce no tracking events.
Attribution therefore lists the tracked touchpoints that preceded a sale. Which touchpoint caused the sale, and which steps of the journey went unrecorded, are outside what the model measures. Teams that treat the model's output as complete over-credit the last inexpensive click, cut the work that created the demand, and interpret the resulting decline as a market problem. Two checks validate any attribution model without relying on its tracking: marketing efficiency ratio, total revenue divided by total marketing spend, which reconciles to the P&L because it requires no attribution at all, and holdout experiments, which measure what a channel caused by withholding it from a comparable group. A model whose channel-level story cannot be squared with the MER trend and the holdout results is describing the tags rather than the business.
In practice
When a publicly traded mattress company expanded from online into retail, customers learned about the mattress online and bought it in stores, where they could try it first. The website showed near-zero conversion and the channel's reported ROAS fell. The team optimized for that number and moved away from the positioning that had built the brand in its first five years, while spending roughly 25x the category leader's digital ad budget, a figure from Fourth Order Intel's own analysis of category digital ad spend data from Statista. The public record shows the outcome: net losses of $89.9 million in 2022 and $120.8 million in 2023. The sales were happening in a channel the tracking could not record, so the attribution data described a failing channel while the store channel absorbed the demand. The reverse case is what attribution looks like when the system works together. Within days of launching Facebook ads for Transparent Labs, searches for the brand spiked from almost nothing to nearly 30,000 that week, and there were 5x more visits to the site than clicks on any of the ads. Regional controls showed a lift on all channels in the regions where ads were active. First-visit attribution would have credited the affiliate site or branded search a prospect entered through; the controls showed the ads created the demand. The brand grew from $2M to $10M in 12 months. The case is documented in [the reality of channel attribution](/insights/marketing-measurement#the-reality-of-channel-attribution).
Where we come in
We measure what a channel caused, not what a tag credited it with, using controlled experiments and portfolio-level models that do not depend on the platforms' own data. It is the same approach behind our work with Digital Realty and Equinix, whom we helped go to market with new software for enterprise buyers, and with Potbelly, where efficiency gains were proven at level budget before spend was scaled.
Start a Revenue Health Pre-Assessment →See it in action
Related terms
- Last-click / last-touch attribution
- All credit to the final touchpoint. It over-credits branded search and remarketing, the steps closest to a purchase that was already underway.
- First-touch attribution
- All credit to the first touchpoint. It over-credits whatever introduced the customer and ignores what closed them.
- Multi-touch attribution
- Splits credit across touchpoints by a rule (linear, time-decay, position-based). More complete than single-touch, and still limited to what the tags record.
- Marketing mix modeling (MMM)
- Estimates each channel's contribution from historical data at the portfolio level, without user tracking. It is unaffected by cookie loss because it never used cookies.
- Cookie / attribution window
- The period after a click or view in which a later sale still counts toward the ad. A longer window credits more sales to the ad, which is why platform defaults are generous.