The Ecomm Analyst

The first-touch versus last-touch debate eats more operator hours than it deserves. People treat it like a deep methodological choice with a right answer, when for most stores it is closer to picking which end of the customer journey you want to flatter. Both models are simplifications. Both are wrong in predictable ways. Knowing exactly how each one is wrong is worth more than picking a side.

Last-touch attribution gives all the credit for a sale to the final click before purchase. Someone discovers you through a Meta ad, thinks about it for a week, then searches your brand name on Google and buys. Last-touch hands the entire sale to Google brand search. The model is the default almost everywhere because it is easy to track and it credits the click closest to the money. Its blind spot is everything that happened earlier. It systematically overcredits bottom-of-funnel channels, brand search, retargeting, email, the stuff people click right before buying, and it makes your top-of-funnel discovery channels look weak because they rarely get the last click.

First-touch does the opposite. It gives all the credit to the first interaction, the Meta ad that started the whole thing in the example above. It is the model people reach for when they suspect last-touch is starving their awareness spend. Its blind spot is the mirror image. It overcredits whatever introduces people and ignores everything that actually closed them, so it can make a cheap top-of-funnel channel look like a hero while the retargeting and email that did the convincing get nothing.

Here is the part that cuts through most of the debate. For a lot of smaller stores with short consideration cycles and one or two real channels, first-touch and last-touch produce nearly the same answer, because the first touch and the last touch are frequently the same touch. If most of your customers see one Meta ad and buy from it that week, arguing about which end to credit is arguing about a difference that barely exists in your data. The debate gets expensive mainly for stores with long journeys and many channels, and even there the honest move is not to crown one model but to recognize neither single-touch model is telling you the truth.

The reason both are wrong is that they assign one hundred percent of the credit to a single touch when real purchases are caused by several. Multi-touch models try to fix this by spreading credit across touches, linear splitting it evenly, time-decay weighting the recent ones more, position-based loading the first and last. These are better in theory and they come with their own catch, which is that the weights are assumptions you picked, not facts the data revealed. A prettier distribution of credit built on guesses is still built on guesses. It feels more sophisticated without necessarily being more correct.

So what should you actually do. Stop expecting any attribution model to hand you truth and start using them as lenses. Look at the same period through last-touch and through first-touch and pay attention to where they disagree, because the disagreement is the information. A channel that looks great under first-touch and terrible under last-touch is doing discovery work, introducing people who get closed elsewhere. A channel that looks great only under last-touch is doing closing work on demand other channels created. Neither is good or bad on its own. You need both jobs done, and seeing a channel’s role is more useful than seeing a single disputed credit number.

The honest answer to which model to run as your default, if you must pick one, is last-touch for day-to-day channel reads, with the explicit awareness that it underrates discovery, paired with periodic incrementality tests to check what is actually incremental rather than just last in line. That combination, a simple default model you understand the bias of plus real holdout testing, beats an elaborate multi-touch model you have to take on faith.

This is also where a tool stops you from rebuilding the same view by hand every week. I work with ThoughtMetric and run it across the stores I manage, so I am not pretending to be a neutral observer here. ThoughtMetric lets you look at the same orders under different attribution models side by side, so the first-touch versus last-touch comparison that reveals each channel’s role becomes a setting you toggle rather than a spreadsheet you assemble. Any order-level attribution tool in the category will let you switch models like this. The value is not that the tool resolves the debate, because no tool can. The value is that it makes looking at a channel through multiple lenses cheap enough that you actually do it instead of defaulting to whichever model your dashboard shipped with.

The takeaway is to stop hunting for the correct model and start reading the gap between models. First-touch flatters discovery, last-touch flatters closing, the truth is spread across both, and the channels whose verdict swings hardest between them are the ones whose role you most need to understand. Pick a default you know the bias of, check it against real incrementality now and then, and spend the hours you save not relitigating attribution theory on the parts of the business that actually move the number.