Should you trust last click attribution at your size

Understand where last click attribution is good enough for a smaller business and where it will quietly push you to cut the channel creating demand.

Why the question comes up

Your reports credit whichever channel the buyer touched last. Someone reads your reports and says this cannot be right. The Instagram work looks worthless. Brand search looks like your best performer even though those people already knew your name. Meanwhile the ads platforms each claim more conversions than the business recorded in total.

Last click is not a mistake. It is a rule for dividing credit and like every rule it is wrong in a predictable direction. Knowing the direction is more useful than replacing the model.

What moving beyond last click costs

Changing the attribution model adds no data. It reallocates credit across the touchpoints you already record. That is worth saying first because attribution projects are often sold as though they reveal hidden customers. They do not.

Real alternatives cost real money. Modelled attribution needs event-level history in a warehouse and someone to maintain the pipeline. Data-driven models inside ad platforms need conversion volume before they fit anything meaningful and on thin volume they produce confident noise. Incrementality testing is the honest option and it costs the most because you have to switch spend off in a region or an audience and accept lost sales for the length of the test.

Who can trust it and who cannot

If you run one paid channel do not pay anyone to rebuild your attribution. There is nothing to attribute. Spend divided by new customers answers every question you have and a multi-touch model will give you the same answer with more steps.

Last click becomes actively dangerous when you run demand creation alongside demand capture. Reels and video and creator work introduce people to you. Brand search and retargeting catch them later. Last click hands the credit to the catcher every time. Read those reports literally and you will cut the channel that fills the top and then watch brand search volume fall a quarter later.

It also fails when the sale finishes off the website. A buyer who finds you on Google and then calls or messages on WhatsApp produces a last click of nothing useful. Businesses selling that way need a source question at the point of enquiry far more than they need a better model.

How credit actually gets assigned

Each ads platform watches for its own click or view identifier and claims a conversion that follows inside its lookback window. Two platforms can both claim the same order because neither can see the other. Your analytics property applies its own rule and usually credits the last channel that was not direct traffic. None of these systems is lying. They answer different questions and they are summed together by people who assume they answer the same one.

Lookback windows matter more than model names. A long window credits touches from weeks ago and makes a channel look strong. A short window credits only the final days. Check the window before you compare two numbers because that alone explains many disagreements.

The number to steer by

The metric that matters is blended cost of acquisition. Total marketing spend for the month divided by new customers for the month. It ignores attribution entirely which is its virtue. Watch its trend as you change spend. If you add budget to a channel and blended cost falls then the channel is doing something real.

The metric that flatters is in-platform return on ad spend added up across platforms. Sum those and you will often account for more revenue than the business actually took. Any model that explains more sales than you made is flattering you.

The common mistake

The expensive error is switching off the channel with the worst last click numbers. It is usually the channel that created the demand. The effect is delayed and counterintuitive. Every remaining platform reports better efficiency because the easy conversions are still arriving. Then brand search volume thins out and retargeting pools shrink and total new customers fall while the dashboards look healthier than before.

Protect yourself by testing rather than arguing. Hold the suspect channel off in one state or one audience for a few weeks and watch blended cost and total orders rather than that channel's own report. If nothing moves you have learned something worth the lost spend. Nobody can tell you the answer in advance and any agency that claims to know it without a test is guessing.

A workable order of trust

Use blended numbers to decide how much to spend in total. Use each platform's own reporting only to optimise inside that platform where its view is complete. Use a source question on your enquiry form for direction because imperfect self-reported data beats no data on offline channels. Reach for incrementality tests when a single channel's budget is large enough that being wrong about it costs more than the test.

This guide explains how something works. It is not legal, tax or financial advice, and rules change. Confirm anything regulatory with a qualified professional before you act on it.