The last-click trap distorting ad-budget decisions

"Instagram Ads ROAS (revenue multiple on ad spend) is low — let's cut it next month." Ever made this call staring at an ad report? That decision might be eroding the revenue foundation of your business.

Because the last-click attribution most e-commerce operators rely on doesn't see the full user journey. This article covers the limits of last-click, the multi-touch attribution concepts that fill the gap, and how to use the four common models for different questions.

Key takeaways#

1. Last-click is good for evaluating purchase-triggered performance, but zero-weights awareness and comparison stages
2. Four multi-touch models (Last / First / Linear / Time-decay) each answer different questions
3. Decide "what you want to know" first, THEN pick the model — that's the right usage

1. What last-click actually measures#

Last-click attribution is the simplest and most intuitive way to split revenue.

"100% credit the channel of the final pre-purchase session."

If a user entered via branded search and bought, that revenue is 100% "Direct / branded search." If they clicked a Google search ad and bought, it's 100% "Google Search Ads." Simple, easy to aggregate, and the default in most analytics tools.

That simplicity is a huge upside. Open a report and you can immediately see "which channel directly produced purchases." This is exactly why it's used so widely as a performance-evaluation metric.

The problem is this intuition often diverges from actual user behavior.

2. When last-click lies#

Consider a concrete journey:

Tuesday: Discovered the product through an Instagram ad → Thursday: Compared it on Google Search → Saturday: Visited via branded search and bought

Ask "which channel produced this revenue?" and most people answer: "Instagram created awareness, Google supported comparison, and branded search closed." Three channels collaborated to produce the purchase.

But under last-click, this revenue goes 100% to Direct (branded search). Instagram Ads and Google Search get zero credit.

How does this distort decisions? A hypothetical. An e-commerce store does ¥6M/month in revenue. Ad spend: Instagram ¥200K, Google Search ¥100K. Under last-click aggregation:

Looking only at this, the conclusion is "cut Instagram, shift to Google Search." Many operators actually make this call.

Now re-aggregate the same data under first-touch (credit the first-visit channel):

The landscape flips: "Instagram is bringing in new customers, who later convert via Google branded search." If you cut Instagram here, the volume of Google branded search itself starts shrinking a few months later — a second-order revenue collapse.

Last-click doesn't lie — but it doesn't show you the whole picture either. That's the reality.

3. The four attribution models#

To complement last-click, the field developed multi-touch attribution — an umbrella term for different ways to distribute credit across the channels that participated in a purchase.

The four canonical models:

• Last-touch: 100% credit to the final channel (same as last-click). Good for evaluating purchase-triggered performance
• First-touch: 100% credit to the first channel. Evaluates which channel brought in new customers
• Linear: Even distribution across all channels involved. Assumes equal contribution
• Time-decay: Heavier weights for touchpoints closer to purchase. Emphasizes the "final push" while not zero-weighting awareness-stage channels

Each has a philosophy, and each has its own "correctness" and "distortion." The important fact: no model is absolutely correct. With the same data, different questions call for different models.

4. Choosing a model — "by what you want to know"#

Here's a simple framework.

The key operational point: don't get stuck on one model. For new-acquisition channel review, use first-touch. For last-minute campaign evaluation, use last-touch. Pick the question first, then pick the model — that's the right usage.

As long as you only look at last-click, this "switching by question" isn't even on the table. When ad-budget decisions depend on a single axis, you can only respond to complex reality by simplistically cutting it.

If you're moving even ¥200K/month in ads, not depending on one attribution model protects the business. Before cutting because ROAS looks low, just checking "what does this look like under another model" avoids catastrophic judgment errors.

One prerequisite: this all assumes you have "revenue by ad channel visible on one screen." Why GA4 makes this question hard is covered in a companion article: GA4 isn't a revenue tool — the attribution blind spot.