How to Choose an Ecommerce Attribution Tool Without Buying Another Dashboard

"We need better attribution."

So you start booking demos. Every tool has a cleaner dashboard than the last one. And every one of them says it can show you what's really driving sales.

But that sentence can mean three different things.

You might be losing tracking. You might have tracking that doesn't match your orders. Or your team might not trust any number at all. Before you buy an ecommerce attribution tool, figure out which problem you have. Then check that your events and orders line up, and know where click-based reporting will always come up short.

If you can't answer those yet, a new tool will just make the same uncertainty look more polished.

What do you actually mean by "attribution tool"?

My view: for a lot of brands, you already have one. It's Shopify, GA4, or your order system. Whatever UTM, click ID, or custom tracking is attached to the click that lands on your site and makes the sale gets recorded as the source.

That's last-click attribution.

Many companies have no other attribution tooling yet, and that's fine. So start with the decision your current setup can't support. Write it down before you open a single demo.

Is it a tracking problem, a reconciliation problem, or a decision problem?

Pull a sample of recent orders and sort them into these buckets:

1. Blank orders with no click IDs at all. Google organic search and organic shopping traffic can carry click IDs. So if an order is truly blank, with not even those identifiers, your tracking may have dropped.
2. A tracked click with a blank order. The click landed, but the sale came through empty. That's an event-capture gap, and it's a different problem. A Shopify app can replace your tracking. A payment step can send the buyer to a third-party site and back.
3. Clean data that nobody trusts. Meta, GA4, Shopify, and finance all say different things, and the budget meeting goes in circles. That's a decision-process problem.

Event capture (GA4-style tracking of each step from landing page to purchase) is how you tell bucket 1 from bucket 2. It shows you exactly where tracking drops.

No tool fixes all three buckets on its own. If bucket 3 sounds like you, start with why your marketing numbers disagree and how to reconcile GA4 and Shopify to the order record.

Diagnose fit before you book demos

Example: you sell a high-dollar product. First-time Reddit clickers don't buy right away. They come back later through Google Search, Facebook, or direct. That's a fair reason to look at MTA. But a few things have to be true first.

Do you capture first-party events?

This one is non-negotiable. First-party events are the site actions you record yourself, under your own domain. Without them, too much traffic has "goldfish memory." The system keeps seeing a returning visitor as new. First-click collapses into last-click, and the MTA shows you nothing you didn't already have.

If you don't have it, build it first. You can do that in-house, with a developer, or through a partner.

Do the tool's orders match yours?

If your order system says a sale was Facebook last-click, the tool's last-click view should say the same. If it doesn't, dig into the gap before you trust any fancier model built on top.

For a baseline: I'd call it extremely normal for third-party client-side tracking (code that runs in the shopper's browser) to catch about 75% of events, purchases included. Add Amazon or phone orders and the share of the whole business is lower. That's my observation, not a pass/fail line. How to check it is in the questions to ask an attribution vendor.

Can it tell a person across visits?

Identity resolution means joining visits and devices into one customer. It will never be perfect. You just need to know the gaps well enough to read the view honestly.

Which metrics do you need?

List them before any demo. Say you want a new-visitor share to judge whether a channel is upper funnel. Then you also need engagement, like bounce rate or pages per visit. Some MTA products don't provide those, and some don't connect deeply with GA4, where that data might live.

Does it connect to what you already use?

Check whether it connects deeply to your order system of record and to GA4. Then list the other sources you'd want in the view, like your store and each ad platform. Anything it can't pull in is a gap in what it can see.

What share of your spend can't be clicked?

Podcasts, TV, and radio. If that's a material 15-30% of spend and central to how the business grows, MTA won't be the whole picture.

Who owns the decision?

If your team will only ever accept last-click as real, an MTA nobody uses is a bad investment. Decide who reviews it, which decisions it feeds, and when a disagreement triggers a test.

These are checks on your own setup. Bring the vendor questions linked above to the demo itself.

So you've passed those checks. Should you buy the whole platform or build part of it?

What should you buy, and what should you own?

The usual pitch is one platform that does everything, start to finish.

But collection and analysis are different jobs.

Here's how I split it. Buy a platform if your gap is collection: server-side events (tracking sent from your server instead of the browser), postback (sending conversions back to the ad platforms), cross-device matching. That's infrastructure, and a platform built for it is the better buy. Everything after the data lands, I build custom around your data sources.

Why? Your reporting, your investigations and your models should fit your own data sources and edge cases. An out-of-the-box dashboard may not handle those.

If the gap is the foundation, see marketing data foundation. If it's what happens after, read custom MTA build requirements. And if you're weighing a platform against building in-house or hiring help, see why your attribution tools disagree, and how to know whether to buy, build or hire.

Can the tool investigate, or can it only display?

A dashboard is a scoreboard. It tells you the score but not how the play happened. You need to be able to ask why a number looks the way it does.

Here's one way to test that.

Take the orders a channel claims. A dynamically inserted podcast ad is a good one. Compare those order IDs across your order system, the MTA, and GA4. Then hand the journeys to an LLM (a large language model, like ChatGPT) and ask: which channel converted them later? How many visits did they make? New or returning? How soon after the podcast ran was the first visit?

Now you have a hypothesis. If the claimed orders behave EXACTLY like the rest of the business, that's a reason to question whether the channel caused them, and maybe to run an incrementality test. In my view, an off-the-shelf MTA that can't cross-reference like this may not justify a four-figure monthly cost. (That's my judgment, not a price quote.)

More on this in a reporting system, not another dashboard. Or watch how to own your ad data and ask it questions with AI.

If a tool can't help you investigate a claim, it's only reporting it.

Where does attribution stop?

Picture this: 15-30% of your spend is podcasts and radio. You install an MTA. It reports zero sales for both.

That doesn't mean they sold nothing. MTA is a click logger. It observes. No click, no record.

A few more blind spots:

• Bots can look like upper funnel. A channel like Reddit can show a high new-visitor share. Without engagement and later return visits, you can't tell real prospects from bots or low-intent traffic.
• "New" starts the day the tool was installed. A tool that relies on its own cookies doesn't know your longer customer history. Its new/returning labels need to match your order system.
• A weak acquisition MER isn't a verdict. (Acquisition MER is new-customer revenue divided by ad spend.) A channel that mostly brings back existing customers can still be incremental, especially if they need a few paid touches to return. And "existing" depends on your definition. Someone whose last order was seven years ago may need paid reach again.

See what MTA can and can't answer, and why a dashboard can be right and still lead to a bad budget decision.

How do you check whether the tool is right?

"The tool says this channel drove the sale. Can we trust it?"

Every attribution model is trying to tell you something about cause.

So the only way to check a model's accuracy is to pair it with a geo test that shows causality. (A geo test turns media off or up in some regions and compares them to matched regions.) One red flag, in my words: a vendor that uses its deterministic pixel-based attribution and claims whatever it reports is incremental.

And watch how a tool handles disagreement. Many dashboards let you switch between click-based attribution rules. But in my experience, vendors rarely run more than one geo control model or more than one MMM. At most, they flag some channels as low confidence, and they rarely show that low confidence as a range of possible iROAS (incremental return on ad spend).

Start with the incrementality testing framework, then check whether a geo test is feasible.

No model owns the truth. A test is how you find out which of its readings to trust.

What's the next useful investment?

Think of these methods as legs of the same stool.

Each one listens for something different. The most complete view of what drives sales may use all of them someday. But not every business needs every tool right now.

1. No first-party event capture, data lake, or identity resolution? Build those first.
2. Foundation in place, and you want journeys beyond last-click? Add MTA. Think first-click over 7-, 30-, 60-, or 90-day windows, in close to real time.
3. Need to know whether a channel causes sales? Run an incrementality test. This matters most when platforms claim more sales than you recorded in total, or when a channel can't be clicked. You don't need an MMM before you use MTA plus testing.
4. Need the whole portfolio, channels working together? Use an MMM (marketing mix model). It can make the orchestra make sense, but it may not give sharp answers until tests calibrate it.

And if you only run Facebook and its numbers broadly match your bottom line? You may not need any of this.

Compare the methods side by side in MTA vs. MMM vs. incrementality testing, or read do you need an MMM yet?

FAQs

Should we replace GA4 with an attribution tool?

Usually not.

GA4, your order records, the ad platforms, and an attribution tool each see different things and define them differently. Decide which question your current setup can't answer. Then check whether the new tool can answer it with orders that match yours.

Can an attribution tool measure incrementality?

No, not on its own. It can show observed paths and help you investigate.

To measure whether a channel caused sales, pair it with a test. For a big channel decision, I'd lean on a geo test.

When does a custom approach make sense?

When you need to combine sources unique to your business, investigate at the order or journey level, or keep a measurement process a standard dashboard can't handle. Start with the decision, your data definitions, and who owns it.

Do we need an MMM first?

No. You can use MTA plus incrementality testing without one.

Got a demo on the calendar? Bring the questions to ask an attribution vendor.

Want to see which tools fit each method? Here’s a table of tools by method.