Why don't your attribution tools agree with each other? Because each one measures something different, in its own way. Like a blind man and a deaf man telling the same story about a bar fight they witnessed. Both stories capture different information.

Quick answer: Ad platforms, attribution tools and lift tests each tell a different story about the same sales. So stop hunting for the one right tool. Decide how you'll compare methods on purpose, and whether to buy a platform, build in-house or hire a partner. Then check what each choice costs as you grow and what you keep if you leave.

Disclosure: Brinker Analytics sells MMM, geo-test and MTA builds and consulting. That's a commercial interest in this topic. Vendor details below come only from each vendor's own site.

Why does every dashboard tell a different story?

Picture this: Meta says it drove a big share of last month's sales. Google says it did too. Your email tool takes credit for some of the same orders. Add them up and you get more sales than Shopify ever recorded!

Sound familiar?

Each source sees its own slice. Each one counts a sale in its own way, with its own window and its own idea of what a "touch" is. So the totals never line up.

I've covered the reconciling part in why your marketing numbers disagree, and the model side in why MTA and MMM don't agree. This piece is about the next decision. Once you accept that the numbers won't match, what do you buy, build or hire to make sense of them?

But first, which way do the numbers tend to miss?

Which way do platforms and holdouts miss?

Pull the platform-reported return for one channel. Then put it next to that channel's last lift test (a holdout, where you turn ads off in some places and compare).

  1. Note the gap between the two.
  2. Note where the channel sits in the funnel.
  3. Do the same for a second channel.

Now look at the pattern. My view: it varies wildly. Often, though, lower-funnel strategies swing higher in platform than in a holdout. And upper-funnel strategies swing lower in platform than in a holdout.

In my experience, that means a dashboard can make your retargeting look great and your awareness spend look weak. The holdout may say the reverse.

In a 2019 Marketing Science study of 15 U.S. Facebook experiments, observational methods "often fail to produce the same effects as the randomized experiments."

And don’t forget that in this case, the “observational method” is click-based reporting. Your clicks may be lying to you!

That's why I don't trust one view on its own.

What are the four ways to get measurement?

Most comparison pages line up brands in a feature grid. But plans and prices change, and a grid skips the bigger question: what kind of measurement are you buying?

There are four broad ways to do it.

1. An all-in-one suite. One platform that offers attribution, an MMM (marketing mix model) and incrementality testing. Northbeam, Triple Whale and Rockerbox are examples as of October 2026. Each says on its own site that test results feed back into its other models. Northbeam's docs say "incrementality results calibrate MMM's estimates." Rockerbox says test results "can calibrate your MTA model and inform MMM as priors." Triple Whale says that with Compass, which it lists as included on Enterprise, "your three measurement signals (MTA, MMM, Incrementality) continuously calibrate against each other." Good for: having it all in one place. Check: what you'll pay as you grow, and how long they lock you in.

2. A testing-first platform. These lead with lift tests and experiments. Results are often used to adjust other reporting. Good for: a brand whose biggest question is "is this channel actually adding sales?" Check: whether you can see the test design, and whether you’re allowed to do more than one kind of test design.

3. Build it in-house with open source. As of October 2026, the core tools are free to use under open-source licenses. For an MMM there's Meridian from Google (Apache 2.0), Robyn from Meta (MIT) and PyMC-Marketing (Apache 2.0). For geo tests there's GeoLift from Meta (MIT) and GeoX from Google. You own the code. You also own the upkeep, so you need someone on your team who can run it and explain it.

4. A flexible, multi-model partner. A consultant or contractor who builds the models with you, in your own systems. More on when that fits, and when it doesn't, in the last section.

Before you pick one, look at how each one prices.

How does the pricing model shape your cost?

Price follows a basis.

Some tools price on your revenue. Some price on your ad spend or on how much data you send them. And a few don't publish prices at all. Here's how the three suites above describe it, as of October 2026:

So what does that mean for you? My view: These tools are usually too expensive for lower-spending brands. And when price rises with volume, higher-spending brands get penalized too.

It's worth saying that some costs really do grow with volume. The first-party data lake that powers an MTA (multi-touch attribution) does scale with your traffic and orders. But geo tests and MMMs don't have to.

Ask what your bill looks like after a great year, not just today.

Who owns the model, code and data if you leave?

Before you sign, find out three things:

Here's the general pattern as of October 2026. Many platform plans are annual commitments. Triple Whale says its paid plans "are 12-month subscriptions." Northbeam says Starter is "month-to-month," while Professional and Enterprise are "annual terms."

Data export is easier to find. As of October 2026, Rockerbox says it works with "BigQuery, Redshift, and Snowflake." Triple Whale lists data warehouse export as an add-on.

The model is a different story. As of October 2026, none of the three answers one question on its pricing or main product pages. If you leave, do you keep the model or the code? Ask them directly.

With open-source tools, the code is already yours. With a partner, ask for a clear off-ramp in writing: the code, plus the training to run it without them.

One model or many?

I think of measurement models like weather models.

A forecaster can look at more than one weather model. When the models agree, you can feel better about the forecast. When they split, that's where you look harder.

Measurement can work the same way. Run more than one model on purpose: more than one MMM, more than one geo design, more than one attribution view. Where they disagree, you've found a hypothesis. Then settle it with a lift test.

Here's what tends to happen instead. Vendors typically build around a house methodology and blend their models into a single answer; they rarely show you where methods disagree. You get ONE number. It's clean. But you may not see the disagreement behind it.

Many dashboards let you switch between click rules: last click, first click, U-shaped, and so on. 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).

And when you ask for something outside the house method? It's very common to hear "no, we don't do that here." It's much less common to hear "we can build it," or "why not look at both approaches?"

Once a test settles a question, feed the result back in. That's what I mean by causal attribution: attribution calibrated to lift-test results. For more, see why use more than one MMM and what an incrementality test can calibrate in an MMM.

Why own the data foundation?

List every source that touches a sale: your store, each ad platform, email and SMS, and analytics.

  1. Pipe each one into a warehouse your brand owns, with an ETL (a tool that pulls the data on a schedule and loads it).
  2. Keep the raw data, not just the summaries.
  3. Point every tool and partner at that same warehouse.

Now a vendor change doesn't wipe out your history. A new model can start from the same data as the old one. And you can compare methods apples-to-apples.

It's less exciting than a dashboard.

It matters more. See marketing data foundations for DTC brands.

Whoever owns the data owns your options.

What should you ask any vendor or consultant?

So you've booked a few demos. Everyone's slides look great.

Use the same short list for every one of them, including a consultant like me. These are about the buying approach. For questions about how an attribution model works, use 10 questions to ask an attribution vendor. For the fuller path through choosing a tool, see how to choose an ecommerce attribution tool.

  1. Does your price scale with revenue, ad spend or pageviews?
  2. What's the contract term?
  3. Do I keep the model, code and data if I leave?
  4. Can I see the test design?
  5. Can you run more than one test design or MMM, and show me where they disagree?

Write the answers down side by side. The gaps tell you a lot.

So where does a partner like Brinker Analytics fit?

Where Brinker Analytics fits, and when it doesn't

Disclosure, again: Brinker Analytics sells MMM, geo-test and MTA builds and consulting. So read this section as my pitch, not a neutral review.

Say you want a lift test this quarter and a second MMM to check the first one. But you don't want a new annual platform, and you want your team to run it next year. That's the kind of work I do. It's billed as a flat fee that doesn't rise with your revenue or ad spend.

Here's what that can look like:

When does it make sense to hire someone? When you want a clear off-ramp to bringing the practice in-house. When you want heavy customization, or a blend of approaches the bigger platforms don't offer as standard. Or when you want a limited or one-off engagement, not an annual or month-to-month agreement.

And when doesn't it fit? If you want a vendor-run platform on an ongoing plan, and none of those three things matter much to you, a platform may suit you better. A good tool plus your own lift tests is a perfectly valid answer.

Pick the setup that shows you where the numbers disagree, and lets you test it.

FAQs

Why don't my ad platforms, attribution tool and lift tests agree?

Each one measures a different thing in a different way. In my view, how far they miss varies wildly. Lower-funnel tactics often read higher in platform than in a holdout, and upper-funnel tactics often read lower.

Is there an ad spend level where I should buy an attribution tool?

No. There is no spend level to tool relationship. Look at how the price grows and what you keep if you leave.

How do attribution tools price?

It depends on the vendor. Some price on revenue (GMV), some on ad spend or data volume, and some don't publish prices. Check each vendor's own pricing page. The examples above were checked in October 2026.

Can I build measurement in-house?

Yes. As of October 2026, open-source MMM and geo-test tools are free to use. Each has its own license. Meridian, Robyn, PyMC-Marketing and GeoLift are examples. You'll also need a warehouse you own and someone who can run and explain the models.

When should I hire a consultant instead of buying a platform?

Consider it when you want a clear off-ramp to doing it in-house, heavy customization or a blend of methods, or a limited or one-off engagement.

What is causal attribution?

It's attribution calibrated to lift-test results. The test tells you what a channel actually added. Then you use that result to adjust how the day-to-day attribution view credits that channel, instead of relying on clicks or paths alone.

Need help? Speak with me here.

Sources: Gordon et al., Marketing Science 2019, https://doi.org/10.1287/mksc.2018.1135. Vendor details checked on October 18, 2026 and October 19, 2026.

Next, why marketing numbers disagree to reconcile the inputs before interpreting the result, what to ask before buying an attribution tool to evaluate the stack beyond the demo, why multiple MMM perspectives help to pressure-test a consequential model recommendation, or five decisions before building a marketing data foundation to keep a warehouse project connected to a real question.