Stack Three Attribution Models

Never let one number, blended, modeled, or multi-touch, run the budget alone.

Every attribution model lies to you in a different direction. Last-click over-credits whichever channel closes the sale. Marketing mix modeling smooths away the ad-level decisions you need to make this week. Blended MER tells you the business is fine, or is not, and nothing about why. Attribution consultant Curtis Howland built his practice fixing DTC brands that had all three running at once and still could not answer "do I cut this campaign," because they trusted whichever number happened to be open on screen. The fix is not picking the right model. It is running three, at three altitudes, and only ever asking each one the question it was built to answer.

What to do: Run blended MER or nCAC as the board-level number nothing else can override. One level down, run marketing mix modeling (Google Meridian or Meta Robyn, both free) alongside a post-purchase survey to split budget between channels, including the ones a pixel never sees. At the base, run a multi-touch tool like Triple Whale or Northbeam to decide which individual ads live or die.

Why it works: Each tier corrects the blind spot of the tier below it, so no single tool gets to run the budget past what it can actually see.

Example: Curtis Howland runs paid media for DTC brands at his firm Misfit Marketing, whose own site puts the range at scaling brands from $5M to $50M a month. He calls this exact structure the Christmas Tree framework: the blended target as the star at the top, marketing mix modeling and surveys as the branches, multi-touch tools as the base, built after auditing brands that were tracking everything and trusting none of it.

Walk it through

Here is the stack in the order it actually gets built, using tools that are real and checkable right now, in July 2026.

1. Lock the top number first.

Blended MER is total revenue over total ad spend. nCAC is total ad spend over new customers. Calculate either one weekly in a plain spreadsheet, no software required. This is the number for the board deck or the investor update, and it cannot be gamed by shuffling credit between channels, because it never touches channels at all.

2. Add the middle tier once you have spend history to model.

Marketing mix modeling wants a couple of years of clean channel-level spend and revenue before its output means anything, so this tier waits on data, not budget. Google's Meridian and Meta's Robyn are both open source, free, and actively maintained right now.

Neither needs a vendor contract to start. The real cost is someone who can wrangle two-plus years of spend data by channel and configure the model without breaking it. Pair whichever one you pick with a post-purchase survey (Fairing or KnoCommerce), asking "where did you first hear about us" rather than "how did you hear about us." First-touch is exactly what a last-click pixel misses, and together the survey and the model catch the channels no pixel sees at all: podcasts, out-of-home, word of mouth, dark social.

3. Add the base tier for daily ad decisions.

Multi-touch tools like Triple Whale or Northbeam connect straight to your ad accounts and your store data and give you ad-and-creative-level numbers you can act on the same day. This is the layer you actually open every morning. It is also the layer most likely to be locally wrong about total impact, so never let it overrule what the top two tiers say about a channel as a whole.

4. Reconcile once a month, not once a day.

Pull all three numbers into one view. If nCAC is climbing while every channel in your MTA tool reports flat ROAS, something you are not tracking is quietly getting worse: seasonality, competitive pressure, or a channel the pixel-based tier cannot see. That gap is information, not noise.

The read

• Blended is truth, not diagnosis. It tells you if the business works. It never tells you why, because it does not touch channels.
• MMM and surveys catch what a pixel can't. Podcasts, out-of-home, word of mouth, and privacy-blocked clicks all show up in blended revenue and nowhere in multi-touch attribution. This tier is how you find them.
• MTA is fast and locally wrong. It is the only tier granular enough for a same-day kill-or-scale call on one ad, and the one most prone to over-crediting whichever channel happened to touch last.

Steal it

Build this in the order you can afford, not the order that looks impressive on a slide. Below roughly $50k a month in spend, a spreadsheet carrying blended MER and nCAC plus a lightweight post-purchase survey is the whole stack you need, and bolting on MTA software earlier just buys you a second number to argue with yourself over. Past that point, add Triple Whale or Northbeam for the ad-level view, and once your spend history runs deep enough to model, layer in Meridian or Robyn for the channel-level one.

Defend it by writing down, in advance, which tier owns which decision. Blended owns "are we still profitable." MMM plus survey owns "which channel gets the next dollar." MTA owns "which ad gets killed this week." The day someone tries to use a same-day MTA number to overrule a quarter of blended nCAC data, the rule you wrote down in week one is what stops the wrong number from winning the room.

Gotchas

• MMM needs history you may not have yet. Meridian and Robyn both want real, clean channel-level data going back a couple of years before the output is trustworthy. Feed either one six noisy months and you get a confident, wrong answer.
• The three tiers will disagree, on purpose. They measure different things at different altitudes. Do not average them, and do not quietly pick whichever one confirms the media buy you already wanted to make.
• Someone has to own reconciliation. Three tools open in three tabs is not a system. It becomes one only once a single person is responsible for reading all three every month and writing down what changed and why.