AdPix marketing mix modeling: what it needs, how it works, how to read it

AdPix marketing mix modeling estimates how many conversions each paid channel actually caused, not how many it happened to be the last click on. You import weekly ad spend per channel, AdPix fits it against the conversions of the same weeks, and every channel gets an ROI and a marginal ROI with a 94% credible interval. The model starts from 26 weeks of data and is included from the Starter plan up. This page covers the data it needs, how to read what comes back, how lift tests sharpen it, and where it is the wrong tool.

Marketing mix modeling (MMM) is a statistical method that explains a business outcome, such as weekly sales, from spend on each marketing channel plus non-marketing factors like seasonality and trend, so you can estimate each channel’s incremental contribution and how it changes as spend rises.

What does AdPix marketing mix modeling answer?

AdPix MMM answers one budget question: how is the whole budget working, and where should the next unit of spend go? It reports what share of conversions each paid channel drives, where each channel starts to saturate, and how the weekly budget could be split (AdPix docs: run an MMM).

That is a different question from attribution. An attribution report tells you which touch came before a conversion, which is correlational and tends to over-credit channels that harvest demand that already existed, such as brand search and retargeting. The attribution models guide shows how much the credit moves between models on the same customer journey. MMM sidesteps user-level paths altogether: it compares spend and outcomes over time.

It is also different from a lift test. A lift test asks whether one specific change on one date worked. MMM covers the whole portfolio, week by week (AdPix docs: MMM or Lift).

Who should use AdPix MMM, and who should wait?

AdPix MMM fits teams that spend on two or more paid channels, have at least six months of tracked purchases, and are willing to change budgets deliberately. A store splitting money between Google Ads, a native ad network and paid social is the typical case.

It is a poor fit, for now, if any of these apply:

• You have under 26 weeks of tracked data. The run closes on the Insufficient tier and shows no ROI.
• Every channel spends the same every week. Flat spend is mathematically unidentifiable, and no amount of history fixes it (AdPix docs: import ad cost).
• You run one channel. A single lift test answers the question faster.
• There are no purchase events on the site. The model has nothing to explain and the run fails.

How does marketing mix modeling work in general?

Marketing mix models regress an outcome on spend per channel and control factors, but they bend two assumptions that a plain linear regression gets wrong. Google researchers put it this way in a 2017 paper: “Advertising usually has lag effects and diminishing returns, which are hard to capture using linear regression” (Jin et al., Google).

• Carryover (adstock). An ad seen this week can still produce a sale next week, so the effect of spend is spread over later weeks.
• Shape (saturation). Each extra unit of spend buys a little less than the one before. Plotted against spend, the response is a curve that flattens.

Modern open-source models work on the same ideas. Google describes its Meridian model as building in “the lagged effect of advertising and saturation” and producing “ROI estimates and credible intervals, response curves, and budget optimization recommendations” (Google for Developers). Meta’s Robyn guide says an MMM “will need a minimum of two years of historical weekly data”, and strongly recommends calibrating the model with experiment results (Meta Robyn).

Two consequences matter for any MMM, AdPix or not. First, the model needs variation: if a channel’s spend never moves, its effect cannot be separated from the baseline. Second, the honest output is a range, and with short histories that range is wide.

How does the AdPix marketing mix model work?

The AdPix model puts each channel’s weekly spend next to the conversions of the same weeks and splits every week into layers: a baseline (organic and direct demand, trend and seasonality) and one layer per paid channel. Each estimate carries a 94% credible interval, and the AdPix docs treat the width of that interval as a large part of the answer (AdPix docs: run an MMM).

What data does the AdPix model need?

Three inputs feed it:

AdPix does not read spend from ad platforms. The CSV needs three columns: date, channel and cost, with optional source, campaign, currency and revenue. One row per week is recommended, and daily rows are summed into weeks. Re-importing the same date, channel, source and campaign updates the row instead of duplicating it. One trap: an empty currency cell is stored as USD whatever your website’s currency is, so write the currency on every row (AdPix docs: import ad cost).

The seasonality calendar is built for the Iranian market. It includes Nowruz, Yalda, the school start in Mehr, Ramadan, Eid al-Fitr and Muharram, among others, and the model learns the real size and direction of each event from your own data (AdPix docs: seasonality calendar). Weeks on a website set to Tehran time end on Friday.

Before a channel can get an ROI, its spend has to move. A weekly-spend coefficient of variation (standard deviation divided by mean) below 0.30 marks the channel Not identified, and so does a variance inflation factor above 5, which flags channels whose budgets rise and fall together. A hypothetical channel spending $2,000 a week on average with a standard deviation of $500 has a coefficient of variation of 500 ÷ 2,000 = 0.25, below the bar.

How do you read the trust tier, ROI and mROI?

Read the trust tier first. Every finished run gets one of three:

The ROI table then shows, per channel, incremental conversions, share of conversions, ROI (incremental conversions per thousand of spend) and mROI (what the next thousand at today’s spend level would buy), each with its interval. The AdPix docs put the rule plainly: move budget on mROI, not on ROI.

A hypothetical example shows why. Two channels over the modelled period:

Channel A has the better average, but it is deep into its response curve: the next $1,000 there buys about 9 conversions, against about 14 on channel B. Shifting $1,000 from A to B gains roughly 14 − 9 = 5 conversions a week, but only if the intervals agree. If A’s mROI interval runs from 4 to 15 and B’s from 6 to 22, they overlap almost entirely and the docs say comparing the point estimates carries no information. The right move then is to calibrate, not to reallocate.

A wide interval means the data cannot pin the number down yet. It does not mean the effect is zero. AdPix only lets you conclude “no effect” when the interval is both narrow and close to zero.

How do lift tests calibrate the AdPix model?

This is the part of the AdPix design that changes how you plan. With less than two years of data, the model cannot tell on its own how much of your sales paid media causes, so a channel’s ROI stays withheld and reads “Needs a lift test” until something calibrates it. The docs sum it up as “A channel’s ROI arrives with calibration, not with time” (AdPix docs: run an MMM).

Calibration can come from three places, all listed in the “Calibration experiments” card at the bottom of the MMM page:

1. A lift test on one channel. Grade A or B results recorded with their full spend change are badged Recommended.
2. The Incrementality page, which finds budget changes in your cost data on its own: a change of at least 20% that holds for at least five of the next seven days is measured like a lift test (AdPix docs: incrementality).
3. A test you ran elsewhere, such as a geo holdout, entered by hand with its start and end dates, daily spend before and during, and the total lift with its standard deviation.

Nothing is applied automatically. You select “Use in model” on each row (or “Activate all recommended”), and the next run uses every active calibration. The docs’ advice for the most informative test: change a channel’s spend by at least 20% and hold it for two weeks. Cuts teach the model as much as increases. The general method behind those tests is in the incrementality testing guide.

What do the budget optimizer and planner do?

The budget optimizer turns the response curves into a proposed weekly split, either to maximize conversions for a fixed budget or to hit a target CPA. It is deliberately cautious: each channel moves only within 70% to 130% of its current weekly spend (the average of the last four weeks) and never above 1.2 times the highest weekly spend the model has seen. Channels with a withheld ROI stay locked.

Say a channel averaged $5,000 a week over the last four weeks, and its highest week ever was $5,200. The 70–130% band allows $3,500 to $6,500, but the 1.2 × $5,200 = $6,240 cap is lower, so the optimizer will propose at most $6,240. Anything the limits cannot place shows up as unallocated budget.

The budget planner goes further, with goals such as “Cut the budget, lose the least” or “Spend while the next conversion costs less than a target”. It tests each plan on every plausible version of the model and reports the “Chance it beats today”; confidence is high only when that share is at least 80%. Plans move at most 20% every two weeks. To check whether a new split pays for itself, run the expected numbers through the ROAS calculator; the ROAS formula guide explains the break-even line.

Which AdPix plan includes marketing mix modeling?

Marketing mix modeling is included from the Starter plan, at 499,000 toman a month, and it is not on the Free plan; a Free website sees an Upgrade plan button instead of the page. The Pro plan, at 2,990,000 toman a month, adds MMM for each channel’s share of sales and deeper channel analysis (AdPix pricing). Plans attach to one website, so one site can have MMM while another does not (AdPix docs: plans and entitlements).

Lift measurement, the calibration partner, sits on the same plans. AdPix is a web analytics platform that replaces Google Analytics 4: realtime, acquisition, engagement, conversion and e-commerce reports, plus attribution, lift testing and marketing mix modeling, with no data sampling and a free plan. The full plan map is in the AdPix features overview.

How do you run your first model in AdPix?

A first model takes an afternoon of data preparation and a few minutes of compute.

1. Export weekly spend from every ad platform for as far back as your tracking goes. Two years of past spend is worth as much the moment it lands as two years of waiting.
2. Download the template on the Cost & ROAS screen, fill date, channel, cost and currency on every row, and import it. Map your platform names to channel families on the Channels tab so thin channels can borrow strength from similar ones.
3. Check the readiness card at the top of the page. It shows per channel the weeks, spend variation and whether the channel is identified, plus an overall tier.
4. Open Marketing Mix (MMM) under Marketing measurement, select “Run model now”, pick Conversions (or Revenue if order values vary widely) and queue the run. It usually finishes in a few minutes.
5. Read the tier, then the ROI table. Note every channel marked “Needs a lift test”.
6. Plan one budget change per uncalibrated channel: at least 20% for two weeks, one channel at a time. Let Incrementality measure it, activate the result, and rerun.

After that, AdPix queues a run by itself once a day for any website with imported spend and no run in the last seven days.

What are the limits, and when should you use something else?

AdPix MMM is honest about what it does not know, which also means it often says less than you hoped. Plan around these limits:

One more check is worth knowing. The MMM page says the model starts from 26 weeks, while the readiness card on the cost import page lists 78 weeks as the bar to run (the same floor the docs give for the legacy engine). If the card blocks a run on your website, follow the card.

For the strategy around these numbers, from the metric chain to budget splits, read the performance marketing guide. If your next budget change is already planned, set it up as a lift test first: that single result is what turns a “Needs a lift test” dash into a number you can act on.

ROAS calculator

Return on ad spend and your break-even point. The free plan covers 100,000 events a month, no credit card. Setup takes under five minutes.

Start freeTry it free