There's a question almost every customer asks at some point during their MMM journey, often phrased a few different ways: "should we break our Search line into branded and non-brand?" or "should we add X to the model?" or "why isn't Y in there?"

These are all the same question underneath: what channels should be in my MMM, and how should they be structured to actually help me make decisions?

This article is the strategic guide to that question. It walks through how to build your initial channel structure and then through the most common decisions you'll face as your MMM evolves. For deep practical detail on any one decision, this article links out to the dedicated guide.

Start with strategy, not data

Before you ask "what data do we have?", ask four questions about what you want the MMM to do for you.

1. What decisions do you actually want this model to inform? Budget reallocation across channels? Approval for incremental investment in a new channel? Defending channel-level spend internally? Setting priors for incrementality experiments? The answer shapes how the channels in the MMM should be structured. A model informing a $2M annual reallocation across 8 paid channels needs different granularity than one that's primarily there to validate that brand investment is paying back.

2. How do you think about your marketing efforts? By funnel stage (top-of-funnel, mid-funnel, bottom-of-funnel)? By content type (brand storytelling vs. direct response)? By brand vs. performance? The way your team conceptualizes marketing usually maps closely to how the MMM should be structured, because how you think about marketing tends to drive how you plan and budget.

3. What level does your team actually manage at? If your channel managers run Meta as a single budget, modeling Meta as a single channel matches your operating reality. If they manage Meta Prospecting and Meta Retargeting as two separate budgets with separate KPIs, the MMM should reflect that, assuming the data conditions are met. This is closely related to question 2, but in some cases the budget split tells you something the conceptual framing doesn't: a separate budget owner is a strong signal it should be a separate channel in the model.

4. What channels would you actually act on if you knew their incrementality? A channel you don't plan to scale or cut isn't generating decision value from being in the model. It might still be worth including for completeness, but don't confuse "in the MMM" with "actively informing strategy."

The strategic foundation here connects directly to your overall paid/earned/owned and brand vs. performance strategy. For that broader portfolio framing, see Channel Mix Strategy & Portfolio Management.

Building your initial MMM channel structure

When you're building your first MMM, the channel structure doesn't have to mirror your reporting org chart exactly, but it should reflect how you actually plan and act on marketing. Five dimensions tend to drive the right splits.

1. Channel. The most natural starting point and the default first cut for most MMMs. Google, Meta, LinkedIn, TikTok, podcast partnerships, affiliate programs, OOH, the platforms or vendors you're spending against. Beyond the channel-level cut, the question becomes which additional dimensions are worth layering in.

2. Type of marketing: brand vs. direct response. Many companies have a functional split between brand and DR teams, with different goals (awareness vs. conversion), different KPIs, and different ad formats. If your team operates this way, modeling brand and DR as separate channels (Linear TV Brand vs. Linear TV DR, for example) usually pays off, because it tells the model these are doing genuinely different work even when they share a platform.

3. Budget ownership. Sometimes the cleanest split is just who owns the budget. If your influencer program is run by a separate team with separate accountability from your paid social team, those are functionally different channels in your business even if both technically run on Meta. The MMM should reflect that ownership split.

4. Funnel stage. TOFU / MOFU / BOFU, awareness / consideration / conversion, or whatever taxonomy your team uses. If you plan campaigns at funnel-stage granularity and budget that way, the MMM probably should too.

5. Content type. Same channel, same budget, different creative can drive very different results. The format × purpose matrix below is a useful way to think about content as a portfolio:

For a deeper look at content type as a strategic dimension, see Marketing Content Types: A Strategic Guide.

Three other factors apply across all of these dimensions:

1. Granularity. A useful rule of thumb is one channel per 10 weeks of data. With 100 weeks of history, that's about 10 channels total. Beyond that, the model starts finding patterns that aren't real.

2. Spend volume. Each channel ideally needs at least 2-5% of total spend (and at minimum about 1%) to give the model usable signal. Smaller channels rarely produce trustworthy estimates and add complexity without insight.

3. Actionability. A channel you wouldn't act on isn't earning its complexity. The MMM is a decision tool, not a reporting deck. Include a channel because you'll do something with the answer, not for completeness.

Two example structures

Disclaimer: These are just illustrative examples AND have more breakouts than we’d suggest. The example is here to help illustrated the types of breakouts to consider.

Example A - This structure splits brand vs. DR within TV, video vs. static within Meta, and breaks out device platform within app campaigns:

• Linear TV DR / Linear TV Brand

• Streaming TV DR / Streaming TV Brand

• Google Branded Search / Non-brand / Competitor / PMax

• YouTube / YouTube Brand

• Google Static Discovery

• Apple Branded Search

• UAC iOS / UAC Android

• Meta Video DR / Meta Static DR / Meta Video Brand / Meta Static Brand

• TikTok / TikTok Brand

• Snapchat / Snapchat Brand

• Direct Mail, OOH, Podcast, Affiliates

Example B - This structure organizes around funnel stage rather than than brand vs. DR. Same idea, different organizing principle: the MMM mirrors how the team plans and acts.

• Google Branded / Non-brand / Competitor / PMax

• Bing Branded / Non-brand

• YouTube / YouTube CTV

• LinkedIn TOFU / MOFU / BOFU

• Meta TOFU / MOFU / BOFU

• SEO, Webinars

Both are valid structures. The right one for your business is the one that mirrors how you make decisions.

The most common decisions to revisit

Once your initial MMM is in place, a handful of questions tend to come up over time as your business evolves and your team matures with the model.

Each section below summarizes the framework. Use the linked deep guide when you're ready to apply it.

Adding a channel you're already running

When we say an existing channel, we mean a channel that has been active in your business for some time but wasn't in your initial MMM. It might have started small, been owned by a different team (events, brand), or have had data quality issues at the time the model was built. Now you want to bring it in.

A common example is an account-based or display platform such as Demandbase or 6sense. If that spend was running but is absent from the model, the right question usually is not simply "can we add it?" but whether you have complete historical data, enough variation and volume, and a stable enough channel definition for the MMM to learn from it reliably.

The most important factor is data completeness: do you have spend and outcome data covering the full MMM lookback period? If more than 10% of the historical period is missing while the channel was active, the model can misinterpret the gap as the channel being newly launched, leading to under-attribution. Beyond data completeness, you'll evaluate three other factors:

• Whether the channel's strategy and execution have been consistent over the period (were there major strategy shifts that will lead to very different results across the historical period)

• Whether it's material to your business (is the volume sufficient and meaningful)

• Whether you'd act on what the MMM tells you about it (would you action against MMM insights)

The cleanest path when historical data is patchy is to run a controlled incrementality test on the channel and use the test result as a prior to inform the model. That gives you statistical grounding rather than relying on noisy or incomplete history.

For the full evaluation framework (the four pillars, the four mitigation paths, and the recommended approach), see How to add an existing channel to your MMM.

Adding a channel you just launched

A new channel here means truly new: no prior marketing activity, no historical data. When a brand new channel launches, the natural instinct is to add it to the MMM right away to measure its impact. The reality is that an MMM works by examining historical patterns. A channel with four weeks of data has the weakest signal in the model and produces estimates that will keep changing as more data lands.

The recommendation: wait 3-4 months before adding a new channel to the MMM. By then, performance has stabilized past the early learning phase, spend has started oscillating in ways the model can read, and the resulting estimates are far more reliable.

If you do want to add it sooner, that's possible, but go in with the right expectations: existing channels' contributions will shift (you're dividing the same pie into more slices), uncertainty ranges on the new channel will be wide, and the first 1-2 monthly refreshes will rewrite the channel's history as more data arrives. Pair the early add with an incrementality test if the channel is material, since that gives the model a grounded prior to anchor on.

For the full timeline, expectations-setting language, and best practices for launching with the MMM in mind, see How to add a new marketing channel to your MMM.

How granular should my channels be?

Most MMMs are built at the channel-and-platform level (Meta, Google Search, LinkedIn, CTV). The question of granularity is whether to go further: Branded Search vs. Non-brand Search vs. Competitor Search; Meta Prospecting vs. Lookalike vs. Retargeting; YouTube Brand vs. YouTube DR.

Three constraints decide whether breaking out is worth it.

• The first is statistical. A useful rule of thumb is that you can support roughly one channel for every 10 weeks of data. With 100 weeks of history, that's about 10 channels total. Adding more risks the model finding false correlations or producing unstable estimates that swing month to month.

• The second is material. A sub-channel needs enough spend (typically a few percent of total) and enough variation over time to give the model a usable signal. Small or flat-spend sub-channels rarely produce useful estimates.

• The third is actionable. If you'd plan, optimize, or reallocate at the sub-channel level, breaking out is valuable. If you wouldn't, the added complexity isn't earning its keep.

Branded Search is the most common case where granularity pays off, because it almost always shows lower incrementality in the MMM than internal attribution implies, and that delta is often the single most valuable insight the model produces.

For the full decision framework, the four-dimension evaluation, and the do/don't checklists, see How granular should your channels be in your MMM?.

A practical example: should you break out CTV retargeting by platform in the readout?

A common follow-up question after a CTV or video campaign is whether the team should break performance out by platform — for example YouTube, LinkedIn, Adobe, and Roku — rather than keeping one combined CTV retargeting line in the readout. The right answer is usually: only if the breakout changes a decision and the data can support it cleanly.

1. Start with the decision, not the curiosity. If the team is prepared to shift budget, creative, audience strategy, or vendor mix at the platform level, a breakout can be worthwhile. If the readout would still lead to the same next step regardless of which platform looked best, the extra granularity may add more noise than insight.

2. Check whether each platform has enough signal. A platform-level cut is most useful when each platform has meaningful spend, enough time in market, and enough variation to produce a readable pattern. If one or two platforms are small or only active briefly, the comparison can become unstable and over-interpreted.

3. Separate reporting detail from modeling structure. It can be perfectly reasonable to show a platform-level breakout in the readout while still keeping a broader grouped channel in the MMM itself. Reporting can answer “what happened within this campaign mix?” even when the model would become less stable if every platform were split into its own standalone channel.

4. Make sure the platforms are actually comparable. A YouTube placement, a LinkedIn audience buy, a Roku CTV execution, and an Adobe-managed retargeting program may differ in format, audience, creative role, and funnel job. Before comparing them side by side, check that you are not mixing demand-creation tactics with demand-capture or retargeting tactics and calling it one like-for-like performance table.

5. Use the breakout to generate the next question. A platform-level readout is usually most valuable when it points to a follow-up action: simplify the mix, scale the strongest platform, test a holdout or pulse-up on the leading platform, or regroup platforms that are too small to interpret individually.

A practical way to explain this internally is: break out by platform when it improves the decision, not just when it increases the number of rows in the report. If the platform-level view will change how you allocate budget or design the next test, it is worth doing. If not, keeping the channels grouped is often the more disciplined choice.

Channels with no media spend

SEO, organic social, email (when not paid), community, and brand-owned channels create real value, but they don't behave like paid channels in an MMM. The model is built on the relationship between spend, impressions, and conversions. Without a spend variable, the model can read correlations between organic activity and conversions, but it can't tell you "if you put $X more here, you'd get Y in return."

Whether to include them depends on what you're trying to accomplish.

If you want a more accurate baseline (so paid channels don't absorb credit for SEO or email-driven conversions), and the data is clean, including them often makes the model more correct.

If your goal is to optimize the organic channel through MMM, this is usually the wrong tool. SEO and email are driven by content quality, ranking, frequency, or CRM strategy, not spend dials. Use channel-specific analytics or experimentation instead.

If you simply want completeness in reporting, that's situational. It's worth doing if the data is consistent and stable. It's not worth doing if including it adds noise without insight.

For the full decision framework and how to interpret zero-spend channels in the model output, see How to think about zero-spend channels in your MMM.

Should I include sales data in my MMM?

Some businesses want to incorporate direct sales activity (telesales, account executives, retail field sales, partner channels) into the MMM alongside paid marketing. Whether that's the right move depends on a few things: whether sales activity has measurable inputs the model can correlate against (calls made, accounts touched, demos delivered, etc.), whether those inputs vary meaningfully over time, and whether you'd actually act on what the MMM tells you about sales the same way you'd act on what it tells you about paid channels.

For most businesses where sales is a real driver of revenue, including it makes the model more complete and prevents paid channels from absorbing credit for sales-driven outcomes. For businesses where sales activity is relatively flat or hard to measure consistently, it can add complexity without insight.

For the full decision framework and how sales data behaves differently than paid channel data in the model, see Should You Include Sales Data in Your Marketing Mix Model?.

How these decisions connect

A useful philosophy for the whole channel structure: include what you can act on, in the form your team manages, with the data quality the model needs.

Three principles flow from that:

1. Don't model what you wouldn't act on. The MMM should reflect the decisions you're trying to make, not your full marketing reporting deck.

2. Match the structure to your operating reality. If a sub-channel has its own budget, owner, and strategic question, model it that way (data permitting). If it's managed inside a larger bucket, model the bucket.

3. Earn granularity. Start broad, build a stable foundation, validate the model is healthy, then introduce sub-channels deliberately as the data and the business support it.

When to revisit your channel structure

Channel structure isn't a one-time decision. A few signals that should trigger a review:

1. Your business has shifted, and channels that were small are now meaningful (or vice versa). The MMM should follow the strategy.

2. A channel manager or new stakeholder is in the room and the current model doesn't reflect their operating view. That's a good prompt to evaluate whether breaking out a sub-channel is now worth the modeling complexity.

3. You've completed an incrementality experiment that contradicts the current MMM read on a channel. The experiment becomes a calibration input, and you may want to reconsider the channel's structure as part of integrating the new prior.

4. You're building toward a budget reallocation decision and the current granularity isn't sharp enough to inform it. Add granularity targeted at the specific decision, not as a blanket pass.

Want to learn more?

• How to add an existing channel to your MMM

• How to add a new marketing channel to your MMM

• How granular should your channels be in your MMM?

• How to think about zero-spend channels in your MMM

• Should You Include Sales Data in Your Marketing Mix Model?

• Marketing Content Types: A Strategic Guide

• Channel Mix Strategy & Portfolio Management

• How MMM and Incrementality Testing Work Together

Want to learn more about marketing mix modeling? Check out other Paramark marketing mix modeling articles here