Many B2B companies reach a point in their MMM journey where the same question comes up: should we include sales activity alongside our marketing channels, or keep it separate? It's a reasonable thing to wonder — sales and marketing often operate as distinct functions with different data owners, different metrics, and different levels of enthusiasm for cross-functional transparency.

The short answer, for most B2B businesses, is yes — but the reasoning matters, and there are cases where it doesn't make sense. Here's a practical framework for thinking it through.

Why this question matters

If your sales team drives a meaningful portion of revenue and you don't include their activity in your MMM, that activity doesn't disappear from the model — it gets absorbed into the baseline. The baseline is meant to represent organic demand: the revenue you'd generate regardless of any marketing or sales effort. When sales activity sits there instead, a few things go wrong:

• Your baseline becomes unstable. It rises and falls with sales pushes, making it hard to interpret what "organic" actually means for your business.

• Marketing channels get mis-credited. When sales runs an aggressive outbound push in the same period that a paid channel is active, the model may credit the paid channel for the lift that sales actually drove — or penalize it for a dip that followed when the push ended.

• Seasonality patterns get distorted. The model may interpret a Q4 sales surge as seasonality, when it was actually a sales team hitting year-end quotas.

In short: an MMM that ignores sales isn't wrong per se, but it's incomplete in a way that affects every output the model produces.

The case for including sales data

A cleaner, more accurate baseline. When sales is modeled explicitly, it's removed from the baseline. What remains is a truer picture of demand that exists independent of any GTM effort — which is what the baseline is actually supposed to measure.

Fairer attribution for marketing. Once the model can distinguish sales-driven revenue from marketing-driven revenue, marketing channels get credited (and penalized) only for what they actually caused. This tends to make the model's channel-level recommendations significantly more reliable.

A complete GTM picture. Leadership — especially CFOs and CMOs accountable for total revenue — benefit from seeing how both functions contribute. An MMM that only shows marketing's half of the picture limits the strategic conversations it can support.

Full funnel visibility. Including sales can reveal whether marketing efforts are being amplified or dampened by downstream sales motions. For example, if marketing generates strong top-of-funnel demand but sales capacity is a bottleneck, the model may start to show diminishing returns on marketing spend that are actually a sales capacity problem in disguise.

The case against (and when it applies)

Including sales is not always the right call. There are a few situations where we'd advise against it or recommend waiting:

• Sales represents less than 2–5% of total GTM effort. Below this threshold, the variable is unlikely to be detectable by the model and may add noise without adding signal.

• You have less than 18 months of consistent sales data. The model's standard training window is roughly two years. Incomplete data leads to unreliable attribution for that channel.

• Sales data quality is poor. If the data can't be provided reliably on a monthly basis — or requires significant manual effort to extract — the operational overhead often outweighs the modeling benefit, at least until the data foundation improves.

• Your model already has a high channel count relative to your data history. A useful rule of thumb: you want roughly 10 weeks of data for every channel in the model. Adding sales to an already-complex model can increase false correlation risk.

• You're a B2C or DTC business. A traditional sales function typically doesn't exist in these models, and this consideration generally doesn't apply — though outbound retention or CX programs can sometimes be treated similarly.

How sales data is modeled

There are two primary approaches, and the right one depends on your goals:

As a distinct channel. Sales is treated like any other marketing channel, with spend (cost of the sales function) and impressions (activity volume) included directly. This gives full attribution to sales alongside your media channels and allows for response curve analysis — understanding at what point adding more sales headcount or activity stops producing efficient growth.

As a non-media control variable. Sales activity is included to control for its effect on the model without directly attributing revenue to it. Think of this as a "sales halo" approach — the model accounts for sales' presence, which improves baseline stability and reduces false attribution to marketing, without producing a standalone sales ROI number. This approach is less custom than it sounds and is a good middle ground when the primary goal is improving model accuracy rather than measuring sales performance.

What data do you need?

To include sales in your MMM, you'll typically need three things:

A time series of sales activity — daily or weekly aggregate data, covering at least 18–24 months of history to match the model's training window.

Geographic breakdown — if your model is hierarchical (broken out by region or market), the sales data needs to match that structure. State or region-level data is usually sufficient.

A spend figure — either fully loaded costs (salaries, commissions, tools) or a variable cost proxy such as commissions or outbound tooling spend only. Both work; the choice affects how cost-per is interpreted in the output.

An "impression" metric — the volume input that tells the model how much sales activity occurred — the most common options are:

• Outbound call volume, emails sent, or LinkedIn touches (best option when available)

• Completed demos or sales-qualified meetings

• Headcount, as a last resort when activity-level data isn't available

Most of this data lives in your CRM (Salesforce, HubSpot), sales engagement platform (Outreach, Salesloft, Apollo), or HRIS if you're working from headcount.

A note on sales cycle length

This is an often-overlooked dimension that has a meaningful impact on model quality. Unlike digital advertising — where the effect typically decays within days or weeks — sales activity often has a much longer lag before it converts to revenue. A two-week median sales cycle is modeled very differently from a six-month enterprise cycle.

When adding sales data, it's worth providing Paramark with a rough estimate of your average sales cycle length. This allows our Data Science team to apply appropriate lag and adstock structures so the model correctly attributes revenue to the period when the sales activity actually occurred, rather than when the deal closed.

What changes when you add sales to the model

It's worth setting expectations before running the updated model, because the outputs will look different from the prior version. Specifically:

Your baseline will decrease. Revenue that was previously sitting in the baseline as "unexplained" will now be attributed to sales. This is a good thing — it means the baseline is now more accurately representing true organic demand.

Some marketing channels will see attribution shifts. Channels that were over-credited (because sales was doing some of the work in the same periods) will see their share of effect adjust downward. This doesn't mean those channels performed worse — it means the model is now more precise.

Sales will appear with its own contribution and cost-per. You'll see sales represented with its estimated share of effect and an efficiency range, which enables apples-to-apples comparison with your marketing channels.

Response curves become available for sales. Like any channel, the model can now show diminishing returns for sales activity — helping identify whether you're under- or over-invested in the sales function relative to its marginal return.

Adding sales data mid-engagement

Sometimes the question of including sales data comes up not at the start of an MMM engagement, but partway through — often triggered by a model output that doesn't quite add up, or a stakeholder who asks why a particular quarter looks the way it does.

A few things worth knowing if this applies to you:

Adding sales is not a full rebuild. The core model structure stays the same. Adding a sales variable is a targeted change that requires rerunning the model with the new input, not starting from scratch. The scope of work is generally much smaller than customers expect.

The model will shift, and that's the point. When sales is added mid-engagement, prior outputs will change. This is expected and is a sign the model is improving, not a sign that previous work was wrong. The prior model was working with the information available at the time; the updated model is more complete.

Stakeholder communication matters. If your leadership has been making decisions based on the prior model, it's worth getting ahead of the changes before they see the updated outputs. Framing the shift as a model improvement — rather than a correction — tends to land better and is also more accurate.

If sales data isn't immediately available, consider flagging known periods of significant sales activity in your insights documentation so stakeholders understand the caveat, and treat the next model refresh as the milestone for formal inclusion once the data is in better shape.

What if sales-driven revenue/leads were removed from the output metric entirely?

The scenarios above assume that sales activity was simply never included as a channel variable — meaning sales-driven revenue was still present in the output metric, just sitting unaccounted for in the baseline. There's a more significant variant worth understanding: some businesses strip sales-attributed revenue out of the output metric before modeling, intentionally excluding it to try to isolate the "marketing view" of the business.

This is a well-intentioned approach, but it creates a more serious set of problems than the standard omission.

The output metric no longer represents your business. When sales-driven revenue is removed, the model is trained on a partial picture of total output. Every relationship it learns — between marketing activity and revenue, between seasonality and demand, between baseline and organic growth — is based on an incomplete signal. The model isn't just missing a variable; it's been trained on data that doesn't reflect how your business actually works.

The total output grows when you correct it. In the standard scenario (sales missing as a variable but present in the output metric), adding sales data redistributes attribution — the total pie stays the same, channels get recalibrated. When excluded revenue is restored, the pie itself gets larger. All channels are now contributing to a bigger total, which means absolute contributions shift in addition to relative shares. This makes the transition more disorienting for stakeholders and requires more proactive communication.

Marketing channel attribution is more deeply biased. Because marketing channels were active during the same periods that sales-driven revenue was being excluded, the model has been learning a distorted relationship between marketing inputs and outputs. Periods where both sales and marketing were running hard may look artificially quiet in the output — which means some channels may be systematically underestimated, and others may have developed inflated attribution to compensate for the missing signal. The direction and magnitude of the bias is difficult to predict without rerunning the model.

Response curves and budget recommendations are unreliable. The diminishing returns curves the model has built for each channel are based on a constrained output ceiling. If the true output is materially larger, the model may be calling diminishing returns earlier than they actually occur — which can lead to under-investment recommendations across the board.

Seasonality patterns will shift. If sales activity peaks in certain periods (Q4 year-end pushes, for example) and that revenue was excluded, the model's seasonal baseline has been calibrated against a distorted view of those periods. Restoring the full output metric will recalibrate seasonality in ways that affect how every channel is interpreted across the year.

The practical implication is that correcting this issue — restoring the full output metric and adding sales as a channel variable simultaneously — is a more significant model change than a standard sales inclusion. Expect larger shifts across all outputs, and plan stakeholder communication accordingly. The framing remains the same: this is the model seeing your full business for the first time, not a sign that previous work was wrong.

One additional question worth investigating before the correction: how was the revenue exclusion done? If it was based on CRM attribution (e.g., any deal marked as outbound-sourced in Salesforce was removed), that attribution carries its own biases. Last-touch and rep-assigned source fields are often unreliable, which means some marketing-influenced revenue may have been excluded, and some sales-influenced revenue may have remained. Understanding the exclusion methodology helps set realistic expectations for how much the model will shift and in which direction.

Impact on incrementality testing

Including sales in the MMM opens up experiment opportunities that weren't previously available. If the model shows sales has strong diminishing returns, a geo-based holdout can be designed to validate whether pulling back on sales activity in a test market produces the effect the model predicts. Conversely, if the model suggests marketing and sales have synergy effects — where both together produce more than the sum of their parts — a joint pulse-up experiment can be designed to test and quantify that interaction.

This is worth keeping in mind as you build out your experiment roadmap. The richer the model's inputs, the more precisely targeted the experiments it can support.

The bottom line

For most B2B businesses with a meaningful sales motion and at least 18 months of reliable activity data, including sales in your MMM will make it more accurate, more complete, and more useful for strategic decision-making. The short-term adjustment of shifting attribution is a small price for a model that actually reflects how your business works.

If you're unsure whether your situation meets the qualifying criteria, or you're trying to determine which approach (distinct channel vs. non-media variable) makes more sense for your business, that's exactly the kind of question worth working through with your growth advisor before the next model refresh.

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