Should subscription MMM be one model or two?

Start with two (separate acquisition and retention models). Move to unified later when the team has experience with the spec. Two-model approach is simpler to set up and easier to debug; unified is more powerful for capturing cross-effects but harder to maintain.

How do we LTV-weight in MMM?

Predict LTV at acquisition using early-signal indicators (plan type, source, engagement in first 14 days, geo). Use predicted LTV as the weight on each acquisition outcome. The MMM's acquisition coefficients then describe LTV contribution rather than headcount, which is the metric subscription finance actually cares about.

What is the right adstock for retention AI visibility?

Longer than for acquisition. Six to twelve weeks geometric half-life is typical. Retention-stage AI exposure (the user looking up help content, comparing alternatives, considering cancellation) operates on a longer cycle than acquisition-stage research.

Does this approach work for B2B SaaS?

Yes, with longer cycles. B2B SaaS acquisition cycles are typically months, retention cycles are typically annual. Adstock half-life should be longer (four to eight weeks for acquisition, twelve to twenty weeks for retention) and the outcome variable is often new ARR rather than new subscriber count.

How to Run MMM for a Subscription Business

The Subscription MMM Problem

Subscription businesses have two distinct marketing problems: acquiring new subscribers and retaining existing ones. Standard MMM was designed for transactional purchases and does not natively handle the dual-outcome structure. Adapting MMM for subscription requires either separate models or a unified spec that handles both.

Step 1: Pick the Modeling Approach

Two viable patterns. Separate models: one MMM for new acquisitions, one for retention or churn. Cleaner spec, more compute, double maintenance. Unified model: single MMM with new and retention outcomes as separate dependent variables in a vector. More complex spec, single refit, captures cross-effects between acquisition and retention marketing. Most subscription brands start with separate models and unify later.

Step 2: Use LTV-Weighted Outcomes

New subscribers are not all equal. A 12-month-LTV subscriber is worth more than a 3-month-LTV subscriber; the MMM should reflect this. Weight new acquisitions by predicted or realized LTV in the outcome variable. The MMM's coefficients then describe LTV contribution by channel, not just acquisition count.

For categories with long actual-LTV measurement windows, use a predicted LTV at acquisition (based on early signals) rather than waiting for full LTV to realize.

Step 3: Add Acquisition AI Visibility

LLM share of voice across acquisition-stage prompts ("best [category]", "[brand] vs [competitor]"). Adstock half-life: two to four weeks. Saturation: Hill with moderate aggressiveness. Enters as a media-equivalent variable in the acquisition MMM.

Step 4: Add Retention AI Visibility

LLM share of voice across retention-relevant prompts ("how to use [product]", "[product] tutorial", "[product] customer support"). Different prompt set, different adstock (typically longer, six to twelve weeks for retention because the consideration to cancel or stay is rare and slow). Enters as a separate variable in the retention MMM.

Step 5: Calibrate With Cohort Lift

For acquisition, standard geographic lift on AI visibility inputs. For retention, cohort-based holdout tests are harder but possible: suppress retention-marketing in matched cohorts and measure the differential churn. Cohort lift tests are operationally complex but produce the cleanest causal evidence for retention spend.

Step 6: Integrate With Finance Reporting

Subscription finance speaks in CAC, payback period, LTV/CAC ratio, and ARR/MRR growth. The MMM outputs should translate to these metrics: channel-level CAC (channel spend divided by channel-attributed new subscribers), channel-level LTV/CAC ratio (channel-attributed LTV divided by channel CAC), and channel-level contribution to ARR growth. The translation is the bridge between the MMM and the way subscription businesses actually manage themselves.

How Presenc AI Helps

Presenc AI provides separate acquisition-stage and retention-stage AI visibility tracking with distinct prompt sets for each. The dual data layer feeds the dual MMM structure (whether two separate models or one unified spec) and produces the AI variable identification for both stages of the subscription business.