Marketing Analytics with Fexingo · 2026-07-02 · 7 min
Key moments - from our scoring
Substance score
62 / 100
Five dimensions, 20 points each
Lucas and Luna explore why traditional attribution frameworks collapse when applied to subscription models, using concrete examples like a B2B SaaS company that showed 4x ROAS via last-click attribution but only 1.2x when measured across a 12-month window with media mix modeling. The core issue: attribution models built for transactional purchases credit the final touchpoint while ignoring early-stage awareness channels (content marketing, podcasts, events) that generate top-of-funnel qualified leads. Subscription businesses face an additional blind spot - these models measure acquisition but not churn, allowing teams to optimize toward cheap, low-LTV customers who cancel within months. The solution requires shifting from conversion-focused attribution to LTV-based cohort analysis, time-decay models that distribute credit across the full journey, and media mix models that treat churn and lifetime value as dependent variables. Implementation challenges include data quality, statistical complexity, and misaligned incentives when marketing teams are still measured on quarterly acquisition targets. The episode discusses practical workarounds like blended metrics (cost per acquired customer with minimum LTV thresholds) and probabilistic attribution models powered by machine learning, alongside emerging tools that make LTV prediction accessible to smaller teams without dedicated data scientists.
Last-click attribution credits the final touchpoint (like a bottom-funnel search ad) with the entire conversion, while ignoring earlier touches like blog content and referrals that generated initial awareness; media mix modeling with a 12-month window distributes credit across the full journey, revealing that top-of-funnel brand-building drove the initial trial signup that enabled the final conversion.
Most attribution models credit the conversion to whatever channel drove the touchpoint closest to signup (e.g., day 10 email), but the actual decision-stage work happened weeks earlier in blog content or referrals; the model misattributes credit and overvalues late-stage campaigns.
A combination of cohort analysis tracking LTV over time by acquisition channel, time-decay models that credit early touches, and media mix models that include churn and LTV as dependent variables rather than just conversions.
A channel with cheap cost-per-acquisition may drive customers with high churn rates who cancel within months, making the true customer acquisition cost much higher; attribution models that only measure conversions never penalize these channels, leading to budget misallocation.
Start tracking LTV by acquisition channel using simple cohort analysis or spreadsheets, assigning acquisition costs to customer cohorts and measuring cumulative lifetime value; this reveals which channels drive high-value versus low-value customers without requiring sophisticated modeling.
Our reviewer’s read on each dimension, with quotes from the episode.
The episode delivers several substantive, non-obvious insights about subscription attribution (the difference between conversion windows and LTV curves, the 4x vs 1.2x ROI case study, the churn-blindness of standard models, cohort attribution gaps), but dilutes them with some repetition and concludes with filler about podcast support and accessible tools that add little tactical depth. For a 7-minute format, the core problems are well-articulated but solutions remain somewhat surface-level.
But the real value comes from month twelve, month twenty-four, maybe even month sixty. Attribution models that only look at the conversion event miss that entire future value curve.
Their last-click attribution model showed a 4x return on ad spend. But when they ran a media mix model with a 12-month attribution window, the actual ROI was closer to 1.2x.
The core argument - that subscription models break transactional attribution frameworks - is well-reasoned and not trivial, but the specific critique (last-click attribution, time-decay models, LTV cohort tracking) represents established thinking in marketing analytics rather than contrarian or first-principles innovation. The 4x-to-1.2x case is concrete but the prescriptive path (move to LTV-based metrics) is familiar to informed practitioners.
Attribution models that only used for transactional purchases - buy a pair of shoes, attribution window closes, done. But a subscription has a completely different lifecycle.
The best approach I've seen is a combination: use a media mix model with a time-decay function that gives partial credit to touches across the full customer journey, but also includes churn and LTV as dependent variables, not just conversions.
Lucas appears to be a practitioner with real exposure to subscription analytics (references case studies, cohort analysis, and media mix modeling), but the transcript provides no biographical detail, company scale, or proof of deep operational experience. Luna is a co-host facilitating rather than a guest. The credibility rests entirely on what Lucas says in the conversation, which is thoughtful but unverified.
I saw a case study from a B2B SaaS company - they were spending heavily on paid search for bottom of funnel keywords.
That's exactly what more sophisticated teams do. They assign acquisition costs to cohorts, then track cumulative LTV per cohort.
The episode anchors its core argument with one concrete example (the 4x vs 1.2x paid search ROI case study) and references Spotify's freemium complexity, but lacks most other specifics: no named companies beyond Spotify, no actual LTV thresholds, no sample time-decay curves, no real spreadsheet templates. The advice is tactical (stop using last-click, use time-decay, track LTV by channel) but underspecified for implementation.
Their last-click attribution model showed a 4x return on ad spend. But when they ran a media mix model with a 12-month attribution window, the actual ROI was closer to 1.2x.
Spotify, for example, has a huge free tier. Their attribution models have to account for the fact that free users eventually convert or churn, and the ad revenue from free users is itself a variable.
Luna asks clarifying questions (SaaS vs recurring, time-decay vs cohort analysis, LTV prediction complexity, freemium models) that push Lucas to elaborate, but rarely challenges or pressure-tests claims. The dialogue is collegial but lacks adversarial follow-ups or skeptical pushback on feasibility or cost-benefit tradeoffs. The closing pivot to podcast sponsorship also disrupts momentum.
You mean like SaaS businesses, or anything with a recurring payment?
So they were essentially subsidizing paid search with brand building that wasn't getting any credit.
Computed from the transcript - who did the talking, and the words that came up most.
Subscription businesses look like a marketer's dream - recurring revenue, known customers, predictable cohorts. But standard attribution models break on subscription economics. Lucas and Luna break down why last-click and even multi-touch attribution systematically undervalue early-stage marketing and overcredit retention campaigns, using a specific SaaS case study. They explain how subscription metrics like churn rate and customer lifetime value interact with attribution windows, and why media mix models with a time-decay function produce more truthful ROI numbers, especially for companies with free trials or freemium tiers. #MarketingAttribution #SubscriptionModels #SaaSMarketing #CustomerLifetimeValue #ChurnRate #MediaMixModeling #TimeDecayAttribution #FreeTrialMarketing #MarketingROI #B2BSaaS #MarketingAnalytics #AttributionModeling #CohortAnalysis #RetentionMarketing #AcquisitionCost #SaaSGrowth #FexingoBusiness #BusinessPodcast Keep every episode free: buymeacoffee.com/fexingo
Transcribed and scored by The B2B Podcast Index.
Lucas: Alright, let's talk about something that drives me crazy in marketing analytics - how standard attribution models break when you're selling a subscription. Luna: You mean like SaaS businesses, or anything with a recurring payment? Lucas: Exactly. Anything with a recurring payment.
A lot of marketers still use last-click or even multi-touch attribution, and those models were built for transactional purchases - buy a pair of shoes, attribution window closes, done. But a subscription has a completely different lifecycle. Lucas: Think about it: the first touch might be a blog post, then a free trial, then an email sequence, then a sales call, then the first payment. But the real value comes from month twelve, month twenty-four, maybe even month sixty.
Attribution models that only look at the conversion event miss that entire future value curve. Luna: So they're overvaluing the touch that closes the deal and undervaluing the awareness channels that bring people in? Lucas: Exactly. And it gets worse when you have a free trial.
Say someone signs up for a 14-day trial, then converts to a paid plan at day 10. Most attribution models credit the conversion to whatever email or ad they saw on day 10. But the real work happened in the blog post or the referral from month one. Lucas: I saw a case study from a B2B SaaS company - they were spending heavily on paid search for bottom of funnel keywords.
Their last-click attribution model showed a 4x return on ad spend. But when they ran a media mix model with a 12-month attribution window, the actual ROI was closer to 1.2x. Luna: Wow.
So they were essentially subsidizing paid search with brand building that wasn't getting any credit. Lucas: Exactly right. The brand building - content marketing, podcasts, events - was generating the top of funnel awareness that led to the free trial signups. But the attribution model gave zero credit to those channels because they were too far removed from the conversion event.
Lucas: And there's another layer. Subscription businesses care about churn. If your attribution model is only looking at acquisition, you might think a certain channel is great because it drives lots of initial conversions. But you don't know if those customers churn fast.
Luna: So you could be acquiring low-quality users from a channel that looks cheap on a cost per acquisition basis, but they cancel within two months. Lucas: Exactly. And the attribution model never penalizes that channel. It just sees the conversion.
The solution is to tie attribution to customer lifetime value - but that's hard because you have to wait months or years to get the data. Luna: What about using cohort analysis? You could attribute based on the month the customer was acquired and then track their LTV over time. Lucas: That's exactly what more sophisticated teams do.
They assign acquisition costs to cohorts, then track cumulative LTV per cohort. But that still doesn't solve the within-cohort attribution problem - which specific channels or campaigns drove those customers? Lucas: The best approach I've seen is a combination: use a media mix model with a time-decay function that gives partial credit to touches across the full customer journey, but also includes churn and LTV as dependent variables, not just conversions. Luna: So the model optimizes for long-term value, not just signups?
Lucas: Exactly. But even that has challenges. You need clean data, a long enough history, and the statistical chops to build the model. Most companies aren't there yet.
Lucas: And there's a practical issue: the marketing team is often measured on quarterly or monthly acquisition numbers. So even if you build a better model, the incentives might still push them toward short-term attribution. Luna: Right. So the attribution model change needs to come with a change in how the team is evaluated.
Lucas: Absolutely. I've seen companies shift to a 'blended' metric: cost per acquired customer with a minimum LTV threshold. So the marketing team still has a conversion target, but only conversions that meet a certain LTV prediction count. Luna: That seems fair.
But then the attribution model needs to predict LTV, which adds its own complexity. Lucas: It does. But it's better than pretending a subscription is the same as a one-time purchase. Luna: Let me ask you this - what about companies with a freemium model where users never pay but still generate value through ads or data?
Lucas: That's an even harder problem. If the user never converts to paid, standard attribution models might show zero revenue from that user. But they might generate ad impressions or word of mouth. Some companies use a 'value per engaged user' metric, but it's not standard.
Lucas: Spotify, for example, has a huge free tier. Their attribution models have to account for the fact that free users eventually convert or churn, and the ad revenue from free users is itself a variable. It's a whole other level of complexity. Luna: So what's the single biggest piece of advice for a subscription marketer listening right now?
Lucas: Stop using last-click attribution. At a minimum, switch to a time-decay model that gives more credit to early touches. And if you can, start tracking LTV by acquisition channel, even if it's just a simple spreadsheet. Lucas: The data doesn't have to be perfect.
Even a rough ltv by channel analysis will reveal which channels are bringing in high-value customers versus cheap conversions that churn fast. Luna: And that's actionable, right? You can shift budget away from the high-churn channels into the ones that build long-term value. Lucas: Exactly.
It's not about having the perfect attribution model. It's about having a model that aligns with your actual business model. Luna: Speaking of alignment - this kind of deep dive into attribution is exactly the type of analysis that takes time and resources. If these conversations have helped you think differently about your marketing, and you'd like to support the show staying ad-free and independent, you can buy us a coffee at buy me a coffee dot com slash fexingo.
No pressure, just helps keep the lights on and lets us keep doing these deep dives. Lucas: Yeah, we really appreciate the listeners who've supported us that way. It makes a difference, honestly. Lucas: Alright - one last thought on subscription attribution.
I think the trend we're going to see more of is probabilistic attribution combined with LTV modeling. Machine learning models that predict the long-term value of a user based on early behavior, then assign credit accordingly. Luna: That sounds powerful. But also a bit intimidating for smaller teams.
Lucas: It is. But the tools are getting more accessible. Some of the marketing analytics platforms are starting to offer built-in LTV prediction. So even if you're a team of two, you might be able to set it up without a data scientist.
Luna: Good to know. Thanks, Lucas. Lucas: Thanks, Luna. See you next time.
Other episodes covering the same guests and topics, from across The B2B Podcast Index.
Lessons from Building LiftLab Media Mix Modeling and Experimentation Platform | John WallaceThe Efficient Spend Podcast · on Multi-touch attribution95 / 100
Episode 107 - From Email List to Membership: Build Recurring Revenue Without Ads with Mike MorrisonCreate Online Business Success! · on Churn rate86 / 100
211: Key Numbers Every Therapist Needs to Track for Client Retention and Business GrowthMoney Skills For Therapists · on Churn rate85 / 100
#124 Why Authentic Content Beats the Algorithm | Yoray HalevyAlways Be Testing · on Multi-touch attribution83 / 100
Userpilot: Why SEO stopped being enough, and how they rebuilt acquisition with ABM in 2026The SaaS Growth podcast · on Multi-touch attribution82 / 100
Why Revenue Is Up But Profit Isn't Moving: The Unit Economics Blind Spot Most Shopify Brands Have - Misha Druzhinin | Why Revenue Doesn’t Equal Profit, Why Scaling Profit Beats Scaling Revenue, The Hidden Danger Of Discounts (#481)Ecommerce Coffee Break · on Customer Lifetime Value (LTV)82 / 100
How Incrementality Reveals True Marketing Impact90 / 100Why Marketing Attribution Misses the Seasonality Pattern91 / 100Why Marketing Attribution Breaks on Subscription Models92 / 100Why Marketing Mix Models Beat Attribution for Long Sales Cycles92 / 100How Unified ID Replaces Broken Third-Party Cookie Attribution85 / 100