Marketing Analytics with Fexingo · 2026-06-29 · 10 min
Key moments - from our scoring
Substance score
74 / 100
Five dimensions, 20 points each
During the holidays, compressed purchase cycles and elevated buying intent create a systematic distortion in how standard attribution models assess channel contribution. Lucas and Luna examine a real case where a brand's cost per acquisition dropped from $45 in November to $27 in December - not because ads got twice as effective, but because holiday shopping velocity changed the timeline from three weeks to three days. Last-click and linear attribution models apply fixed weighting rules regardless of seasonality, meaning a customer exposed to a November TV spot or podcast sponsorship often doesn't convert until December, when a retargeting ad or search query becomes the "last touch" and claims all credit. Research shows last-click models overattribute December conversions to paid search by roughly 30 percent while television brand campaigns receive virtually no credit. The discussion covers practical solutions: running separate attribution models for peak versus off-peak seasons, adjusting conversion lookback windows based on actual time-to-conversion data (14 days for December, 60 days for November), and implementing custom seasonal attribution weights using regression models or Markov chains. For resource-constrained teams, manual overrides based on brand search spikes during holidays offer a spreadsheet-level alternative to expensive media mix modeling. The stakes are significant - one brand avoided a $500,000 misallocated spend decision by recognizing this distortion.
Holiday shopping velocity collapses your purchase cycle from three weeks to three days, making conversions happen faster. Standard attribution models misinterpret this - they credit whatever touchpoint happened right before purchase (usually a December retargeting ad or search query) with full conversion value, when in fact November brand campaigns created the initial intent that the December ad merely capitalized on.
Last-click models overattribute December conversions to paid search by approximately 30 percent while giving television and social brand campaigns virtually zero credit, even though those channels built the awareness that drove the December purchase behavior.
Segment your attribution data by season - create one model for November-December and another for January-October. Adjust your lookback window based on actual time-to-conversion analysis: use 14-day windows in December (faster cycles) and 60-day windows for November campaigns, then manually assign 15-25 percent of brand search credit back to earlier awareness campaigns based on search volume spikes.
No - compress it instead. Run a time-to-conversion analysis to find the actual distribution: if 90 percent of December conversions happen within 10 days of first touch, use a 14-day window in December. For January when cycles extend to 20 days, extend the window to match. Let data determine the window, not intuition.
Brand X avoided shifting an additional $500,000 from TV to search in December 2025 after their media mix model revealed that TV was responsible for 24 percent of revenue, not the 12 percent their linear attribution model had shown.
Our reviewer’s read on each dimension, with quotes from the episode.
The episode delivers several substantive, non-obvious claims about attribution blindness during seasonal periods, including concrete examples (Brand X's $500k budget shift averted, CPA swings from $45→$27→$34) and actionable mechanics (time-to-conversion analysis, dynamic lookback windows, seasonal decay factors). However, roughly 20-25% of the runtime is spent on padding (supporter callout, cross-references to other episodes, conversational throat-clearing) and the deeper technical exploration of seasonal regression models is sketched rather than deeply explored.
if you're using a standard multi-touch attribution model that doesn't adjust for seasonality, that December CPA drop looks like your ads suddenly got twice as effective. You might even increase spend based on that signal.
Brand X's CMO told me that if they'd followed the attribution model blindly, they would have shifted another $500,000 from TV to search in December 2025. They didn't, because the MMM gave them the confidence to stick with a balanced mix.
The core insight - that attribution models systematically misallocate credit during seasonal compression because purchase cycles shorten - is genuinely counterintuitive and not widely discussed in mainstream marketing discourse. The specific workarounds (dynamic lookback windows, seasonal decay factors, Markov chains with time-of-year variables) show original thinking. However, the episode leans slightly on familiar MMM vs. last-click comparisons and doesn't fully explore contrarian angles (e.g., whether seasonal adjustment might mask channel decay).
The purchase cycle compresses. A customer might go from awareness to purchase in three days in December, versus three weeks in March. The model sees that short window and assumes the last touch was the decisive one, but it's often just the final nudge after weeks of priming.
paid search's contribution dropped to 28 percent, and TV jumped from 12 percent to 24 percent
Lucas demonstrates practitioner credibility through named client case study (Brand X), concrete outcomes (14% YoY lift vs. flat competitors), and hands-on technical knowledge (regression models, Markov chains, data stitching). Luna appears to be a co-host/knowledgeable peer rather than a guest, but engages at appropriate depth. Lucas's grounding in real CMO conversations and actual budget decisions indicates senior operational experience, though his exact title/scale of current work isn't explicitly stated.
a mid-market ecommerce brand I've been following saw its cost per acquisition drop from about $45 in November to $27 in December last year
Brand X's CMO told me that if they'd followed the attribution model blindly, they would have shifted another $500,000 from TV to search in December 2025
Episode is rich with specific data: CPA figures ($45→$27→$34), dollar impact ($500k budget swing averted), attribution model output percentages (42% vs. 28% for paid search; 12% vs. 24% for TV), conversion probabilities (0.3 vs. 0.7), and time horizons (3 days vs. 3 weeks, 14-day windows, 60-day lookback). The Brand X case is named and detailed enough to be credible. Only minor gap: no external studies cited with publication year/source beyond vague reference ('study from a few years back... from a retail analytics firm').
its cost per acquisition drop from about $45 in November to $27 in December last year. Then in January, it jumped back to $34
paid search driving 42 percent of December revenue. But when they ran a media mix model with a seasonal variable, paid search's contribution dropped to 28 percent, and TV jumped from 12 percent to 24 percent
Luna asks clarifying follow-ups (e.g., 'So you're saying the attribution model itself is creating a false signal?', concern about overcorrecting on lookback windows) and introduces competing ideas (Markov chains, seasonal decay factors). However, the conversation rarely pushes back on Lucas or challenges his framing; Luna mostly validates and builds on his points. Missing are sharp pushback moments (e.g., 'But doesn't seasonal adjustment hide poor channel performance?'), deeper probing of trade-offs, or genuine disagreement. The structure feels more like a curated Q&A than an adversarial or exploratory dialogue.
So you're saying the attribution model itself is creating a false signal, because it's treating all months as equal?
But isn't there a risk of overcorrecting? If you shrink the lookback window in December, you might miss the contribution of early December ads that convert in late December.
Computed from the transcript - who did the talking, and the words that came up most.
Most attribution models treat every day of the year the same - but consumer behavior shifts dramatically during the holiday season. In this episode, Lucas and Luna break down how standard multi-touch attribution models systematically misattribute December conversions, using real data from a mid-market ecommerce brand that saw its CPA drop 40 percent in December only to rise 25 percent in January. They explore why last-click models over-index on holiday ads while underweighting the November brand campaigns that actually drove the purchase. The hosts also discuss how media mix models with seasonal variables can correct for this distortion, and why marketing teams should run separate attribution weights for peak and off-peak periods. If you've ever wondered why your January campaign performance looks terrible even though the strategy didn't change, this episode will explain the seasonal attribution blind spot and how to fix it.
Transcribed and scored by The B2B Podcast Index.
Lucas: Let me start with a number that should bother anyone who runs marketing analytics: a mid-market ecommerce brand I've been following saw its cost per acquisition drop from about $45 in November to $27 in December last year. Then in January, it jumped back to $34. Same targeting, same creative, same offer structure. So what changed?
Luna: Holiday shopping frenzy, right? People are just in a buying mood in December. Lucas: That's the obvious answer, and it's partly true. But here's the problem - if you're using a standard multi-touch attribution model that doesn't adjust for seasonality, that December CPA drop looks like your ads suddenly got twice as effective.
You might even increase spend based on that signal. Then January hits and your model tells you performance tanked, so you cut budget. Luna: So you're saying the attribution model itself is creating a false signal, because it's treating all months as equal? Lucas: Exactly.
Most attribution models - whether last-click, linear, or time-decay - apply the same weighting rules regardless of the time of year. They don't account for the fact that a customer who sees your ad in November might not convert until December, and when they do convert, the model gives full credit to whatever touchpoint happened right before the purchase. That's almost always a December ad - a retargeting banner, a promotional email, a search ad for 'free shipping codes'. Luna: So the November upper-funnel campaigns - the brand awareness video, the podcast sponsorship, the out-of-home - they get zero credit in a last-click model, even though they built the intent that December ad capitalised on.
Lucas: Right. And the magnitude of this distortion is bigger during the holidays because the purchase cycle compresses. A customer might go from awareness to purchase in three days in December, versus three weeks in March. The model sees that short window and assumes the last touch was the decisive one, but it's often just the final nudge after weeks of priming.
Luna: I remember a study from a few years back - I think it was from a retail analytics firm - that showed last-click models over-attribute December conversions to paid search by about 30 percent, because search is typically the last click before purchase. Meanwhile, television and social brand campaigns that ran in November got virtually no credit. Lucas: That study aligns with what I've seen. The brand I mentioned - let's call them Brand X - ran a pretty typical holiday mix: national TV spots starting mid-November, social video in November and December, paid search and retargeting through December.
Their multi-touch model, using a linear attribution, showed paid search driving 42 percent of December revenue. But when they ran a media mix model with a seasonal variable, paid search's contribution dropped to 28 percent, and TV jumped from 12 percent to 24 percent. Luna: That's a massive swing. So the MMM basically said, 'Hey, the TV ads that ran five weeks before Christmas were responsible for nearly a quarter of the revenue,' but the attribution model couldn't see that because of the time lag.
Lucas: Exactly. And this isn't just an academic curiosity. Brand X's CMO told me that if they'd followed the attribution model blindly, they would have shifted another $500,000 from TV to search in December 2025. They didn't, because the MMM gave them the confidence to stick with a balanced mix.
Their total holiday sales were up 14 percent year-over-year, while their average competitor in the same category saw flat to negative growth. Luna: So the fix seems obvious: run a media mix model alongside your attribution model, especially during seasonal peaks. But MMMs are expensive and require a lot of data history. What about teams that can't afford a full MMM?
Lucas: There are simpler workarounds. One is to segment your attribution data by period - run separate attribution models for November through December and for January through October. You can build two sets of weightings: one for peak season, one for off-peak. Even a basic last-click model becomes more useful if you acknowledge that the conversion window is shorter during holidays and adjust your lookback window accordingly.
Luna: Right - instead of a standard 30-day lookback window, maybe you shrink it to 14 days in December because the purchase cycle is compressed, and extend it to 60 days for the November campaigns that need more time to convert. Lucas: Precisely. And you can do that in most analytics platforms - Google Analytics 4, Adobe Analytics, even some of the dedicated attribution tools allow custom conversion windows by channel or campaign. It's not perfect, but it's a low-cost way to reduce the seasonal distortion.
Luna: But isn't there a risk of overcorrecting? If you shrink the lookback window in December, you might miss the contribution of early December ads that convert in late December. Lucas: That's a valid concern. The key is to use data to determine the right window.
You can run a time to conversion analysis for each month: plot the distribution of how many days passed between first touch and last touch. If you see that 90 percent of December conversions happen within 10 days of first touch, a 14-day window is safe. But if January conversions take 20 days on average, a 14-day window would undercount. So the window should be dynamic - tied to the actual conversion velocity of that period.
Luna: Another approach I've seen is using a weighted attribution model that incorporates a seasonal decay factor. Instead of a standard time-decay that gives more credit to recent touches, you adjust the decay curve based on historical conversion patterns for that time of year. Lucas: That's a more sophisticated version of the same idea. I've seen some of the enterprise-level attribution platforms offer something they call 'seasonal attribution weights' - basically a lookup table that says, 'For a conversion in December, a touch that happened 30 days ago should be weighted at 0.
4 instead of 0.2 because historically, November ads drive December conversions.' It's not common yet, but it's gaining traction. Luna: It feels like one of those areas where the data science team can really add value - building a custom seasonal model using historical conversion data, rather than relying on the out-of-the-box attribution model that treats every day the same.
Lucas: Absolutely. And listen, this whole conversation is the kind of thing that can genuinely save a marketing team from making a bad budget decision. If these marketing analytics conversations have sparked something you've actually used in your own work, that's the whole point. Luna: It really is.
We keep this show ad-free, and listener support is what makes that possible. Lucas: A couple of dollars a month is genuinely what keeps these going - buy me a coffee dot com slash fexingo, if you've gotten something out of them. Luna: Yeah, it makes a real difference. So, back to the seasonal attribution question - you mentioned that some teams are building custom seasonal models.
What does that actually look like in practice? Lucas: Good question. Let's say you have three years of conversion data. You can train a simple regression model that predicts the probability that a given touchpoint contributed to a conversion, with a variable for the day of year.
That gives you a seasonal weight for each channel. For example, you might find that a Facebook ad viewed on November 15 has a 0.3 probability of contributing to a conversion that happens on December 10, while a Google search ad on December 10 has a 0.7 probability.
Then you build your attribution model using those probabilities as weights, rather than a fixed time-decay curve. Luna: That's essentially a custom Markov chain model with seasonality, isn't it? Where the transition probabilities change depending on the time of year. Lucas: Exactly right.
And the beauty of it is that it doesn't require a huge data science team. A decent analyst with Python or R can build it in a couple of weeks. The hardest part is getting clean, stitched-together user-level data across devices - which is a whole other episode. Luna: Speaking of which, we did cover cross-device blind spots in episode 68.
But for teams that don't have that level of data integration, are there any simpler signals they can use? Lucas: One simple signal is just comparing your paid search brand versus non-brand volume. During the holidays, brand search spikes because people are looking for specific stores or products they've already been exposed to. If you see brand search conversions skyrocket in December, that's a strong signal that your earlier brand campaigns are working.
You can then manually adjust your attribution - maybe assign 20 percent of brand search credit to the November TV or social campaigns that drove the awareness. Luna: That's a manual override, but it's better than nothing. And it's something any marketing team can do in a spreadsheet. Lucas: Right.
And the key is to do it before you make budget decisions. If you wait until January to look at your attribution data and wonder why your CPA jumped, you've already missed the chance to optimise. The insight needs to be applied in real-time, or at least in the planning phase. Luna: So really, the takeaway is: don't let your attribution model treat December like it's any other month.
If you do, you'll overinvest in bottom of funnel channels during the peak and underinvest in the brand-building that actually drives the peak. Lucas: That's the one-sentence summary. And it applies beyond holidays, too - think about back to school season, Black Friday, Valentine's Day, any period where consumer behavior is fundamentally different from the baseline. The same principle holds: your attribution model needs to account for seasonality, or it will systematically mislead you.
Luna: It's one of those cases where a more nuanced understanding of your data can save you from a really expensive mistake. Lucas: Exactly. And the brands that figure this out - they're the ones that consistently outperform their competitors during the most important sales periods of the year.
Other episodes covering the same guests and topics, from across The B2B Podcast Index.
How Kubernetes Topology Spread Constraints Create Scheduling HotspotsDevOps Daily with Fexingo · features Luna95 / 100
How B2B Brands Wreck Pipeline with Unsyncroned CRM DataThe Marketing Operator Podcast with Fexingo · features Luna92 / 100
Why API Webhook Payloads Should Be Signed Not VerifiedThe Developer Tools Podcast with Fexingo · features Luna90 / 100
How to Sell Against a Competitor Already in the BuildingSales Leadership with Fexingo · features Luna85 / 100
Why B2B Brands Are Using AI for Account PrioritizationThe Growth Operator with Fexingo · features Luna84 / 100
Enterprise Software Buyers Now Demand a Vendor Data Portability GuaranteeB2B SaaS Talks with Fexingo · features Luna82 / 100
Why Incremental Lift Testing Beats Attribution91 / 100How Brand Lift Studies Reveal True Marketing ImpactHow Offline Sales Drive Online Marketing ROIHow Contextual Targeting Replaces CookiesHow First-Party Data Reshapes Attribution