Performance Attribution Models

I spent over $3M on Meta ads in January. And I use 3 attribution models: Ad platforms are notorious for taking credit for view-through conversions they didn't drive. They do it to bait you into spending more. The issue is that your top 1-2% of ads should drive ~50% of your spend and revenue. If you're relying on bad attribution, you won’t be able to find them. This is why 8-9 figure brands (that NEED their tracking to be faultless), use 3 attribution models: 1. Multi-touch attribution (MTA) - for ad and campaign level optimization. This is your Triple Whale. Great for knowing which ads are performing best, which ones to scale, which to cut. Not as good for comparing channel to channel. It also will overcount total revenue, which you need to be careful about. To make sure your account is well optimized, plot CPA vs Spend on a scatter plot. The top ads should be in the low CPA, high spend zone. 2. Post-purchase survey - for channel level allocation. Get a 35%+ response rate, extrapolate to all new customers, and calculate your cost per new customer response per channel. This tells you which channel to push into. Click-based attribution overvalues lower-funnel performance by up to 250%. Post-purchase surveys catch what click attribution misses - top-of-funnel creative can drive 13X more incremental acquisitions than bottom-of-funnel. 3. Marketing Mix Model (MMM) - for validating direction. You can't use this daily, but it confirms your post-purchase survey is sending you the right way. Then you use post-purchase on a daily basis to optimize channel allocation. Some channels drive low-quality customers that look good on ROAS but don't stick around. MMM helps you optimize for 12-month profit as opposed to just immediate return. The other thing to know is that view-through attribution is poor signal. Make sure your attribution is set up for 7 or 14 day click, depending on your purchase funnel. One day view will overcount. Here's what this gives you: When performance drops, you know exactly where to pull budget to create the smallest impact on revenue while keeping the company profitable. When things are going well, you know exactly where to push budget to scale effectively. Bottom line: -> Use MTA for ads and campaigns.  -> Use post-purchase surveys for channel allocation.  -> Use MMM to validate you're heading the right direction. This is how 8-9 figure brands figure out where every dollar should go.

Marketers are skeptical of attribution models. And honestly, they should be. Most are built on shaky assumptions, like giving all the credit to the last ad someone clicked before buying. Many are black box. I'm always in search of research on better ways to measure marketing's impact. So this week on The Marketing Architects Podcast we covered a study titled, "Bayesian Modeling of Marketing Attribution" by Ritwik Sinha and David Arbour from Adobe Research and Aahlad Manas Puli from NYU. The researchers modeled customer journeys probabilistically, looking at things like ad decay, exposures across different channels, and purchase probability. All of that came together to change the chance of a sale over time. One finding: When users saw more than 20 ads in a short window, the chance of a sale went down. Another takeaway: Search and display ads had extremely short half-lives. Their influence faded fast. The model also assigned strong credit to owned and offline channels, which traditional digital attribution methods often ignore. (❤️📺) The Bayesian model doesn't just assign credit, it gives us a sense of how much a channel mattered, how long its effect lasted, and how confident you should be in the results. Even if your brand isn’t ready to adopt a model like this, it's interesting to learn about. And backs up why it's important to invest in multiple models and perspectives. Links in the comments to listen to the podcast + read the study.

For those breaking into quant finance: When I started studying Financial Engineering I thought that: "Factor models are for building strategies that beat the market." That CAN be one use (firms have been implementing factor investing for decades). But the more strategic and universally valuable use, risk management, is available to every investor, not just those pursuing factor investing strategies. The misconception comes from how factor models are taught: In academia, we learn: - Fama-French shows size and value factors tend to generate excess returns - Momentum strategies tend to capture persistent trends - Therefore: tilt your portfolio toward these factors to beat the market This narrative focuses on RETURN GENERATION. My previous post showed this: Leveraged ETFs like UPRO take 3x the market risk but deliver only 2x the returns. I used factor models to: - Estimate systematic risk exposure (beta = 3.04) - Decompose returns into market factor vs. idiosyncratic components - Assess whether the returns justified the risk I wasn't trying to beat the market. I was trying to UNDERSTAND the risk. This is how sophisticated investors typically use factor models: Use Case 1: Risk Measurement. "My portfolio has a beta of 1.2 to the market, 0.3 to value, -0.1 to momentum" Measuring exposures, not predicting returns. Use Case 2: Performance Attribution. "Last quarter: +5% total = Market +3%, Size +1%, Value -0.5%, Stock-specific +1.5%" Understanding where returns came from. Use Case 3: Portfolio Construction. An asset allocator has dozens of stock picks from various analysts. Factor analysis reveals unintended concentrations. Optimization maintains picks while neutralizing unwanted exposures. Use Case 4: Risk Budgeting. A pension fund targets 6% tracking error. Factor models show how much comes from intended tilts vs. stock selection. Notice: None of these use cases require PREDICTING which factors will outperform. They're about MEASURING what risks you're taking and WHERE returns come from. Anyone managing portfolios can use risk factor models to: - Measure exposures explicitly - Decide which are intentional vs. accidental - Understand what drives performance - Manage risk effectively For those learning quantitative portfolio management: Risk Factor models are fundamentally MEASUREMENT tools. This is why my course (built with Edgar Mauricio Alcántara López) starts with risk measurement (Module 1), then teaches you HOW to build and use the simplest risk factor models. Want to learn how factor models work in practice? Check factor attribution in Module 6.2: https://lnkd.in/eMSixrfr All with Python code and real data. This course was created independently. All views are my own. How have factor models changed your understanding of portfolio risk? #QuantFinance #PortfolioManagement #RiskManagement

We tested Meta’s new incremental attribution model. The results broke a few assumptions. Out of 4 campaigns last week, 1 was optimized for incremental attribution. The other 3 were ASC and ABO campaigns: And yet: - 1-day click CPA was 48.2% lower than the ASC - Incremental CPA was 49.0% lower than the ASC - New audience ROAS was atleast 55% higher than the ASC Why does this matter? Because 1-day click is the most conservative attribution model, it only counts conversions that happen fast and post-click. It’s historically been the cleanest proxy for incrementality. So lower 1DC CPA is exactly what short consideration DTC products need - faster conversions, lower CAC, and better cash efficiency 🚨 How It Works: Meta now runs always-on holdout tests in the background. It splits users into treatment and control groups and continuously asks: “Did this person convert because they saw the ad, or would they have purchased anyway?” The difference between those groups is considered incremental lift and that becomes the basis for attribution, optimization, and reporting. It’s not about what happened within a time window anymore. It’s about what happened because of the ad. 🚨 Why This Matters - Eliminates the guesswork of choosing between attribution windows - Shifts focus from tweaking settings to scaling what actually works - Enables budget consolidation and simpler account structures - Reduces dependency on exclusions and segments, Meta already accounts for it - Moves us toward causal measurement inside the algorithm itself We’ll continue to validate this. But so far, incremental attribution has outperformed our default benchmarks. Including ASC.

💡 Everyone's measuring #marketing #performance. But I find that a lot of great marketers are not combining the right #measurement protocols to tell the real story behind the numbers. The pressure to prove #ROI is intense. And yet most teams are either drowning in #data they can't action, or relying on metrics that only tell part of the story. The problem isn't effort, but it may be a lack of experience in using the right framework. There are two models that are essential in a marketer's toolbelt - Marketing Mix Modeling (#MMM) and Multi-Touch Attribution (#MTA). They're not competitors. Frankly they solve different problems and together, they give you a more comprehensive understanding of #marketing performance than either can alone. 🧠 Marketing Mix Modeling (MMM) :: Your top-down view. ✨ It uses aggregated data such as spend, revenue, pricing, seasonality, even external economic factors to model how your entire marketing mix drives business outcomes over time. → Mechanics: Statistical regression across channel-level data, typically requiring 2+ years of historical to be reliable. → Use Case: Annual budget planning, scenario modeling, and measuring channels that are hard to track individually. → Primary Limitation: It won't tell you what's happening in your campaigns right now. It's a strategic lens, not a real-time one. 🧠 Multi-Touch Attribution (MTA) :: A bottom-up analysis. ✨ It tracks individual user journeys across digital touchpoints such as impression, clicks, search, conversions and distributes credit across each interaction. → Mechanics: User-level data stitched together across sessions and platforms to map the path to purchase. → Use Case: Real-time digital campaign optimization, creative testing, and understanding which touchpoints are actually moving people through the funnel. → Primary Limitation: It's increasingly fragile in a privacy-first world, and it systematically undervalues anything offline or upper-funnel. As with any valuable framework, there is great benefit in pairing these two models together in partnership as they truly fact check one another. This is what's called a Unified Marketing Measurement, using MMM to set your strategy and allocate budgets at a macro level, while MTA helps you optimize the execution of your digital campaigns week to week. MMM tells you where to invest. MTA tells you how it's performing. One gives you the long-term baseline. The other gives you real-time signal against it. It may sound like a lot, but it doesn't have to be. Start with the #analysis that fits your immediate need and build the other alongside it. Let them inform each other over time. Marketing measurement doesn't need to be perfect from day one. It just needs to be pointed in the right direction. Are you using one, both, or something else entirely? I'd love to hear how your team is approaching measurement right now. #MarketingMeasurement #MMM #MTA #MediaMix #MarketingAnalytics #DataDrivenMarketing

"If I only looked at last-touch attribution, I would have killed everything driving our growth." Kacie Jenkins 🎁 uncovered a scary truth about B2B marketing metrics: Sendoso's best-performing channel is direct website traffic. But traditional attribution missed that those "direct" visitors had already: + Interacted with partners + Opened nurture emails + Seen organic content + Taken a product tour + Engaged at events + Received a gift The pipeline was there. The attribution wasn't. If you saw their multi-touch data, you'd see something fascinating about these "direct" visitors... Most of them had interacted with the exact channels that looked like they were failing. The same channels a finance team would have flagged for cuts. This pattern kept showing up: High-intent buyers were consuming 7-8 different marketing touches. None of them showed up in pipeline reports. Then they'd visit the website directly and convert. Without multi-touch analytics, every investment driving those conversions looked worthless. That's when they made a radical change to their attribution model. The results transformed not just their pipeline reporting, but their entire relationship with finance. Your "worst performing" marketing channels might actually be your best. Most CMOs get forced cut them before they ever find out. If you're looking to transition away from being a lead-gen machine, this is the way.

Is Multi-Touch Attribution collapsing into something even stranger? If you have been in marketing long enough, you know the dream. Track every touchpoint, assign perfect credit, and scale spend with total confidence. But the more you work with real data, the harder that dream is to believe. Some blame cookie loss. Others point to fragmented user journeys. But the deeper issue has always been the same. Attribution models are built on correlation, not causation. And it is easy to mistake one for the other. In-platform dashboards and most MTA tools show correlation. Sometimes it overlaps with true causation. But often it does not. It is easy to miss the forest for the trees. You believe the correlation, you scale spend, and suddenly your entire account is off track. You have probably seen it. You increase spend maybe 20 percent and your ROAS might tank. Lately, we are seeing attribution start to evolve. Platforms like Meta are introducing their new “incremental attribution” model. Under the hood, they are likely using Bayesian Structural Time Series (BSTS) to build synthetic control groups and estimate lift. It is a meaningful step forward from last-click attribution. But even BSTS has limits. Without real-world randomization, synthetic models can still be biased by seasonality, hidden confounders, or external shocks. It is better than pure correlation, but it is not the same as true causality. Working with big brands, we see many default to 1-day click attribution inside Meta and Google. Sometimes, that is the right move. A fast-moving buyer journey may genuinely favor short attribution windows. Other times, a longer view or view-through attribution paints a more accurate picture. It all depends on your specific customer path and it is something you can only validate through proper holdout experiments. Here is what we have learned helping brands navigate this shift: 1. Incrementality studies are a snapshot, not a prediction. A 3x iROAS today does not mean you can triple spend tomorrow and expect the same result. 2. Scaling safely requires measuring marginal ROAS across different spend levels. Growth is not linear. Incremental revenue per extra dollar or impression usually declines as you scale. 3. Attribution defaults can either support or sabotage your growth, depending on how well they match your actual buyer journey. Measurement must adapt, not assume. The future of measurement is not about finding perfect attribution. It is about understanding where each model is strong, where it is weak, and when you need real experiments to map your next move with confidence. MTA + Incrementality + MMM + Self- Reported Attribution = Success in 2025. Attribution is not dead. But confusing correlation for causation has stalled more growth than bad creative ever has.

Last year, I asked 200+ marketers one question: "How many of you believe there's a single source of truth that tells you everything about your marketing performance across all channels?" Almost no hands went up. They're right to be skeptical. If there is one thing that I've taken away from managing $2B+ in total marketing budgets, it’s that every measurement method tells a different story. - Google Analytics says one thing.  - Ad platforms say another.  - MMM disagrees with both. And the problem gets worse. - Third-party cookies might disappear. - iOS tracking is limited. - AI bidding systems are black boxes. We're facing unknown returns on ~$700 billion in annual digital ad spend. The result: stakeholders get puzzled and frustrated because they don't know what the real ROI of your marketing campaigns is. The answer I found most effective is not to rely on the 1 “perfect" methodology. It's triangulation - getting 3 methodologies working together in orchestration: Lift Testing - the gold standard for establishing causality. Run geo-experiments on the biggest channels to understand what conversions you wouldn't get without advertising. Statistically accurate and privacy-proof, but difficult to scale with opportunity costs. Marketing Mix Modeling - holistic regression linking all inputs to outputs. Captures seasonality, promotions, pricing, offline media, competitor activity. Gives you baseline conversions if you stopped all marketing. Privacy-proof but limited granularity. Multi-Touch Attribution - user-level methodology tracking touchpoints in conversion paths. Real-time and granular for daily optimization. But myopic - only sees UTM-tracked visits, misses offline marketing, inflates attribution on bottom-of-funnel channels. This is what combining them looks like: - Start with a Bayesian MMM as your absolute framework. - Inject your lift test results as prior knowledge to calibrate the model - this ensures your MMM output stays grounded in reality instead of producing unrealistic attribution from finding a local optimum. - Then layer MTA underneath on a relative basis. Take your realistic channel-level results from the calibrated MMM and use MTA to break them down to campaign-level granularity or below within each channel. This gives you a complete picture: holistic insights for strategic budget allocation and granular data for daily optimization decisions. Most companies pick one methodology and force it to answer every question. That's like using a hammer for surgery. Each method has blind spots. Together, triangulation fulfills most of your marketer needs. I've seen this approach work across industries through my work at Rocket Internet and with @Growth Vision Partners clients. The single source of truth is still a myth, and there is no silver bullet. But triangulation gets you as close as you’ll ever get to knowing your marketing’s real ROI.

What's the difference in these pictures? They're all the same client looking at the same time period with nothing changed. ROAS & conversion rate are materially different though. Delayed attribution. When you're looking at data with attribution windows longer than one day, unless you're actively factoring in those delayed conversions, you're not getting a full picture. This causes issues especially when you're looking at recent performance and comparing in YoY. When we first reported this data on the 16th of January, YoY performance looked terrible. 21 days out? Performance is up YoY. If you're not factoring in delayed attribution, you're getting an incomplete picture which can lead to bad decisions. This is also why I love working with 1 day click data in a platform like Northbeam. It allows for more nimble decisions because your data is back in in one day. (They also forecast out 7D, 30D, 90D and LTV numbers) If you're not using a platform like Northbeam, you can measure in-platform delayed attribution by looking at a timeframe of data periodically afterward, like you can see below. You can then get a delayed attribution multiplier you can use to get an idea of what the actual performance will be in the future.