Aaiz Khan

Dropping part 2 of our internal training series on Marketing Attribution Models. 25+ pages. High-level and conceptual. I created this for our data and engineering teams who need to understand the measurement concepts behind the data they're modeling. Download it. Save it. Share it w/ your team. Google Link is in the comments. ----- Why this matters: Attribution tells you who touched a customer before they converted. It does NOT tell you what caused the sale. That's a different question entirely. Most people treat attribution reports like truth. They're not. They're model outputs based on rules someone picked. ----- The core point: Attribution is just math that assigns credit to channels. Last-click, first-click, linear, time-decay, position-based, data-driven. Same customer journey run through six models = six different top performers. None of them are wrong. None of them are right. They're just different views of the same data. ----- What's covered: Foundations - What attribution actually is (and isn't) - Where attribution models live: GA4, ad platforms, MTA vendors Model types - Last-click, first-click, linear, time-decay, position-based, data-driven - Same journey showing 6 different credit assignments Click vs view - Why this matters more than people think - View-through inflation (platforms love it) Where it breaks - Non-digital touchpoints - Cross-device journeys - Logged-out users - Attribution windows Platform attribution - Why every platform grades its own homework - Adding up platform revenue = 3x your actual revenue When to use which model - Different models answer different questions - Common mistakes that lead to bad budget decisions Engineer context - GA4 clickstream data structure - Key attribution fields in BigQuery ----- We don't talk about attribution much at Power Digital. We lean heavily on incrementality-based measurement for strategic decisions. But attribution still matters for day-to-day optimization and real-time use cases. And if you're building data models that touch marketing performance, you need to understand how credit gets assigned. This pairs w/ the incrementality guide we dropped last week. Together they cover: who touched the customer (attribution) vs. what actually caused the sale (incrementality). What's your team's biggest attribution headache? #attribution #marketinganalytics #measure

76

8 Comments

Let's clear something up. The Kodak analogy fits companies, not industries. Kodak invented the digital camera in 1975. But they stuck to film profits and didn’t pivot. That's a company problem—not an industry death sentence. Photography didn’t die with digital. It exploded. More photos in a day now than the entire 20th century. Format changed; need didn’t. PyMC Labs and Colgate-Palmolive’s SSR method is genuinely groundbreaking. Hits 90% of human test–retest reliability on purchase intent using LLMs. Hours instead of weeks. Costs drop fast. You get scores plus reasoning. This is real. This changes how we work. But here’s what nobody’s saying: AI learns from biased data—and then amplifies it. Every dataset carries social stereotypes. Think of it this way: AI gets trained by humans who say 'this one's better.' But what they decide comes from their own background. That bias gets baked in and applied everywhere. Humans? Inconsistent. Illogical. Emotional. We say one thing, do another. We’re unpredictable. AI finds statistical patterns in messy data. Correlations, not causation. But it can't replicate all those lived, complex experiences behind real purchase decisions. Especially in India, where price thresholds shift by area; where family consensus wins over personal preference; where vernacular flips meaning. The Photoshop example is perfect. When it launched in 1990, people said designers were done. Wrong. Photoshop killed grunt work—retouching, color separations—and designers built new fields: UI/UX, motion graphics, 3D, brand systems. The tool democratized access but elevated the craft. For market research? SSR won’t replace researchers, but it’ll change what we do. Agencies selling panel volume? Trouble. Researchers using AI for speed, then ground-truthing with human samples? We are adopting Hybrid models that are faster and more accurate. We are creating new offerings: > Real-time dashboards with human calibration > AI segmentation plus ethnographic depth > Governance frameworks for synthetic vs human > Cultural-fluency audits catching what embeddings miss At ResearchFox, we're leaning in. More AI testing, faster failures, reserving human panels for validation that needs lived experience. The craft shifts from fielding to curation. Knowing when to sprint with AI, when to slow for human truth. Blending both for smarter business decisions. Want to see this in action? DM me for a pilot slot. #MarketResearch #ArtificialIntelligence #FutureOfWork #ConsumerInsights Pranab Sen Samar Baghel Rajesh Yadav Shweta Gupta Praveen Mettelu Sunitha Barlota Divyanjali Sinha Anil Kumar Sakala

7

This new piece from Jack Millership at Zappi really resonated with me. It challenges one of the industry’s most deeply held assumptions: the role of normative benchmarks. Yes, data stability and reliability matter. But so does relevancy. And the balance between the two is something I’m often wrestling with for my clients. A few quotes that stood out: 🔹 “Norms aren’t just numbers – they reflect shared values and expectations across cultures, communities, and generations.” 🔹 “Consumer priorities, loyalties, and the very tone of effective communication are evolving faster than ever before.” 🔹 “The mathematical definition of a norm gives a business confidence; the cultural interpretation should lead us to question benchmarks that lean too heavily on historical data.” The big provocation? 👉 Are we clinging to outdated benchmarks for reassurance rather than insight? Full article here: https://lnkd.in/eeBJ43yT

3

2 Comments

Looking Beyond the Dashboard Observation Over the past few posts, I’ve been reflecting on the intersection of data, marketing, AI, and human behaviour. What began as thoughts on signal vs noise has revealed something deeper about how we actually use data at work. Many of the challenges we face in analytics today are not purely technical. They are challenges of thinking, framing, and interpretation. Insight We have more tools than ever. More dashboards. More data. More automation. Yet clarity doesn’t always increase at the same pace. A dashboard can tell us what is happening in great detail. But it rarely tells us why it’s happening — or what to do next. That part still depends on human judgement, context, and the questions we choose to ask. Reflection Analytics doesn’t end at the dashboard. It begins when we interpret it. The real value appears when we move from “numbers on a screen” to “behaviour, causes, and possible actions.” Quote “The goal is to turn data into information, and information into insight.” — Carly Fiorina #MarketingAnalytics #HumanSideOfData #DataThinking #AnalyticsMindset #AIandWork #SignalVsNoise #Hiring

2

Approximately 43% of organisations use GA4 globally. Yet only a fraction truly utilise the data available to its full potential. By integrating GA4 with BigQuery (as simple as a few clicks) and interrogating raw, unsampled data through well thought out queries, organisations can unlock a wealth of intelligence to drive smarter marketing strategies. One particularly powerful use case? Pathway Analysis 🚀 Especially for e-commerce-focused businesses looking for deeper behavioural insight from their GA4 data. Ever wondered: - Which digital journeys lead to the most missed revenue? - Which journeys drive the most revenue? - Where do high-intent users drop off before converting? - How does behaviour differ across devices before purchase? In this example, my colleague Utkarsh Balooni transforms, queries, automates, and visualises GA4 data in BigQuery using Python — delivering the insights needed to answer all of the above. The result? ✅ Data-driven insights to dictate your CRO roadmap ✅ Data-driven insights to power your web A/B testing programme Dexata Rajneesh Gautam Charlie Nicholls Utkarsh Balooni Karan Arun More Elena M. Vijayaganapathy Veeramani Calvin Brady Krishna Nayak

23

1 Comment

Another insanely smart tool from my colleague 𓃋 Tyler Gargula. If you are a marketer struggling to understand buyer intent, this is a game changer....and it's free for you to use!( for now 😉.)

6

1 Comment

We saw a brand lose 20L+ in Ad Spend in 2 weeks and then ran diagnostics. One of our Clients stared at their Month End metrics and realised that scaling had gone horribly wrong. When they asked us to analyse it. We found that Meta Ads consistently gave them wrong signals. Failed to predict campaign exhaustion and continued over attributing to itself even as Total Order started tanking. On the surface, everything looked fine: → CTR, CPC, CPM stayed stable → Meta kept reporting 3.5x+ ROAS → “Scale more” was the default recommendation But underneath: → Incrementality had disappeared → Marginal ROAS turned negative → Meta was cannibalising demand from Google, Organic & Retention The real warning signs were there all along: → True ROAS was falling → Channel share was skewing heavily toward Meta → Blended performance was quietly eroding → Meta was increasingly over-attributing conversions. Those signals just weren’t visible inside Ads Manager. They learned the hard way that the only way to scale is adopting a multi channel strategy. Spending more isn’t always the answer.

5

Most organizations are not set up to act on the AI signals they are already receiving. That is not a technology problem. Every major martech platform is racing to solve the data and context layer. The signals are getting cleaner, faster, and cheaper by the month. The actual problem is decision architecture. Who decides what those signals mean? How does marketing interpret them differently than sales? What happens when product disagrees? Without a governing framework underneath, those signals just create faster, more expensive misalignment. You cannot buy or subscribe to organizational alignment. You have to build it. The companies winning right now are not the ones with the best AI tools. They are the ones who built a decision operating system first, and then let the tools plug into it. This is exactly the problem I built BX-AI-OS to solve. #Martech #AIGovernance #DecisionArchitecture #BXAIOS

5

In a world where shoppers make decisions in a blink, the traditional marketing funnel is officially obsolete. Kantar’s Leigh O'Donnell (she/her) shares her perspective with The Drum on how today’s purchase journeys are anything but linear. Instead, they're fast, fragmented, and fueled by emotion. In today’s always-on marketplace, discovery, conversion, and delivery can all happen in an instant. As Leigh notes: “Someone can become acquainted with a brand, convert it and have it delivered to their house… That’s a subterfuge for the classic path to purchase.” Backed by Kantar’s ShopperScape data and real-world insights, Leigh challenges marketers to rethink how they engage consumers across channels, especially in an era where loyalty is fleeting and attention spans are short. The funnel may be gone, but opportunity moves faster than ever. Learn how Kantar is helping brands pivot from path-to-purchase to moment-to-moment: https://lnkd.in/eqqfXfjt

3

Brands In AI Search era, should seriously be asking what sources are AI platforms citing when mentioning our competitors but not us? Citation Gap Analysis allows brands to identify the topics, publishers and individual URLs influencing AI responses, then prioritise where to invest in content, Digital PR and partnerships. It's not just getting more citations. It's getting cited by the sources that actually influence AI answers. GEO can show us where competitors are earning AI influence. #GEO #AISearch #SEO #DigitalPR

13

Compelling research from my colleague Shishir S. Pathak at the Boston Consulting Group (BCG): leading consumer companies invest heavily in #digital -- capabilities like cloud engineering, Gen AI copilots, composable technology architecture & data management.

3

As teams reset priorities for the new year, one pattern keeps repeating: Most CRO programs fail because they optimise outcomes without fixing decision systems. Heatmaps get reviewed. Reports get shared. Experiments get discussed. And then opinions take over. Real leverage starts when optimisation stops being episodic and becomes default behaviour. AI doesn’t help by being smarter. It helps by making hesitation, friction, and intent visible — continuously. Until optimisation becomes a system, not a project, teams will keep mistaking activity for progress. Curious how others are approaching CRO resets this year.

20

1 Comment