Strong data-driven marketing and analytics professional who has a penchant for the creative & business side of advertising as well.
Well-versed in BigQuery, Looker Studio, Marketing Mix Modeling, Incrementality Testing, AB testing
I bring a healthy mix of research orientation & a client-focused approach to the table. 
Been 7+ years working with Google Ads, Facebook Ads, Bing Ads, Linkedin Ads, Google Analytics and Google Tag Manager. 
Having spent over $500K a month for hundreds of clients across different sectors and countries and having scaled scores of them, I have learnt a thing or two about growth hacking. 
Experience working with:
BI using Power BI/Looker Studio & SQL & Python for exploratory and predictive data analytics.
Meridian, Lightweight MMM, Robyn, PYMC (open-source MMM tools)
Causal Inference, Causal Impact, Difference-in-Difference, Synthetic Controls

I have just completed a video walkthrough on implementing Google's open source Matched Markets package for geo-experimentation, the latter itself being a gold standard in incrementality testing.

To put it simply, matched markets is a method that helps you find comparable markets so that you can divide your geos into test and control groups reliably without depending on pure intuition or surface level metrics. 

In the video, I will show step by step how to:
✅ Set up a Google Colab notebook
✅How to prepare your input datasets (cost/sales data & geo eligiblity)
✅How to configure your input parameters 
✅How to set up the design of the test 
✅How to interpret the results
✅How to visualize the daily point lift estimates as well as cumulative incremental lift
& more 

To get the video, just 
"Like" this post and comment "geo-testing" or "geo-experiment" or "matched markets" 
And I will share the link with you along with the Colab notebook 
#marketinganalytics #marketingmeasurement #marketingscience #geoexperiment #timeseriesanalysis #datascience #businessanalytics #marketingeffectiveness

Kishalaya Mukhopadhyay

Marketing Data Science | Paid Media | Analytics | Martech | Marketing Measurement & Effectiveness | Proud Cat Parent

Was exploring the Bayesian Geolift code by CausalPy by PyMC Labs and liked how elegant it is. 
Like all incrementality tests using quasi-experimental methods, it relies on estimating the counterfactual and in this case it does so by using the Synthetic Control Method. 

So this is what it involves: 

1️⃣ It builds a weighted combination of control markets 
2️⃣ It compares actual data to this "synthetic data" that didn't get the intervention
3️⃣ The gap = The true causal impact

Instead of guessing, you get:
Exact lift: "Campaign drove $2.3M in incremental sales"
Statistical confidence: "94% certain the effect is between $2.1M - $2.5M"
Percentage impact: "Campaign generated a 5.2% lift"

💡 CausalPy's approach gives you full uncertainty quantification. 
You can say "There's a 99% probability this campaign was profitable" - not just point estimates.

Will be doing a video walkthrough of the process with a custom example (not just the explainer code already in the docs). If interested, leave a comment saying "CausalPy" below :) 
#marketinganalytics #marketingmeasurement #geolift #geoexperiment #geoholdout #incrementality #marketingscience #marketingeffectiveness

PyMC Labs

🔬 Marketing Science in Action: Designing a Geo Holdout Experiment Just implemented a Matched Markets design using Time-Based Regression (TBR) for incrementality testing. 

THE PROBLEM: Traditional attribution can't answer: "What would have happened without the ad spend?" a.k.a the counterfactual.

THE SOLUTION: 
Create a controlled experiment at the geographic level
→ Treatment geos get incremental budget 
→ Control geos stay at baseline 
→ The difference = True causal impact 

THE METHODOLOGY: 
1️⃣ Data: 20 markets, 119 days historical performance 
2️⃣ Algorithm: Matched Markets optimization to maximize correlation 
3️⃣ Validation: Statistical tests 
4️⃣ Output: Treatment/Control assignments + statistical guarantees 

THE RESULTS: 
✅ 93.6% correlation (exceptional matching) 
✅ 90% confidence, 80% power 
✅ Can detect minimum 3:1 iROAS 
✅ All diagnostic tests passed 

THE MATH THAT MATTERS: 
With high correlation, variance reduces by (1-ρ) 
→ 0.936 correlation = 93.6% variance reduction 
→ Meaning: 4x more statistical power vs random assignment 
→ Translation: Need 1/4 the budget to prove the same effect

WHY THIS BEATS TRADITIONAL APPROACHES: 
❌ Last-click attribution: Correlation, not causation 
❌ MMM: Takes 1-2 years of data, slow insights 

✅ Geo experiments: Causal 'proof' in 4-8 weeks 

Built using Google's open-source matched_markets library in Python.  

For marketing analytics teams: This is the gold standard for incrementality. 

Time to move beyond pure correlation. 

If you want a step-by-step video walkthrough of the implementation, just 
Like this post
Comment "matched markets" or "TBR" 
and I will share the link 

#MarketingScience #CausalInference #DataScience #IncrementalityTesting #MarketingMeasurement #Python #MarketingAnalytics

📢 2025 is soon coming to an end, I so decided to reshare the most important videos of 2025 from my YouTube channel with my amazing analytics and marketing community (especially those who have missed out). 

Almost all my content is free, based on my experience and experiments as a marketing analytics professional. Gives me immense pleasure when others find my content useful. 

These videos cover topics like 

✅ Bayesian Statistics 
✅ Meridian MMM
✅ GeoLift and CausalImpact 
✅ Lift and Incrementality Tests 
✅ GA4-BigQuery Attribution Models 
✅ Marginal ROAS calculation 
& more 

If you want the document with all the links readily available for your convenience, 
just "Like" this post so that it reaches more people 
and comment "video" and I will share the doc with you 
#marketinganalytics #marketingmeasurement #marketingmixmodeling #attribution #marketingscience #businessanalytics

My Top Videos From 2025

This is what the time series modeling pipeline in BigQuery ML looks like. 

The ARIMA model is the most computationally expensive module, which is why the model is named ARIMA_PLUS.

The ARIMA_PLUS modeling workflow carries out the following steps:
✅ Identify the time-series data frequency.
✅ Manage irregular time intervals.
✅ Resolve duplicated timestamps by averaging their values.
✅ Fill missing points using local linear interpolation.
✅ Detect and correct spike or dip outliers.
✅ Find and adjust sudden level shifts.
✅ Identify and account for holiday effects.
✅ Capture multiple seasonal patterns using STL (Seasonal-Trend decomposition with Loess) and extend the seasonal component with ETS (double exponential smoothing).
✅ Estimate the underlying trend with an ARIMA model, using the auto. ARIMA procedure for hyperparameter selection. This process trains many candidate models in parallel and chooses the one with the lowest Akaike Information Criterion (AIC).

I made a video recently on AI functions in BigQuery, that leverages the TimesFM model; if interested, check out the link below:
https://lnkd.in/d3BQHvBh

#marketinganalytics #bigquery #ml #machinelearning #datascience #dataanalytics #dataengineering #forecasting #timeseries #predictiveanalytics

This is a causal diagram (DAG) from a PYMC Labs blog (link given in comments) showing how different factors influence sales units through a utility-based demand model. This structure forms the basis of a discrete choice model.

𝐂𝐚𝐮𝐬𝐚𝐥 𝐅𝐥𝐨𝐰
Determinants of Utility:
Price (gray/observed) → Utility: Higher prices typically decrease utility
ACV (gray/observed) → Utility: All Commodity Volume approximates the availability of the product in the market. 
Base (unobserved) → Utility: Baseline preference/brand strength independent of price
Price Elasticity (unobserved) → Utility: The sensitivity of demand to price changes

𝐃𝐨𝐰𝐧𝐬𝐭𝐫𝐞𝐚𝐦 𝐄𝐟𝐟𝐞𝐜𝐭𝐬:
Utility → Market Share: Consumer utility determines competitive market share (likely through a discrete choice model like multinomial logit)
Market Share + Total Sales → Sales Units: Your product's share of the total market determines actual unit sales

#discretechoicemodel #marketinganalytics #marketingscience #marketingmeasurement #salesanalytics #causalinference

I recently posted about the AI.FORECAST and AI.EVALUATE functions in BigQuery. An even newer, currently in Beta, function is AI.DETECT_ANOMALIES. 

This lets you detect anomalies in time series data by using BigQuery ML's built-in TimesFM model. 

Anomaly detection is a pretty big field within time series analysis, with many bespoke approaches and even packages already available, so it would be interesting to see just how well this pre-built functionality performs compared to the legacy methods. 

I was playing around with the sample code that BigQuery has given in its documentation (link shared below) and I found a quick little tweak that you can use to visualize better.

The sample code was using a Boolean flag for anomaly, I instead converted to a binary flag, making it possible to plot it along the y axis. 

Want the code? I explain this and more in a video on all the new AI functions in BigQuery. Just comment "AI" or "BigQuery" below and I will share the link. 

#marketinganalytics #bigquery #forecasting #predictiveanalytics #timeseries #marketingmeasurement #businessanalytics

These are my predictions for developments within marketing, in particular whatever goes by the name "performance marketing", in 2026. 

Hear me out in case you disagree with any point.

1. Data becomes the only other arm of strategy apart from creative branding - this may sound confusing so let me explain. 

I think most will agree that data has always been very important in performance marketing, but what is changing now is that the entire journey of data, from collection (think signal engineering) to advanced analysis (think ML-based classification/segmentation, customer value prediction, event study modeling etc.) to measurement (marketing mix modeling, incrementality testing, AB testing) is becoming deeply entrenched in strategy, so much so that the complete data journey becomes indistinguishable from strategy itself. 

Hitherto, platform-specific aspects/skills used to be important in strategy too (like "domain knowledge" about Google or Meta Ads), but that will cease to be the case with automation of optimization actions reaching its zenith. Here's a version of Murphy's Law adapted to marketing: 'Anything that can get automated inside the ads platforms, will get automated". 

The only other element of strategy that remains completely separate and will perhaps become even more important after the Meta Andromeda update (diversity >>> iteration) is creative branding. 

2. Incrementality becomes an indispensable part of marketing ops - Rather than being a good-to-have add-on, incrementality will become the litmus test for growth marketing. Without it, strategies can no longer be approved; "successes" no longer be celebrated. 

3. As a corollary to the last point, except for exceptional single platform specialist agencies, 360 degree agencies with all-round capabilities (genuinely skilled ones, not low-quality shabby "we do everything for peanuts" types) will outperform. This is because of their innate ability to accept "Truth" via incrementality and shift budgets and actions to platforms or tactics that DO work without any selfish or self-preservation related agenda coming in the way. 

Mediocre single-platform agencies will become more conservative, adopting "protectionist" or "apologist" or "denialist" stances when presented with strong evidence from scientific tests that show other channels are doing better. And eventually fizzle out / perish. 

Questions? Disagreements? Additions? 
Please add in the comments - let's discuss! 
#marketingstrategy #performancemarketing #marketinganalytics #marketingmeasurement #causalinference #mediaplanning #adagency

🚀 𝐌𝐞𝐭𝐚 𝐉𝐮𝐬𝐭 𝐑𝐞𝐛𝐮𝐢𝐥𝐭 𝐈𝐭𝐬 𝐄𝐧𝐭𝐢𝐫𝐞 𝐀𝐝𝐬 𝐁𝐫𝐚𝐢𝐧 — 𝐇𝐞𝐫𝐞’𝐬 𝐖𝐡𝐲 𝐌𝐚𝐫𝐤𝐞𝐭𝐞𝐫𝐬 𝐒𝐡𝐨𝐮𝐥𝐝 𝐂𝐚𝐫𝐞

Meta is rolling out a new generation of AI systems that fundamentally change how ads are delivered, ranked, and optimized. And the upgrades aren’t small — they’re some of the biggest leaps Meta has made in years.
Here’s a quick breakdown of what’s happening and why it matters. 👇

💡 1. 𝐌𝐞𝐭𝐚 𝐆𝐄𝐌 — 𝐓𝐡𝐞 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐯𝐞 𝐀𝐝𝐬 𝐑𝐞𝐜𝐨𝐦𝐦𝐞𝐧𝐝𝐚𝐭𝐢𝐨𝐧 𝐌𝐨𝐝𝐞𝐥
✅ Think of GEM as Meta’s new “super-brain” for ads.
✅Built like a foundation model (similar to LLMs, but for recommendations)
✅Trained on thousands of GPUs
✅Processes massive volumes of signals
✅Delivers far more relevant, personalized ad experiences
✅For marketers, this means smarter delivery and better performance with the same creative and budget.

💡 2. 𝐌𝐞𝐭𝐚 𝐋𝐚𝐭𝐭𝐢𝐜𝐞 — 𝐓𝐡𝐞 𝐑𝐚𝐧𝐤𝐢𝐧𝐠 𝐄𝐧𝐠𝐢𝐧𝐞
This is the “giant library” that organizes and ranks billions of content and ad signals.
With updated architecture, Lattice better understands context, behavior patterns, and intent — a huge unlock for more accurate targeting and optimization.
✔️ More predictable outcomes
✔️ Better signal interpretation
✔️ Stronger learning from user journeys

💡 3. 𝐀𝐧𝐝𝐫𝐨𝐦𝐞𝐝𝐚 — 𝐓𝐡𝐞 𝐒𝐲𝐬𝐭𝐞𝐦 𝐓𝐡𝐚𝐭 𝐌𝐚𝐤𝐞𝐬 𝐈𝐭 𝐀𝐥𝐥 𝐖𝐨𝐫𝐤
✅A new hardware/software stack that gets recommendations to users faster and more efficiently.
✅This improved infrastructure allows Meta to scale training 23× faster, meaning models improve sooner and more often.
✅For marketers: your campaigns benefit from a constantly improving learning system.

💡 4. 𝐒𝐞𝐪𝐮𝐞𝐧𝐜𝐞 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 — 𝐌𝐨𝐝𝐞𝐥𝐢𝐧𝐠 𝐓𝐡𝐚𝐭 𝐔𝐧𝐝𝐞𝐫𝐬𝐭𝐚𝐧𝐝𝐬 𝐌𝐞𝐦𝐨𝐫𝐲
Meta’s newest modeling technique learns from behavioral sequences over time — not just one-off actions.
✨ Better prediction of intent
✨ Better understanding of how people move through funnels
✨ More accurate delivery to high-value users

🚀 Why This Matters to Marketers
Meta is quietly making its ads ecosystem smarter, faster, and more predictive.
The bottom line?
✔️ Better performance
✔️ More efficient spend
✔️ More accurate predictions
✔️ Higher-quality audiences
✔️ Smarter optimization with fewer manual tweaks

#performancemarketing #marketinganalytics #attribution #advertisingeffectiveness #paidmarketing #paidmedia #facebookads #metaads #marketingmeasurement #growth

Found this chart on the State of PPC 2025 report on ppcsurvey.com. It tells a very interesting story - it shows how satisfaction with various AI functionalities compare with respect to the depth of usage. Interestingly, the survey found that despite frequently used, AI tools that generate insights and recommendations are not as useful to PPC / paid marketing professionals. 

This shows despite all the technology available, AI hasn't been able to make too much progress (yet) when it comes to understanding context and giving suggestions that actually help decision-making. 
Miles to go, friends, miles to go.....
#AI #AIinmarketing #performancemarketing #ppc #googleads #metaads #facebookads #automation

If you are a 6 figure+ brand and you are struggling to scale, it is most likely an incrementality problem. 

I know you'll say: "This isn't about analytics or measurement - this is about strategy".

I get it. But the reality is that in an increasingly saturated market with rising competition/CPCs/CPMs on one hand and the decline of attribution's accuracy, incrementality is itself the new strategy! 

Because what good is a "strategy" if you don't even know the marginal returns from a particular channel. How will you decide where to invest. i.e. you won't know how to implement your "strategy"!

As Meta's new report correctly points out, it isn't just good to have - incrementality should be treated as a "new line item" in your measurement stack. 

As the attached image shows very well, rule-based attribution models are the fastest but have the lowest level of statistical rigor and moderate scope. Experiments sit on the other end of the spectrum, with minimal scope but very high statistical rigor and moderate speed. 

This is from Meta's new report titled "Building a suite of truth" - but I think I will differ in one aspect - statistical rigor is equal when it comes to MMM or custom attribution models, whereas speed is much faster in experiments/quasi-experiments commonly used in incrementality testing. 

What's been your experience? At what stage is your brand or agency in this hierarchy of truth?

The link to the report is given below:
https://lnkd.in/dcxhH8yG

#MetaAds #FacebookAds #GoogleAds #performancemarketing #incrementalitytesting #marketingmeasurement #marketingscience #marketinganalytics

AI functions from BigQuery are a rage right now, but let's dive a little deeper. 

Their most recent addition are functions like AI.FORECAST. They use the built-in model called TimesFM. This is a time series forecasting foundation model.

Foundational models (they simply mean they are pre-trained on huge amounts of data) may sound great on paper but they come with their unique set of challenges. 

Watch this video to know more. I have left the research papers in the comments. 

#marketinganalytics #marketingmeasurement #performancemarketing #digitalads #marketingscience #bigquery #forecasting

Every data/measurement expert wishes there were at least one CMO who looks at them like Trump looks at Zohran Mamdani here. 😄 
#agencylife #marketingmeasurement #marketinganalytics #attribution #marketingmemes

Black Friday weekend's here, let's go muthafuggas 🤣 🤣 🤣 🤣 
He who refreshes faster wins 😉 
#blackfriday #bfcm #performancemarketing #googleads #metaads #marketingmemes

Cloudfare at it again - in & out 
#shopifydown #cloudfare #marketingmemes #shopify #performancemarketing

The different stages of anxiety - if Michael Jackson were a Google Ads marketer.... 😄 IYKYK
#googleads #performancemarketing #ppc #paidmedia #paidmarketing #marketingmemes

taplio

Kishalaya Mukhopadhyay's Linkedin Analytics

Kisholoy Mukherjee

Performance ScoreKisholoy Mukherjee's LinkedIn Performance

Kisholoy Mukherjee's Best Posts (last 30 days)

Ready to grow your LinkedIn presence?

taplio

.css-6ytdur{background-image:linear-gradient(284deg, rgba(39, 43, 72, 1) 0%, rgba(33, 47, 183, 1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Kishalaya Mukhopadhyay's Linkedin Analytics

Kisholoy Mukherjee

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Kisholoy Mukherjee's LinkedIn Performance

Kisholoy Mukherjee's Best Posts .css-1emoo84{font-size:var(--chakra-fontSizes-md);font-weight:400;color:var(--chakra-colors-gray-600);}(last 30 days)

Ready to grow your LinkedIn presence?

Kishalaya Mukhopadhyay's Linkedin Analytics

Performance ScoreKisholoy Mukherjee's LinkedIn Performance

Kisholoy Mukherjee's Best Posts (last 30 days)