Data Science Expert Working For Amazon Ads Reveals Her Top Secrets For Unlocking $30 Million In Real Marketing ROI

Around the world, wasteful spending has been the bane of the advertising industry as corporate executives pour billions into ad campaigns across the massive, multibillion-dollar industry, yet they see no tangible results. Instead, what they see are just clicks, impressions, and leads—vanity numbers that do not translate to real revenue or profit. A recent Alvarez & Marsal report, “Drowning in Data, Starving for Insight,” confirms this huge threat: 65% of CEOs do not trust their chief marketing officers (CMOs) when it comes to being financially accountable due to their inability to directly link spending to revenue. 

So, this is no longer an issue about measurement, but that of unproven ROI, which has now made any marketing budget be seen as a liability. This systemic failure, therefore, requires a ruthless response—the cold, tough reasoning of data science, a trait that is lacking in traditional marketing but present in Shruti Dash, an analytics and insights consultant at Amazon Ads. Starting her career in engineering before switching to data science, Shruti did not just work her way up, but her background, which is deeply rooted in hard science, has paved the way for her. At Amazon Ads and earlier at Flywheel, she eliminated guesswork and brought clear data-driven evidence to even the trickiest marketing budgets, creating AI-powered causal models that uncover what truly causes a sale instead of just what happened last. Some of these strategies were formed on the basis of her research articles,“ An Econometric Framework for Estimating the Joint Elasticity of Advertising and Promotions on Retail Sales” and the “Challenges and Prospects of Using Synthetic Control Groups in Measuring Digital Advertising Effectiveness: A Data Quality Perspective.” Both papers underscore her immense contributions to the growing body of research in her industry. In recognition of these efforts, Shruti received the Data-Driven Product of the Year award in the Advertising Tech category at the just-concluded American Business Expo Award, a prestigious global competition honouring innovations, leadership, and impact in business and technology.

In this exclusive interview, Shruti Dash shares insights on the model she applies to turn messy data into solid evidence, addressing the 65% CEO trust-gap issue, the specific methods and causal reasoning tools that influenced ad spend with Amazon, and why the future of marketing success will not just depend on showing results, but will involve the use of AI and econometrics to reason like a data scientist.

You started your journey in electrical and electronics engineering, shifted to analytics. Now, you’re focused on delivering real marketing ROI for big brands. How did that shift come about? And how did your studies and certifications help you make this kind of impact?

Studying engineering taught me to view the world as interconnected systems. This also goes for data analytics, because when datasets contain bias or errors, business results can’t be trusted.

So, I wanted to apply the same kind of structure to how I make decisions. This made me pursue a master’s degree in quantitative management at Duke University’s Fuqua School of Business, North Carolina, where I specialised in business analytics. While I was there, I studied econometrics and machine learning. I also learned how to evaluate uncertainty and understand complex relationships.

In order to enhance my technical skills further, I earned certifications in Python, R, SQL, and data science. These courses helped me use the concepts I learned to solve problems in real life. They also allowed me to build, test, and apply models myself. All these things, therefore, played a huge role in every success I have achieved so far.

That’s an interesting journey! At Duke, you also learned how to combine engineering thinking with econometric analysis. How did those skills help you achieve real-world results at Flywheel, boosting revenue by $1.5 million?

When I got to Flywheel, the internet marketing service was growing so fast; it needed systems to bring fragmented data together. So, I helped it set up an analytics platform using Snowflake, a cloud platform for storing and analysing data, and connecting it to Streamlit dashboards for instant data visualisation. We then used AWS Lambda, a serverless compute service, and Step Functions, a serverless orchestration service, to automate data collection, cutting processing time significantly.

Next, we rolled out uplift models with XGBoost and Random Forest, separating actual incremental effects from regular sales. While XGBoost is an open-source machine learning library, Random Forest is an ensemble ML algorithm. So, combining them sharpened targeting precision considerably and generated extra ad investment.

While at the company, I also began exploring the early applications of AI in marketing. I examined how predictive tools could uncover hidden factors behind customer actions way ahead of standard measurements. 

The results validated all the lessons I learned at Duke: structuring systems correctly makes it possible to get consistent insights and track profits.

Now, you’re replicating your Flywheel successes at Amazon Ads. How have AI tools and econometric systems driven significant gains for the platform’s advertisers globally?

The size of Amazon Ads is staggering, with billions of impressions in categories such as fashion, fitness, and luxury. I have helped clients shift from post-hoc reporting to causal measurement so that they not only know what happened, but also why.

We did that by constructing a hybrid system of causal inference, econometric forecasting, and automated optimisation pipelines.

With the integration of AI-assisted solutions and econometric model processing, we were able to handle millions more data points.

The outcome has been tremendous, with ROI accuracy increasing by two digits. For advertisers they have been able to save millions through smarter spending. In the case of Amazon, it reinforces its image as a credible and data-driven partner. On a personal level, it has been gratifying to watch something I have helped design impact decisions across various markets.

Beyond that, you’re creating a new framework to optimise ad promotions. You have also published research on an econometric framework for estimating the joint elasticity of advertising and promotions and the use of synthetic control methods in digital advertising. For specialists in your field, this is a rare feat. So, how do you ensure that what you discussed in your papers is the same thing your framework does in real life? And why is this good for advertisers? 

Many attribution systems analyse ads and promotions as if they work in isolation, but that is wrong because it skews the outcomes. So, I created my framework and carried out the research to solve this problem. They measure how ads and promotions interact, whether they enhance or replace each other. In my research, I talked about how things like the quality of data, effects of spillover, and choices of aggregation can affect casual estimates. 

Then my framework uses what we call econometric modelling with elasticity estimation to understand the short-term growth and long-term impact of a brand. This approach changes how teams plan spending in industries like fashion and sports wear, where promotions overlap week after week. The initial pilots influenced ad spend with Amazon. To achieve that, spending was reallocated toward areas with genuine added value, rather than relying on superficial metrics. But it is not just about that figure. It is about understanding the real financial picture. 

So, my works talk about the same thing. When you see how each factor contributes to growth, you stop focusing on meaningless KPIs. Instead, you start working toward proven results. This is where the true change happens. Guesswork fades away. Then marketing starts working like magic.

As an award-winning data science and AI analytics expert who has done a lot of work at Flywheel and Amazon Ads, what do you think will be the future of this framework? And how do you believe marketing analytics will evolve in the near future?

Apart from fashion, we are now expanding our system optimising ad promotion to other categories such as home and consumer electronics. These areas are known to have complex promotional plans and multichannel arrangements. So, we made our system to be as flexible as possible so as to accommodate this complexity. This will allow any advertiser using Amazon Ads to determine the actual incremental value of their expenditure. This strategy would result in smarter media spending, saving millions of dollars throughout Amazon's entire global network. 

However, I have a greater mission of ensuring that these tools are available to all. I would like to develop open-source and easy-to-use versions of these techniques to allow smaller brands to gain clear insights as well. 

Having said that, I believe that adaptive measurement is the future of marketing analytics. We need tools that continually learn, identify biases, and adapt as circumstances change.

Published on: Friday, February 27, 2026, 02:38 PM IST