How Greenlake helped Delivery Hero enhance their marketing budget allocation


Our client is a multinational conglomerate in the food delivery space. Their marketing tech department was starting to systematize their data science functions and Greenlake was tasked by their Director of Engineering for Marketing Tech to help develop a stable marketing spend optimization model and to help audit their existing vendor for TV spend attribution.


Our team developed a stable scalable attribution model for marketing spend and developed optimal spend recommendations yielding 5-10% potential improvements in each country.

The audit of the vendor helped save a considerable amounts in year on year direct costs and avoided misdirection of a significant amount of marketing budget.


Marketing channel spend modeling and optimization

A recurring theme in organizations with large marketing budgets is the optimal budget allocation amongst various marketing channels (Facebook, SEM, display). Various models can be envisioned to address this problem (from simple linear models to non-linear ones that address aspects such as the fact that many channels saturate as the amount of spending on them increases).

The challenge was to develop a short term spend model that
  • is generalizable to various markets across the world
  • is stable across time (in the absence of dramatic shifts in market conditions, the model should not recommend dramatically different strategies each month)
  • enables readjustments of marketing spend in a manner that is more effective (higher acquisitions/orders per dollar invested)
Hence the first part of the engagement required the development of a model that accomplished the goals above.


This class of problems requires a deep understanding of mathematical modeling, constrained numerical optimization, machine learning and data science skills. The class of models used in this domain are unavailable in standard tools like scikit-learn, tensorflow, MxNet. Hence solving this problem requires an ability to create a scalable solution to a core business problem from basic principles - a challenge that Greenlake excels at.

In collaboration with the client team, we developed a model which
  • yielded stable short term spend impact models that had predictive accuracy
  • enabled rapid code reviews and enabled faster transition from development to production environments.
In addition, a set of optimization routines created around this model helped obtain systematic answers to the following
  • how to spend a fixed budget across the available channels in an optimal manner
  • how to spend a fixed budget across the channels while being constrained by bounds on the spends in each of the channels


The project thus yielded a stable and scalable model which was extensible to a number of spend channels. The recommendations improved spend effectiveness across various countries by 5-10% thereby assisting the operational effectiveness of marketing managers.


Auditing TV spend effectiveness measurement vendors

In marketing, measuring the effectiveness (whether short term or long term) of money spent on TV advertising is a challenging problem. It is made more complex owing to the way incentives are aligned in the marketing solution vendor space - indicating that a channel has no/low spend effectiveness would invalidate a clients use of a tool that measures spend effectiveness on that channel. The client had been using a solution from a vendor focusing on TV, and was uncertain as to its validity.

Hence the mandate for the engagement was to audit a vendor's - without having much access to its core algorithms (as the details are proprietary to the solution vendor). This is further complicated by the fact that there are no actual "ground truth" measurements to validate these metrics (!)


The engagement team developed an extremely elegant set of experiments which used the vendor tools measurements regarding the effectiveness of various TV spots - to demonstrated a lack of consistency of the tool with itself. In essence, the experiments devised by the team at Greenlake indicated that the numbers being generated by the vendor tool for each individual TV spot were inaccurate/inconsistent by several orders of magnitude (over 200%!).


The audit outcomes thus invalidated the measurements from a TV-ad spot effectiveness measurement tool used by firms across the globe. This revealed a significant savings opportunity for the client (a yearly license fee in direct savings and further savings from the re-allocation of funds in directions unbiased by incorrect indicators). In addition it revealed an open problem, enabling the client to prioritize their efforts on an opportunity which, if addressed, would provide significant competitive advantages.

Have a similar problem?

If your firm is focused on the market-tech space, let's discuss opportunities to accelerate your business.