How We Supercharged Growth at Heetch with Mix Media Optimization

During the last few years, I’ve been lucky enough to dive deep into the world of data at Heetch, wearing the hats of both data analyst and scientist. Now, if you haven’t heard of it yet — Heetch is a fast-growing ride-sharing platform dedicated to offering a user-friendly, safe, and efficient transportation alternative. In the ride-sharing space, where growth is the North Star metric, my role goes beyond crunching numbers or running complex algorithms. It’s about uncovering the story within the data and translating it into actionable insights to propel our company’s growth trajectory.

Being part of a growth team means I’m always in the thick of it, analyzing, testing, and iterating. Each dataset is like a puzzle, harboring a hidden narrative waiting to be uncovered. It’s about connecting the dots in a way that’s as much an art as it is a science, to not only meet but exceed our KPIs, elevating the customer experience in the process.

Today, I’m excited to share a slice of our journey, specifically around a project that’s as complex as it is rewarding: automating the optimization of the media mix in a company that’s deeply rooted in digital acquisition. It’s about making every penny count, ensuring that each dollar spent on marketing is a dollar that drives growth.

In the coming sections, I’ll pull back the curtain on the intricate dance of variables and equations that is the incremental CAC and how it’s become the linchpin of our media strategy at Heetch. We’ll delve into its modeling, explore the pathways to optimizing the media mix, and discuss the impacts and implications in the ever-evolving landscape of digital marketing.

Understanding Incremental CAC: A Key to Digital Acquisition Efficiency

When you’re in the business of digital acquisition, like we are at Heetch, there’s one overarching goal: maximize the number of new passengers with the resources at hand. However, as straightforward as it may sound, the process is layered, nuanced, and demands a meticulous approach. With a plethora of digital advertising networks at your disposal, the challenge lies in deciding whether investing in network A or network B will yield the most significant impact on your KPIs.

First, let’s have a look at what happens at a single network level.

Picture this: you’re at a local market buying tomatoes. The price per kilogram is fixed. If you put 1€ in, you’ll get 1kg. If you put 2€ in, you’ll get 2kg… you get the drift. This is a simple, linear relationship between cost and quantity. If you decide to buy in bulk, the vendor might even cut you a deal, and with 9€ you could get 10kg — the more you buy, the less you might pay per kilogram.

However, this logic is not applicable in the world of digital advertising. The price won’t decrease with quantity — it operates more like an auction where the price goes up as demand increases. The more ads you want to buy, the more you’ll end up paying for each one.

This brings us to the concept of Customer Acquisition Cost (CAC), the total cost to acquire a new customer through a specific marketing channel. But here’s the catch — CAC isn’t constant. As you scale up your ad buys, each additional ad costs more but yields fewer new customers. This is just because you’ve already captured the ‘easy users’ from this network, and the network must exert more effort to acquire new customers. This effect is called Network Saturation.

So, how do we navigate this landscape where costs are rising, and returns are diminishing? Enter Incremental CAC. Unlike traditional CAC, which is basically an average CAC over all your new customers giving a broad view of acquisition cost, Incremental CAC zooms in on the cost of acquiring an additional customer. For example, if you’re working with a network that brings you 1000 new customers for 5000€, your CAC is 5€. However, your incremental CAC is the amount you will have to add if you want a 1001st customer, and it will likely be way higher (maybe 10€, maybe 50€, or more…) depending on network saturation. This concept is key as it is at the core of media mix optimization.

Now, you’re likely working with a diverse media mix to optimize your reach and impact. A media mix is the combination of various advertising channels — think of it as a well-balanced diet for your marketing strategy. Relying on a single channel is like eating only tomatoes — sure, they’re tasty, but you’re missing out on the nutrients provided by other foods.

Optimizing this mix is where the magic of Incremental CAC comes into play. A well-optimized media mix is one where the Incremental CAC is equal across all channels. It ensures that you’re not overinvesting in one channel while underutilizing another. When mix media is balanced, it means each euro of increased budget would yield the same result, regardless of which channel it’s allocated to. This is the key to ensure the mix media is optimized, and we’ll now see how we implemented this strategy for Heetch digital acquisition.

Modeling Incremental CAC at Heetch: The Inside Story

At Heetch, the task of modeling the Incremental CAC to optimize the media mix is approached with a combination of specialized tools and refined methodologies. The process, systematic and data-driven, unfolds in several well-defined steps.

The foundational element is data concerning app installs, rides, and associated costs. This data is gathered and installs are attributed using Adjust, a leading Mobile Measurement Partner (MMP). For each advertising network, weekly data is compiled, detailing the total number of new riders acquired and the total cost incurred.

The modeling of CAC is executed in Python. Each advertising network’s weekly data, spanning the previous 52 weeks, is fed into a logarithmic model (a*log(bx + 1)) to explain the relation between weekly new riders and spends for each network. This model is specifically chosen for its ability to accurately represent the saturation effect — the phenomenon where the cost of acquiring additional customers increases as the audience pool diminishes.

While all data is crucial, not all of it holds equal relevance. Recent data is often more indicative of current trends and market dynamics. To account for this, a weighting function is applied to the data within the model. This linear weighting scales the influence of each data point based on its recency, ensuring that the model is sensitive to the most current market conditions and responsive to emerging trends.

The focus then shifts to the incremental CAC. By determining the rate of change of the CAC concerning costs, the incremental CAC is derived, offering granular insights into the additional cost associated with acquiring each subsequent customer.

Optimization of Media Mix

Once you reached an efficient model of incremental CAC, media mix optimization is quite straightforward, utilizing a Python library specifically designed for this purpose. The goal is clear: distribute the advertising budget across different networks to maximize the number of new riders, given a specific total budget — in this case, we use the total budget from the previous week as a benchmark.

This optimization also results in on finding the ‘operating’ incremental CAC. It ensures that each network is allocated a portion of the budget that aligns with this incremental CAC, maximizing efficiency.

However, we’re mindful of the risk related to important changes in the budget repartition and the need for a smooth transition in reallocating budgets. To mitigate the risk related to significant changes in the mix media, budget shifts are capped at a 10% increase or decrease for each network. This approach, while not immediately hitting the optimal point, ensures stability and allows the mix media to adapt progressively. Over consecutive weeks of incremental adjustments, the budget split will anyway converge on the optimal allocation, balancing efficiency and risk mitigation.

Openings

In this exploration, we’ve focused on optimizing the media mix within networks in a specific country, but it’s essential to recognize the scalability and adaptability of this approach. The same principles and methodologies can be applied with varying granularity. For instance, within a single network, the budget allocation among different campaigns can be optimized using the exact same method. Dive deeper, and the focus can shift to the distribution of budget among ad groups within a campaign. It’s a flexible strategy, ripe for customization to fit specific contexts and objectives.

A caveat is paramount — the quality of attribution. The efficiency of the optimization is intertwined with the accuracy of attribution. In the world of data, the adage “garbage in, garbage out” holds profound significance. If the attribution is flawed, the optimization derived from it will be equally skewed. We’re navigating challenges due to Apple’s App Tracking Transparency (ATT) framework, but the strategies to mitigate its impacts are a topic for another discourse.

A potential pathway to navigate this challenge lies in media mix modeling. It’s a sophisticated approach that can potentially bypass the constraints imposed by attribution challenges, offering an alternative route to optimize the media mix effectively. However, this too, is a rich topic deserving of its detailed discussion.

In conclusion, while the journey to media mix optimization is intricate and laden with both opportunities and challenges, the incremental steps towards refinement are marked by strategic adjustments, data-driven decisions, and an unwavering focus on maximizing efficiency. The landscape is ever-evolving, and as we continue to adapt and refine our strategies, optimizing our resources utilization to maximize growth remains our North Star.