Ensuring Even Ad Spend on the Zalando Homepage: How Our New Bidding Algorithm Maximizes Value for Advertisers and Shoppers

Zalando Marketing Services (ZMS) is Zalando's advertising platform. It helps brands create and manage campaigns on Zalando, increasing their visibility and improving performance at every stage of the marketing funnel, from awareness to purchase, within the Zalando marketplace.

At ZMS, we're constantly innovating to optimize the advertising experience on Zalando homepage. A key element of this is ensuring sponsored ads receive optimal exposure while maintaining a seamless shopping experience for Zalando users. This article dives into the challenge of achieving even ad spend and introduces our new bidding strategy designed to address it.

Homepage content selection flow with real-time bidding

The Challenge of Uneven Ad Spend on the Homepage

Imagine you're an advertiser running a campaign on the Zalando homepage. Your goal is to maximize brand awareness by getting as many user views as possible for your ad. You allocate a specific advertising budget for your campaign within a defined timeframe.

However, a hidden hurdle exists: uneven ad spend. Currently, ad placements on the homepage are determined by a real-time bidding system. This system can lead to situations where your ad budget is exhausted early in the campaign period, limiting your potential reach.

The Consequence? Lower-than-desired ad views and potentially a missed opportunity to connect with your target audience.

The ZMS Solution: Introducing the Adjustment Factor Bidding Strategy

Our ZMS product team understands the importance of efficient ad spend for both advertisers and Zalando. That's why we've developed a new bidding strategy, based on closed feedback loops.

Imagine you're on a road trip in an electric car. You have a set amount of battery power to cover a specific distance. To reach your destination efficiently, you can't just use all your power at the beginning and speed down the highway. Just like with uneven ad spend, this would leave you stranded before reaching your goal. Instead, an electric car on a long trip with varying terrain needs to adjust its speed throughout the journey. It might go faster on flat stretches to maintain an average speed and conserve battery for steeper hills. Similarly, our new bidding strategy avoids the "all-or-nothing" approach, ensuring advertising budget is used efficiently throughout advertisers’ campaigns to maximize reach.

Here's how it works:

• Budget allocation: advertising budget is allocated based on the traffic forecast of customers on the Zalando platform and distributed in hourly buckets.
• Monitoring budget allocation: The system continuously tracks the remaining budget for the campaign relative to the expected (the expected amount is the amount remaining at any give time if the budget were to be spent evenly over the hour).
• Dynamic bid adjustments: Based on this comparison, the bidding strategy automatically adjusts the advertiser ad's bid price. If the advertiser’s campaign is overspending, the bid is lowered. Conversely, if it's underspending, the bid is increased.
• Equilibrium through feedback control: This dynamic adjustment process ensures the ad budget is spent evenly, maximizing the number of potential viewers throughout the campaign duration.

Converging to even spending

Technical Deep Dive

Let's now break down the math behind our new bidding strategy. Let \(t\) be the fraction of the hour passed at a given point in time. Let \(spent_t\) be the fraction of budget spent at time \(t\), and \(spent_t^{even}\) the ideal fraction of budget spent at time \(t\) with even spending. Note that \(spent_t^{even}\) is equal to \(t\) due to even spending. The ratio between these values, \(r_t = spent_t / spent_t^{even}\), captures how close we are from even spending, and we want to achieve a value close to \(1\).

In our previous bidding strategy, bid was directly proportional to the following factor:

$$ 1 - spend_t \cdot (1 - t) = 1 - r_t \cdot t \cdot (1 - t) $$

Taking the derivative of the factor w.r.t. \(t\), we obtain \(r_t \cdot (2t - 1)\), which is negative for \(t < 1/2\) and positive for \(t > 1/2\). In other words, regardless of the value of \(r_t\) (i.e. over- or under-spending), the bid would decrease in the first half of the hour (negative slope w.r.t. \(t\)) and increases in the second half of the hour.

In the new bidding formula, bid is directly proportional to the following factor:

$$ \frac{1 - spent_t}{1 - spent_t^{even}} = \frac{1 - r_t \cdot t}{1 - t} $$

Taking the derivative of this factor w.r.t. \(t\), we can see that it is positive for \(r_t > 1\) and negative for \(r_t < 1\). That is, this formula ensures that:

• The bid increases when underspending.
• The bid decreases when overspending.
• The bid remains constant when spending is even.

As a result, this bidding strategy converges to an even spending over time and achieves an equilibrium price under a given market condition (supply, demand, competition).

Benefits for Advertisers and Shoppers

By leveraging the new bidding strategy, advertisers gain several key advantages:

• Maximized reach: Achieve a more even distribution of ad views throughout your campaign, leading to a higher likelihood of reaching your target audience.
• Cost efficiency: Reduce your cost per view (CPV) by ensuring we use an optimal bid (not the max bid), while meeting your goals.
• Greater value: Get more value from your advertising budget, leading to a potentially higher return on investment (ROI).

Shoppers also benefit from this strategy:

• Improved relevance: By allowing for a wider range of ads to compete for display, shoppers are more likely to see ads relevant to their interests.
• Seamless experience: The strategy maintains a balanced ad-to-content ratio, ensuring a smooth shopping experience on the homepage.

Validation Through A/B Testing

To validate the effectiveness of this bidding strategy, we conducted a comprehensive A/B test with budget-split. The results were clear:

• Increased advertising views: Ads using the new strategy achieved a 10% increase in views compared to the previous approach; with a linear drop of CPV of around 10%.
• Increased clicks on advertising content: the absolute number of clicks on ads increased by 23%.
• Enhanced click-through rate (CTR): The ratio between clicks and views improved by 11%, suggesting greater relevance of advertising content for Zalando customers.
• Non-significant impact on metrics of overall Homepage customer experience; a great indication of success since we are delivering more sponsored content without harming overall homepage customer experience.

Uneven ad spend can hinder advertiser efforts and limit the value proposition for Zalando. The new ZMS bidding strategy effectively addresses this challenge by ensuring a balanced distribution of ad spend. With this approach, inspired by the principles of closed feedback loops, ZMS empowers advertisers to maximize the effectiveness of their campaigns while maintaining a positive shopping experience for our valued users.

Ad spend: old vs. new algorithm

We're hiring! Do you like working in an ever evolving organization such as Zalando? Consider joining our teams as a Software Engineer!