Maximizing Amazon DSP Efficiency by Solving Constrained Budget Allocation Problem

With the launch of Pereptua’s Amazon DSP Optimization Software, we are excited to roll out the support for Line Item Budget Optimization. Manually monitoring and adapting budget allocation strategies to the performance trends across multiple line items can be cumbersome and suboptimal. Perpetua's Amazon DSP Line Item Optimizer is designed to take into account the historical performance trends along with the configurations such as order KPI, total order budget and budget pacing.

How does it work ?

Perpetua's Amazon DSP Line Item Optimizer supports two modes of optimization:

1. Maximize Efficiency

2. Optimizing Efficiency While Maximizing Spend

In the “Maximize Efficiency" mode, the optimizer focuses solely on improving the overall efficiency of the order. Whereas in the “Optimizing Efficiency While Maximizing Spend” mode, the optimizer solves a dual optimization problem with a goal of maximizing performance while ensuring that budget pacing is in-line with the expected spend rate.

The optimizer supports a wide range of efficiency KPIs including:

• ROAS

• TOTAL ROAS

• eCPDPV

• NTB ROAS

• NTB Cost Per Purchase

• NTB Percentage of Total Purchases

• Video Completion Rate

• Click Through Rate

• eCPM

• Cost Per Click

Let’s dig into it:

To understand this better, let us consider an order with a total budget of $500 that is set up to “Maximize Efficiency” with a goal KPI of ROAS.

The Greedy Approach:

It can be seen that line item - ‘LI-4’ is the most efficient line item whereas line item - “LI-0” is the least efficient one. In a greedy world, we would want to allocate all of $500 to “LI-4”.

Why being Greedy is Suboptimal?

- The greedy approach of allocating budgets assumes a linear relationship between budget and efficiency. In other words, it assumes that an x% increase in budget, and hence spend, would result in a sales lift of same magnitude. This is generally not the case.

- It is also important to note that advertising spend on a line item is a function of factors like frequency, viewability rate, audience size, etc. As a result, simply allocating budgets do not guarantee spends, which makes the greedy approach less desirable.

- Finally, from a pure statistical standpoint, the greedy approach assumes sample efficiency to be equal to the population efficiency. However, given the stochastic nature of auction dynamics, a frequentist approach to estimate efficiency can lead to suboptimal performance.

Data Driven approach:

Perpetua’s Amazon DSP Line Item Optimizer is designed to solve a constrained optimization problem with the objective of maximizing the overall order efficiency. Instead of optimizing for global optima in one shot, it aims to iteratively inch closer to the global optima.

For instance, an iteration of our Amazon DSP budget optimization would result in an allocation as follows:

It can be seen that the optimizer rewards the more efficient line items “LI-3” and “LI-4” with budget upgrades while penalizing the less the efficient line items based on realized ROAS. However, unlike the greedy approach, Perpetua’s Amazon DSP Line Item Optimizer takes a more measured approach to allocating budgets across line items. Our iterative approach solves the constrained optimization problem, while gathering more evidence on each iteration enables us to optimize the order in a more robust way.

Impact:

- As a part of our alpha testing, we enabled DSP Line Item Budget Optimizer for 385 orders across different advertisers.

- The budget utilization improved by an average of 18% across participating orders.

- The average daily spends increased by 16% with over $68000 in additional monthly ad-spends.

- Efficiency improved by more than 20% for 180 DSP orders out of the total 345 orders that participated in the alpha testing.

- More than 70% of the participating orders witnessed a boost in budget utilization as well as efficiency.

Start Optimizing your DSP Advertising today by requesting a demo with our team!