Pacing Alerts for Ensuring efficient delivery of programmatic advertising budgets

Introduction

Almost every impression we see now is being delivered through programmatic advertising. Programmatic advertising has taken the ad tech market by storm with better optimizations and by automating most of the bidding process. Given this automation, one of the key factors that are in our control to optimize and run effective campaigns is Pacing. To understand what pacing is all about, let’s look at an example:
Suppose a campaign was booked with a specified amount of budget. Now a simple approach would be dividing the budget equally for every day throughout the campaign flight duration and spending according to that. But what if there are not as many required conversions as needed by end of the flight. In that case to achieve the required conversions the trader might increase the budget for the last few days to secure more impressions and might overspend thus exceeding the booked budget eventually leading to reduced ROI.

Also, this is rarely the case as traders generally tend to vary the pace of the budget allocated to the campaigns depending on how the campaign is performing. The problem with this approach is that traders tend to allocate most of the budget to campaigns in the starting phase so they can rely on trailing conversions which is the ultimate goal of the campaign but then they are left with little or no budget by the end of the campaign and they might need to overspend in case they are not able to hit the required KPI. Here’s where pacing comes into play, using the right pacing we can achieve the campaign goal well within the specified budget.

Ad pacing refers to the automatic balancing of impressions for a Flight or Ad so that the campaign hits a desired metric over the course of its lifetime.

In this blog, we will discuss a daily alerting mechanism for traders around overpacing or underpacing and also recommend the ideal impressions/cost the campaign should spend for the remaining flight duration of the campaign.

Data

In order to calculate the required pacing amount for each day for a campaign we need the following configuration details of a campaign: Flight of the campaign (for how many days it has to run), campaign budget (both in terms of cost and impressions), daily spending stats of impressions and cost (to calculate things at a daily level), media cost projections, current CVR or CPM or whatever KPI is being optimized/tracked as a goal of the campaign, etc.

The methodology used for alerting

To deep dive into the methodology and rules we use to alert the traders let’s look at some key definitions first:

• Required impressions/cost: This is the number of impressions/cost required for a campaign to spend for the remaining of the flight per day calculated by dividing the remaining impressions/cost with the remaining days in flight.
• Overpacing : If the campaign spent more than what it was intended to on the last day it is termed to be overpacing for that day.
• Underpacing : If the campaign spent less than what was intended, it is termed to be underpacing for that day.

The above pacing stats can be calculated with respect to both impressions as well as the cost for every campaign. Subsequently, based on the criticality of each campaign and the degree of under/over pacing, and some rule-based selection criteria (described below) the alerts are pushed to traders to act upon.

• Select the top 15% of all the campaigns based on the absolute pacing indicator of impressions (takes into account the most overpacing/underpacing campaigns)
• If Campaign is overpacing wrt impressions and underpacing wrt cost then filter out such campaigns and don’t show the alerts for these campaigns
• If Campaign has exhausted its budget then required impressions will be a negative value for the rest of the days. We exclusively alert the traders for all such campaigns with a message indicating that the campaign has exhausted its assigned budget and it should stop delivering

Fig. Tableau Dashboard with ideal pacing for the rest of the flight of the campaign

The primary end goal of the alerting process is to reduce the volatility in budget spends which particularly happens towards the end of the campaign flight. We have seen some positive initial results so far to that end. Going forward, the plan is to use forecasting-based methods to include specific recommendations around actions to take to address under/over pacing as part of the alerting mechanism. Some of the techniques we plan to explore include unsupervised and supervised machine learning, reinforcement learning, time series analysis, etc. Stay tuned!