By James Dickman, CTO – Advocado
By Vibhav Gupta, AWS Pre-Sales Lead – Quantiphi
By Matheus Arrais, Partner Solutions Architect – AWS
Over the past few decades, there’s been a shift in user behavior; prompting marketers to explore multiple mediums of communication, such as social media, television, and OTT, to capture user attention. As a result, marketers are finding it difficult to understand the factors that impact ad conversions, and provide visibility into marketing campaigns.
To address this challenge, Quantiphi worked with Advocado, an advertising focused data platform provider, to develop a solution for ad attribution of web traffic to respective offline and online sources on AWS. The solution leverages Amazon Forecast for predicting web traffic.
In this post, we explore how marketers can leverage this solution to track and understand which source of communication works best for its target user base and use it to help plan and optimize future ad campaign spends.
Quantiphi is an AI-first digital engineering company and AWS Advanced Tier Services Partner with competencies in Migration, DevOps, Data and Analytics, and Machine Learning.
When using different communication mediums, it is critical for marketers to understand which ones work best with their target audience to plan and optimize future ad campaign spend.
Marketers need insights that enable them to attribute the lift in website/store traffic to specific channels, events, and ad features at a brand, market, and regional level. They also need to understand how factors, such as ad length, time of day, location of viewership, date, device, and external factors—like weather conditions—contribute to the success of an ad campaign.
An attribution solution is key to capturing campaign metrics and understanding the type of interactions that influence conversions in order to calculate ROI for individual campaigns.
To address these industry challenges, Quantiphi helped Advocado develop a machine learning-based custom ad attribution solution on AWS. The solution fills the visibility gaps with unique data that advertisers, media companies, and agencies can use—empowering marketers with insights on campaign effectiveness and website lift attribution.
The key differentiators of the solution include:
- Attribution as explainability: Derive insightful metrics on ad placement and ad attribution using the solution’s model explainability concept.
- Lift calculation by predicting web traffic: Calculate realistic lift values using the forecasts generated by DeepAR and forecasting model from Amazon Forecast. The forecasts are given out using historical data to estimate the number of visits on a website at a given time and location.
- Explains any number of ad features: Understand contribution of different advertisement features with respect to lift using the solution’s model on model approach. This will help understand whether a feature of an ad is impacting lift positively or negatively.
- Extracts insights from multiple ads: Perform analysis on multiple advertisements of the same brand instead of only one to determine the set of features that work best for increasing brand lift and visibility.
Customer Use Case: Advocado
Quantiphi worked with Advocado, an advertising focused data platform provider, to build a custom solution for ad attribution. The ad attribution solution enables Advocado’s customers to forecast attribution to specific channel sources to manage website traffic and optimize future marketing campaigns.
Prior to working with Quantiphi, Advocado was trying to solve the ad attribution problem using Amazon SageMaker DeepAR forecasting algorithm. Their challenge here was deciding which features of an ad to include to provide the best forecasts of web traffic.
To build a custom ad attribution solution for Advocado, Quantiphi leveraged Amazon Forecast to predict and track web traffic on customers’ websites pre and post-commercial air to capture digital lift in real-time. Quantiphi leveraged AWS Glue to transform multiple historical data sources, like audio watermarks, consumer sentiments, Google Ads, and more to ensure ML models had the right set of data.
The solution follows a two-step process:
Step 1: Time Series Forecasting Using Amazon Forecast
The solution involves cleansing and preparing the data using AWS Glue to be compatible with Amazon Forecast and setting up experiments for different channels to estimate the attribution.
For each experiment, predictors are trained with pre-built algorithms or by leveraging AutoML features to define the calculations of attribution for each channel in order to predict web traffic and optimize marketing campaigns.
Sample time-series graph and real data of prediction are shown in Figure 1 below:
Figure 1 – Website traffic sample with predicted quantiles.
Step 2: Explainability Model for Ad Attribution
Using a model-on-model approach, Quantiphi built an industry-specific explainability model to understand and attribute the lift with each feature at an individual brand level. This approach helped marketers understand the model predictions better and attribute the lift to each feature, individual ads, and brand.
The values obtained also generated insights on how the ad length, location, and airtime contribute to the lift.
Sample force plots providing event-level information are illustrated in Figure 2 below:
Figure 2 – Force plot to explain ad attribution.
Using a model exclusively for the explainability of features for attribution provided flexibility to work with multiple features of an ad. This helped marketers understand the features that work best and their importance in optimizing the overall strategy for marketing campaigns.
The solution architecture is shown below in Figure 3:
Figure 3 – Solution architecture.
- The AWS Step Functions are triggered by the Amazon CloudWatch events at regular intervals (monthly, for example). Then, the Step Functions orchestrate the execution flow.
- Step Functions initiate a glue job to transform the data to the format used in Amazon Forecast.
- The transformed data is saved in Amazon Simple Storage Service (Amazon S3).
- Then, the Step Functions sequence the AWS Lambda functions to trigger the Amazon Forecast service.
- The exported forecasts are saved in Amazon S3.
- Metrics and logs generated during data transformations and inferences are stored in Amazon CloudWatch. Amazon CloudWatch is also used for alerting once the processing is complete. These logs are used by the customer to get details on the lift predicted and contribution of each feature.
Quantiphi’s ML-based custom solution empowers Advocado customers with deeper insights to help them optimize website traffic, ad spends, and campaign strategy; including ad conversions, location-based campaign effectiveness, duration, time, and channels where the ad was played.
The solution predicts the right attribution to distribute the credits of digital lift to the right marketing campaigns and bridges the visibility gaps with unique data that advertisers, media companies, and agencies can use.
Now, Advocado’s customers can leverage this solution to gain insights on campaign effectiveness and website lift to boost campaign engagement. The solution is also designed to work with additional time series and third-party data in the future.
Overall, the solution helped advance Advocado’s machine learning initiatives through the development of a deep learning neural net to attribute incremental digital lift to TV airings.
As new means of consumer engagement emerge, it becomes increasingly essential for marketers to adopt an ad attribution solution that can help attribute lift to specific channels and optimize their marketing strategy.
Quantiphi and Advocado worked together to build an ad attribution solution on AWS that not only forecasts web traffic and attributes to individual channels, but also understands features that work across advertisements to optimize future campaigns.
To learn more about Quantiphi’s solutions, contact the team.
Quantiphi – AWS Partner Spotlight
Quantiphi is an AWS Advanced Tier Services Partner and leading artificial intelligence-first company driven by the desire to solve transformational problems at the heart of the business.
*Already worked with Quantiphi ? Rate the Partner
*To review an AWS Partner, you must be a customer that has worked with them directly on a project.