AWS Architecture Monthly

What are the general architecture pattern trends for advertising and marketing in the cloud?

• Scaled data pipelines processing first-party data for advertising intelligence use cases, such as ad measurement and event logging. Often customers are adopting serverless containerized compute to reduce the time and increments used for highly variable (spikey) ingest patterns. We also see customers adopt a stateless, streaming log analytics pattern to enable the use of Amazon Elastic Compute Cloud (Amazon EC2) Spot Instances to reduce costs and automatically scale.
• Customer data platforms that unify first-party data into “golden” user records with identity resolution for anonymized and known data. They also allow for audience segmentation, analytics, and the activation of defined audiences. Industry customers with scaled, real-time datasets typically adopt NoSQL databases like Amazon DynamoDB or Aerospike in these cases.
• Shared data lakes and data clean rooms with permissions that enable privacy-safe data sharing and collaboration between entities, without revealing underlying raw data, and in compliance with data sovereignty laws.
• Machine learning pipelines that take advantage of Amazon SageMaker to reduce heavy lifting through automation of feature engineering, training, and deployment.

When putting together an AWS architecture to solve business problems, specifically for advertising and marketing customers, what are some of your considerations?

1. Reduce development time spent on undifferentiated work, like managing commodity services like databases, Kubernetes clusters, or streaming services. This frees up your engineers to develop new features and products.
2. Automate your application deployment and testing.
3. Migrate your applications to the cloud. The ability to deploy a new service, instance type, database, or toolset in minutes allows you to experiment faster and with lower risk.

Additionally, we must also consider that advertising and marketing is unique in that applications must achieve these three goals:

1. Scale to millions of transactions. Our applications must be able to scale up and down automatically, without limitation or decrease in performance, based on demand.
2. Deliver performance as cost effectively as possible. Optimal cost efficiency in the cloud is obtained with cloud-native architecture. This starts with being stateless, using Spot Instances, and extends to the use of serverless compute to further reduce cost.
3. Normalize data so that it can be used for advanced machine learning and analytics. This includes use cases like inventory prediction, media analysis, audience modeling, data clean rooms, or federated measurement.

Do you see different trends in advertising and marketing in the cloud versus on-premises?

Customers usually face the same challenges, but customers using the cloud can resolve challenges quicker with lower costs. We seecustomers in the cloud investing in these workloads:

• Implementing data lakes with AWS Lake Formation to allow central governance and data sharing.
• Investing in clean room services with solutions provided by the Amazon Partner Network.
• Investing in machine learning platforms to deliver optimization, segmentation, attribution, targeting, and personalization with integrated services and tools.
• Using machine learning services such as SageMaker and Amazon SageMaker Studio allows customers to establish a machine learning foundation in their organizations. These services automate their machine learning pipelines, which reduces the time from development to production.
• To reduce costs, cloud customers port their applications to Amazon EC2 instances with AWS Graviton2 processors. These are the fastest processors in the cloud and enable up to a 40% improvement in price performance compared to similar generation x86 instances.

On-premises infrastructure provisioning and application development are generally much slower, which impacts time to market significantly and reduces engineering capacity for product development and delivery.

A common practice for on-premises customers is to expand into new geographic regions using the cloud. This is especially true when industry customers onboard large advertising partners. They risk losing customers to competitors if they are unable to provide the required services in the required time frame.

We also see on-premises customers investing in refactoring their workloads to be stateless and to containerize their workloads. This optimizes their compute resources and prepares them for a future cloud migration.

What’s your outlook for advertising and marketing, and what role will the cloud play in future development efforts?

We have two predictions about the role of the cloud for advertising and marketing:

1. Uncertainty in how advertisers will be able to market to consumers on a personalized, one-on-one basis will continue. The resulting loss of revenue will force publishers to adopt a walled garden approach, which will require users to log in to access their favorite websites and applications.
2. Increased legislation to protect consumers is likely. Using the current availability of third-party identifiers to build consumer profiles while results can still be attributed to individuals is critical. This also applies to building the capability to define, target, and activate audience segments. Transforming organizations that are dependent on third-party cookies to be effective requires immediate and urgent action.
   

We have four recommendations about the role of the cloud for advertising and marketing:

1. Stop moving data. Create a data lake that can be governed and shared in a single location with query federation and analytics initiated at the point of consumption. You’ll be more compliant with data privacy regulations and reduce heavy lifting and transfer fees.
2. Evaluate capabilities like Amazon SageMaker to accelerate innovation and reduce heavy lifting for machine learning use cases like audience modeling and inventory forecasting. With the pace of innovation in the industry today, customers are moving faster using these capabilities. This allows data science teams to spend more time on actual data science.
3. Prioritize cloud-native architectures and instance types if you are considering migrating to the cloud. Examples like Nielsen Marketing Cloud’s serverless data pipeline processing highly variable (spikey) traffic at volumes of 250 billion events per day is a good indicator of what’s possible.
4. Experiment with Graviton2, which enables up to 40% price performance on comparable x86 instance types as a primary cost-optimization and performance driver.
   

Anything else to add?

Visit our industry website at https://aws.amazon.com/advertising-marketing/ to learn more, and see additional customer examples.
Reach out to me — louwgerr@amazon.com — if you want to connect and ask questions.