An Overview of Amp.ai as an Optimization-as-a-Service

Ajay Bhoj

Nov 29, 2018

In an earlier post on Growth Amplified via Autonomous Optimization, we described the kind of problems solved by Amp.ai, and why Amp.ai is an improvement over existing methods like A/B testing. In this post, we'll take a closer look at how Amp.ai offers a versatile, plug-n-play Optimization-as-a-service by synthesizing solutions from five domains.

Introduction

Amp.ai offers a new paradigm for optimizing key business metrics, as discussed in our post on Amp.ai vs A/B Testing: A paradigm shift. It differs from A/B testing in a fundamental way, as it is able to continuously learn from past interactions to optimize metrics in current and future sessions. In doing so, it utilizes all your variants by applying each of them in just the right segments, which are learned automatically. For more information, see our earlier posts on A/B Testing vs Autonomous Optimization and How to Design Autonomous Optimization Projects.

Amp-and-ABtesting-v2.001.png#asset:850


How is Amp.ai used?

When you use Amp.ai, you primarily interact with our client libraries and the console. All else is managed by Amp.ai in a seamless fashion.

Amp.ai as an Optimization-as-a-Service

From an engineering standpoint, building an end-to-end service like Amp.ai involves addressing multiple technological challenges simultaneously. Developing such capabilities in-house can be extremely resource intensive. Irrespective of how metrics are computed (per session or per user), one has to grapple with six key questions while building such an end-to-end learning system. These are:

  1. How to ingest data from the applications to be optimized
  2. How to manage and sanitize big data
  3. How to track metrics and analyze big data
  4. How to model metrics from past data
  5. How to deploy and use models
  6. How to monitor the above for reliability, correctness and availability

All of the above issues are tightly coupled and interact with each other in subtle ways that can lead to failures in optimization. For example, if there is systematic data loss when ingesting data from an application due to buggy client-side code, there can be an inherent bias in the data, which can lead to poor models and failure of metric optimization. With Amp.ai, we make it very simple to hook into a well-tested, continuous optimization service powered by reinforcement learning, where a lot of pitfalls around adapting in-house ML pipelines with reinforcement learning algorithms can be avoided.

By subscribing to Amp.ai, you can access a powerful service for optimizing your business metrics, with simple high level abstractions that require no prior knowledge of AI/ML algorithms or systems. Amp.ai addresses all of the six questions above by packaging solutions from five domains into a single service:

Blog-figures.002.png#asset:824

Each of the solutions above addresses multiple problems that are encountered on the path to successful optimization. They are:

The solutions above are closely synchronized with each other to let you iterate rapidly on how your customers experience your product or service, and optimize your key business metrics using a systematic, data-driven approach.

Summary

Amp.ai provides immense savings and value by giving customers the ability to rapidly prototype and test ideas, while continuously learning from past data, and optimizing metrics tied to all levels of the business funnel. It packages smart client libraries, big data processing, real-time analytics and continuous policy learning and deployment into a plug-n-play service. Thus, Amp.ai dramatically lowers the barrier to entry for software developers looking to endow their applications with intelligent decisions, without having to invest in such capabilities in-house.

For more information, contact us here or email us at support@amp.ai. Schedule a demo here