A shoppable video platform for AiBUY

A shoppable video platform for AiBUY

Itransition developed a complex technical solution for a video ecommerce platform with AI-powered product recognition among 1.5 million images and optimized internal processes based on Agile development practices, thus cutting feature delivery costs by 30% and achieving 2.5x higher productivity.

Case study

A shoppable video platform for AiBUY

Context

The customer is AiBUY, a US leader in shoppable media technology and the owner of an interactive video ecommerce platform that uses overlay technologies and machine learning. The platform allows brands to create shoppable videos and grow their influence by cooperating with partners, affiliates, and resellers on the local and global markets.

Their goal was to bring the next innovative iteration of their product to an MVP stage and transform their patented ideas into scalable and flexible services with the help of the latest technologies. Therefore, they needed a reliable technology partner to define and carry out the delivery strategy. Itransition became that partner as the customer previously collaborated with us on another project and was impressed by our Agile expertise and track record in ecommerce development and machine learning consulting.

Solution

In the course of four years of intensive collaboration, we implemented the following solutions and processes:

  • ML-based product recognition in images and videos
  • The AiBUY Overlay management portal
  • A microservices-based architecture
  • DevOps setup
  • Agile process optimization for multiple distributed teams

Shoppable video and image overlays

AiBUY Overlay is our customer’s proprietary web platform allowing users to create videos for all online channels, showcasing products and services paired with calls to action and interactions for frictionless buying.

The solution utilizes machine learning technologies. Based on the item detected in an image or a video to be overlaid, the system powered by computer vision software scans images found in marketing affiliate networks, automatically analyzes them, and provides the closest matching products from the database. The automatic recognition results could be moderated by administrators, and the underlying neural network could be retrained to fit changing business needs.

Our team complemented the solution with the following features:

  • Customized overlays for the AiBUY platform, YouTube, images, mobile apps, and HTML content
  • Integration with leading ecommerce stores, Shopify and Magento, and the Stripe payment system
  • Integration of third-party video players using Video.js
  • Setup and integration of Salesforce Commerce Cloud

During our collaboration, we made sure to take into account the users’ feedback regarding possible improvements and such metrics as views count per video, conversions per video, smart store impressions, the number of unique users, views by country, etc. As a result, the AiBUY Overlay platform offers the following features:

  • In-media checkout: integration with Shopify, Magento, and Stripe allows paying for items right in the Overlay.
  • Link-out: users can visit a third-party website to get additional information or make a purchase.
  • Lead generation forms: a convenient communication tool that leads potential customers right to the Overlay.
  • Hot spots: admins can highlight specific products or information at certain timestamps in the video.
  • Cue points: admins can set cue points at a specific time for viewers to receive messages and calls to action during a video playback.
  • Analytics: the video and smart store analytics system lets admins get insight into interactions with buyers and sales and trends metrics.
  • Linked smart stores: related stores get triggered at the video’s end or as an ad in the product reel.
A fashion show video with an overlay linking to mentioned products
An overlay with a fashion item ready for in-video purchase

Architecture redesign

When we started out, the AiBUY Overlay was operating as a frontend in JavaScript, the management portal and the admin area as a middle tier in PHP, and the monolith Service Bus as a backend in Java. This caused a few issues:

  • All services were grouped around the monolithic Service Bus and had uncontrolled access to various internal and external data sources and services.
  • To change a certain part of the business logic, AiBUY engineers would have to update several services each in different languages.
  • The services used outdated versions of libraries and languages.
  • Without a cloud environment, the services were deployed on the company’s own hardware in a local data center.
  • There was no centralized logging and collection of performance metrics because of the high expenses for external services and the difficult maintenance for internal ones.
The legacy architecture

In the course of the architecture redesign, we:

  • Introduced microservices
  • Migrated the solution to the cloud
  • Integrated third-party services
  • Isolated the logic from the monolith to microservices and integrated the frontend with the related services to avoid logic duplication and begin legacy code migration
  • Refactored the Java code preserving the main business logic, eliminating three years of the accumulated technical debt in the process
  • Further minimized the technical debt by covering every iteration with unit and integration tests and rewriting parts of the used monolith apps
  • Created a centralized repository of configuration and sensitive data
  • Implemented a new communication approach for connecting different system parts using RabbitMQ
  • Standardized development approaches

Moving to AWS

Itransition’s team provided AWS consulting services and replaced external services, such as CDN and video transcoding, with AWS to simplify the infrastructure and reduce support and usage costs. The services no longer communicated directly with each other but instead used API Gateway/Load Balancer API and other infrastructure perks to enable seamless updates and horizontal scaling if necessary. As a result, the implemented infrastructure was no longer monolith as we transferred the entire backend to microservices.

The redesigned architecture

The microservices communicated via HTTP, where a synchronous response was needed. To cover asynchronous messages, we enabled communication on top of AWS SQS/SNS using AWS Lambda. Only the Management Portal and AiBUY Overlay were left publicly available and distributed via a CDN.

The microservices-based system architecture

By the end of the redesign, the project ecosystem included 12 environments running independently and containing more than 30 individual microservices, serverless lambdas, frontend single-page applications, and PHP services each.

Process

We set up collaborative processes within our team and demoed them to the customer’s decision-makers, set up processes for their distributed teams, and generally organized the mutual collaboration using the best of Agile methodology.

SAFe

Itransition assigned a certified Agile delivery manager to the project, responsible for identifying communication interconnections and bottlenecks. We also introduced the Scaled Agile Framework (SAFe) as it allowed us to link business and development, combining Agile advantages with systems thinking and lean product development. We conducted intensive SAFe injection workshops for all teams.

The project members were grouped on two levels—team and program ones. At the team level, we arranged all the processes and regular releases, while at the program level the business owner, the product manager, and system architects worked closely together.

The team level organization

Organized planning helped the customer get a realistic picture of the entire project, predict issues and bottlenecks, and make data-based decisions for future iterations. We managed to synchronize alignment, collaboration, and delivery between multiple Agile teams and reach the following improvements:

  • 30-70% faster time to market
  • 50% productivity increase
  • 75% reduction in defects
  • 50% improved delivery predictability
  • 20-25% increase in client satisfaction

Kanban cadences

Once the MVP was released, the teams moved from SAFe planning cycles to Kanban cadences and by-feature releases a few times a week to ensure the necessary release speed. This provided a shorter feedback cycle and the ability to deliver increments for several request flows from AiBUY.

When AiBUY requested a new feature, Itransition allocated a dynamic team to develop and move it to production. Meanwhile, the customer’s team could quickly make changes to the development plan and add new tasks without waiting for two sprints until a planned iteration was delivered.

Multiple by-feature releases

Moving from large scheduled releases to fast incremental ones brought the following benefits:

  • 10x faster time to market
  • 2.5x higher productivity
  • 20-25% increase in client satisfaction
  • 70% improved delivery predictability

DevOps

When our engineers joined the customer’s DevOps team, they set out for fully automating delivery and moving the entire process to a new TeamCity application. They were able to cut feature delivery costs by 30% by writing new customized scripts for the features developed by the previous software vendor.

Within DevOps consulting, our team also adopted CI/CD and bug feature release principles. Within the new transparent CI process, any change in the code resulted in automatic assembly and unit testing launch. If the steps were completed without any issues, the updates were deployed to the cloud.

Results

During this three-year-long partnership, Itransition was fully responsible for the shoppable video platform development, coming up with the most effective solutions to AiBUY’s new ideas.

During our collaboration, we have achieved the following results:

  • We deployed machine learning features providing automated product recognition among 1.5 million images.
  • We fully redesigned the platform’s architecture from a monolithic to microservices-based one and migrated it to the cloud.
  • We integrated the solution with third-party payment systems, video platforms, Magento, and Shopify.
  • We sped up release deployment from 40 hours to 30 minutes.
  • We achieved 2.5x higher productivity with better resource utilization and lower downtime.
  • We introduced frequent by-feature releases, with the feature delivery costs cut by 30%.

We have no reservation in recommending [Itransition] to tech startups, venture investors and other companies who need process consultants that offer the right mix of capabilities and experts to deliver high-quality software solutions while shortening time-to-market and ROI cycle.

Ryan Scott

CTO, AiBUY