Book a scoping call
Case study · Ecommerce and AI · 2019 to 2022

Recostream: the recommendation engine we built in-house and GetResponse acquired

Recostream set out to make advanced recommendation models available to ecommerce sites and content platforms of any size, with a one-line integration and results you could verify in Google Analytics. Stratoflow built it as its own product. GetResponse, a global online marketing company, acquired it in December 2022.

20-30 ms
to return a new recommendation under very high event volumesProduction figure
5-10%
typical sales uplift observed in stores running Recostream recommendationsMeasured in Google Analytics
Dec 2022
acquired by GetResponsePublic announcement
Client
Our own product; acquired by GetResponse in December 2022
Sector
Ecommerce and content platforms
Years
Built and operated 2019 to 2022
Stack
Java services, machine-learning recommendation models, event ingestion at high volume, a JavaScript integration and an advanced API, Google Analytics and Google Optimize integration
Offer today
AI Integration for Java Systems
Challenge

Recommendations that any store could install and any analyst could verify

The mission was to make advanced recommendation models available to platforms of any size. Development had to be strictly data-driven and rigorously tested for both recommendation accuracy and latency. Integration with an ecommerce platform, a content system or a mobile application should ideally need no technical resources on the client side. And the performance of the engine had to be open to validation with third-party tools rather than taken on trust.

Approach

Models improved by measurement, integration by one line of JavaScript

We researched and improved a series of machine-learning recommendation models against real store data. The integration method for non-technical users was a single JavaScript import line that works with any ecommerce platform; more complex integrations used an advanced API.

The dashboard was integrated with Google Analytics and Google Optimize, so store owners could see the objective impact of recommendations on their own metrics and run comparative A/B tests without development resources. Every claim the product made about itself could be checked in a tool the customer already trusted.

Results

Measured uplift, production latency and the acquisition

5-10%
sales uplift most stores observed from Recostream recommendationsGoogle Analytics
20-30 ms
recommendation latency despite very high event volumesProduction
Acquired
by GetResponse, a global online marketing company, in December 2022PR Newswire
What it taught us

Most AI projects fail on data and integration, not on models

Recostream is why our AI Integration offer starts with the evaluation set and the integration path rather than with model choice. The models mattered, but the product succeeded because a store owner could install it in a minute and verify it in Google Analytics. That is the discipline we bring to Spring AI, retrieval and agent work in regulated systems: measurable, integrated, auditable.

Recostream results along the shopper path from product page to purchase: 5% of product page views ended with a recommendation click, 7% of clicked recommendations led to a purchase, and 10% of all products added to the basket came from a recommendation
Related

The offers this product proves

Next step

Book a scoping call

Thirty minutes with an architect, not a salesperson. You leave with a written view of scope, price model and whether we are the right team for it.

Book a scoping callSee the proof first