Smart Retail Company Realizes a 650% ROI Using Innotescus
Bossa Nova is a Pittsburgh-based company delivering AI superpowers to accelerate the future of shopping. Their retail AI engine provides customers with the fastest, most accurate image-based analytics available. With the help of Bossa Nova, retail organizations can increase sales and reduce costs by gaining critical insights about what shoppers are seeing in stores.
And what does that solution run on? You guessed it – data. Over the years, Bossa Nova has amassed a library of labeled data – 134 million labeled images to be exact – each containing many products now recognizable by their AI engine. This has enabled them to build a solution capable of recognizing over 300,000 unique products in various positions, orientations, and lighting conditions in stores all over the world.
With a mixture of computer vision, Machine Learning, and big data analytics, Bossa Nova automates a myriad of customer operations, keeping each shopper and vendor happy.
That’s a lot of data! And Bossa Nova has no plans to down. New products are introduced to the market every day; stores in London don’t contain the same products as stores in Dallas; and their technology has applications beyond retail. All of those factors point in one direction – Bossa Nova has much more data to collect, label, and manage, and that’s no small task.
Before turning to Innotescus, Bossa Nova used their own modified instance of a popular open source platform to manage and annotate their data. But their initial solution wasn’t up to the task. It was such a burden to maintain, and difficult to teach new users, that only a few team members used it. Having scaled up their team on Innotescus, Bossa Nova has seen a 3x improvement in their productivity, resulting from efficiency gains in their entire operation. They report a 650% ROI from their initial subscription to the Innotescus platform.
Can you imagine how many unique products fill a big box retailer? That’s the number of classes Bossa Nova has to use on any given annotation task. For each product, annotators have to search through tens of thousands of classes to find an exact match, and that’s no small feat when any given brand can have 20 variations of a single product. Before using Innotescus, Bossa Nova built their own tool – separate from their old annotation tool – that allowed annotators to search for the correct product ID with a few key descriptors. With Innotescus, that tool is no longer necessary, because it’s built right into the platform! Our fast, intuitive search handles tens of thousands of classes without breaking a sweat, and doesn’t require annotators to remember the exact order of words or even letters; just type in what you can and we’ll do the rest!
An important part of Bossa Nova’s operation is the ability to associate custom metadata values with each product for analysis and profiling later on in the pipeline. It isn’t just the ability to define custom metadata that’s important to Bossa Nova, though. With hundreds of products in each image, the ability to quickly populate each field is crucial; if entering one piece of metadata is too slow, entering thousands of pieces is a big problem. In response, and with input from Bossa Nova, Innotescus implemented comprehensive hotkeys for entering and navigating between pieces of metadata, so Bossa Nova could optimize metadata entry, and that’s exactly what they’ve done.
When you’re working with a team of annotators and tens of thousands of classes, label errors are bound to pop up. With their old solution, Bossa Nova was able to review individual objects to make sure nothing got missed, and with Innotescus, they can review their work in old and new ways. Aside from the designated review process, which lets supervisors define and configure any number of separate review workflows, Bossa Nova can review annotations with ease from the image viewer, and even our new dive charts, which show images and annotations as thumbnails to quickly glance over and correct when necessary. Quality is an ongoing effort, and we’ve made sure Bossa Nova is able to incorporate it at every step of their operations.
Onboarding & Team Analytics
Their old solution made onboarding nearly impossible, and analytics on team member productivity were out of the question. Because of Innotescus’ cloud-based platform and simple, straight-forward design, Bossa Nova is able to easily invite, quickly onboard, and seamlessly supervise new members to scale up their operations smoothly, and that’s exactly what they’ve done.
teaming up with Innotescus
There were several features that we desired and had enjoyed at various annotation platforms but none that had them all. [Innotescus has] worked hand-in-hand with us to develop those useful tools to have a one-stop shop of annotation data…Their work is high-quality and their care is top-notch.
– Jacob Wells, QA Team Lead
There is a huge potential for machine learning to make our lives easier, and Bossa Nova’s work in smart retail shows us how. Their novel, high performing product recognition technology will solve countless problems for retailers, who will be able to more efficiently and effectively improve customer experiences across the globe. But before that could happen, Bossa Nova needed to upgrade their tooling to prepare for a continuous and large volume of data to collect, label and process to ensure an up-to-date solution. That’s where Innotescus came in. Before switching to Innotescus, Bossa Nova’s team experienced countless inconveniences and workarounds that amounted to a serious hit to their ability to organize and scale their work; after migrating their operations to Innotescus, Bossa Nova saw a 3x improvement in their productivity, and realized a 650% ROI in a matter of weeks. With Innotescus, Bossa Nova’s team has the tools and freedom to execute on their core mission – helping bring powerful machine learning solutions to retailers across the world.
We love this stuff.
The industry insights that fuel our desire to create the best-in-class machine learning annotation platform.
The What and Why of Annotation Augmentation: Three Key Benefits
Everybody knows dataset bias exists. What’s more, most folks are very adept at identifying it. But once your bias is defined, how do you act on it? For Innotescus users, the
Curating Data for Transfer Learning with NVIDIA’s TAO Toolkit and Innotescus
This blog walks through using NVIDIA’s TAO Toolkit and Innotescus’ data curation and analysis platform to improve a popular object detection model’s performance on the ‘person’
March 2022 Release Notes
This month’s updates are all about UI enhancements, buffing up the Image Viewer, and managing annotations more easily. Enhanced User Interface Putting the “You” in UI means giving