Introduction
Scoutbee: Better Data. Better Decisions. Better World.
Scoutbee drives better business outcomes by giving companies the actionable insights they need to perfect the supply base and advance strategic initiatives, such as risk management, ESG and innovation. The Scoutbee Intelligence Platform (SIP) uses graph technology and predictive and prescriptive analytics to deliver holistic supplier visibility that helps procurement make confident supplier decisions, drive cross-functional efficiency, and optimize their existing technology investments. Scoutbee’s AI-powered data foundation connects teams to any data point – internal, external, third-party, and more – and any data combination necessary to orchestrate a resilient, competitive, and sustainable supply base.
In order to deliver these solutions, Scoutbee uses large-scale information extraction using deep learning models on unstructured data found on the web. They then use machine learning to rank and score each item for relevance to make it easy and accurate to search for supplier information across their knowledge graph.
Scoutbee is a global company with employees from 20+ countries. Learn more at https://scoutbee.com, and follow Scoutbee on LinkedIn, Twitter and YouTube.
The Challenge
Using Proprietary Knowledge to Drive ML Product Development
Scoutbee has a deep understanding of the personas in the supply chain industry and how they interact with data associated with the supply chain.
That knowledge presented Scoutbee with a major opportunity to innovate and deliver significant value to customers at scale. The big idea? Using domain-specific proprietary machine learning models to automate supply chain information collection, cleaning and enhancement (a task that would normally take a customer hundreds of hours to do by themselves), and then make that information easily available and searchable for Scoutbee customers.
However, in order to train their models, Scoutbee needed to create very specific labeled datasets using their own proprietary data to ensure that they would have a strategic advantage over other publicly available search engines. And they needed to do so in a cost-effective manner.
The Solution
Scoutbee Finds Label Studio
To develop targeted, highly-accurate datasets, Nischal Harohalli Padmanabha, Scoutbee’s Vice President of Data Engineering and Data Science, needed a flexible labeling platform that could do rich HTML annotation, along with the ability to support active learning and large scale inferences. They were particularly interested in finding a platform to support human-in-the-loop reviews to ensure model quality and SLAs.
Discussing their criteria for a labeling platform, Nischal states “As part of choosing a technology partner at Scoutbee, we always evaluate several options, so that we can make a good decision. Our first thought process is to always look into the open source world, to see if there are tools we could host and contribute to that we can work with.” This led them to take a look at Label Studio, the most popular open source labeling platform. After evaluating both Label Studio Community and Enterprise editions, they determined that Label Studio Enterprise was the right solution for them. Ultimately, “based on our requirements (and of course, our budget for the tool), we decided to move forward with Label Studio,” says Nischal.
Major Revenue and Efficiency Improvements With Label Studio
Scoutbee has seen significant success with their ML-driven products while using Label Studio Enterprise to both train large-scale models and provide adjustments to their models currently in production. This includes significant results like:
2-3
x
increase in revenue generated through ML-based products
20
x
reduction in time taken to label data and train and maintain models while keeping quality
at SLA level
>90
%
model accuracy across millions
of documents
In speaking about his experience with Label Studio, Nischal says "All of the investments we've made, including tooling, licenses, annotation, crawling, extraction, and training of machine learning models, and running the team, come at a cost. But the revenue and value we bring to our customers via Label Studio have more than compensated for it.”
Conclusion
What’s Next for Scoutbee and Label Studio?
Using Label Studio in conjunction with their proprietary data to build their knowledge graphs along with all the ontology work that they have put in place, Scoutbee is in a great position to reap the benefits of working with large language models (LLMs). They plan to give their customers the ability to interact with supply chain data by asking questions using natural language. This will open up new opportunities to enhance the product user experience, as well as new use cases and capabilities for their customers.
Nischal is excited about the future. “We also have plans to support not only text-based but also image-based information extraction in the future to further enhance our knowledge graph. AI at Scoutbee is quite exciting at the moment.”
All of the investments we've made, including tooling, licenses, annotation, crawling, extraction, and training of machine learning models, and running the team, come at a cost. But the revenue and value we bring to our customers via Label Studio have more than compensated for it.
Nischal Harohalli Padmanabha
Vice President of Data Engineering and Data Science
