We’re delighted to announce that GLYNT has released its third annual accuracy report! As in previous years, there has been strong technical progress here at GLYNT. For our users, we’re always trying to improve by maximizing our Peak Accuracy and minimizing the effort and time it takes to get there.
Peak Accuracy is defined as the accuracy level GLYNT’s machine learning achieves if appropriately trained. We’re averaging 99%+.
To get to that level, GLYNT needs to train on less than two documents on average. This is a drop from five documents last year, a 65% productivity gain. And a drop from 10 documents in 2020!
Key TakeawaysThere are two big takeaways from the past 12 months: Humans introduce errors. GLYNT has built an advanced ‘Few Shot’ ML system, which uses the user-provided association of desired fields and data targets (eg key value pairs) as training data. One document of associations is great. Additional human interaction with the training set can introduce errors. So we’re working hard to minimize the number of documents used in training GLYNT. A data quality framework that extends beyond documents-to-data extraction accuracy is needed. GLYNT produces high quality sustainability data for companies across the globe. This data is structured and ready to consume. Behind the scenes GLYNT has untangled confusing utility jargon, normalized data formats and more. Customers need verification that these additional steps are done well. So we have developed additional data quality and accuracy metrics.
Of course we’re zealots for accuracy and data quality here at GLYNT. Take a look at our Annual Report. Send us your questions. We’re constantly scrutinizing our accuracy and worrying about all the details. So our customers can confidently share our data with investors, auditors and regulators.
Get the Annual Accuracy Report