Evolution AI uses generative AI to extract data from PDFs and scans. Our platform is pre-trained on more than 25 million documents and can learn to read new documents without substantial additional training.
Artificial intelligence means our products are self-learning; no explicit configuration is required to learn how to extract information from a new document. Just point and click—our products learn by themselves!
Our revolutionary ‘zero-shot’ algorithm enables users to extract data without lengthy training and configuration. The algorithm marks a significant breakthrough for IDP technology, which has historically required many document training examples to extract data from a new document type.
Zero-shot algorithms can transfer understanding from previously-seen training examples to completely new tasks. The algorithms can take on new tasks without additional training data, ensuring the model can quickly adapt to processing your company’s documents.
Zero-shot learning is powerful but not infallible. To ensure optimal performance, our AI also learns over time. The AI model improves its understanding whenever a user corrects an error, and will never repeat it.
The software performs self-learning automatically. Unlike competing IDP systems, there is no need for a cumbersome ‘training’ stage.
Testing of our self-learning technology shows that after processing just 25 documents the self- learning model quickly improves accuracy to over 90%, reaching 98% once the system has seen 200 documents (Figure 1). During the first few days of production usage, accuracy will quickly reach 98-99%. Any remaining errors can be corrected by human annotators and accuracy will continue to improve over time.
Train the model yourself - no need to get your IT team involved! Our easy-to-use user interface allows you to quickly train the model to your company’s requirements, with our excellent customer service available as needed.
Evolution AI can apply several data post-processing rules to improve and enrich data quality, such as standardised date formats and calculating inferred fields. If you need something unique, just tell us. We do everything to ensure you have your data, your way.
Data needs to be accessible to the right systems to create value. Many companies are concerned about implementing AI technology - but we offer several accessible options for integration. For example, flexible integration via our REST API makes it easy to upload documents and send the output to downstream systems. Alternatively, you can integrate by file transfer using SFTP.
Unsure about which integration method would work best with your pre-existing company architecture? Get in touch with us at hello@evolution.ai.
Evolution AI understands accurate data is of the utmost importance: our software supports fully configurable QA workflows. That includes multiple teams, spot or comprehensive checking, customising confidence thresholds, double-checking mandatory fields, and even requiring double-user sign-off prior to data extraction.
A series of automated checks and balances also identifies anomalies in the data. For example, ‘time series analysis’ is used to detect greater variation than would normally be expected and flag potential errors.
Example workflow for extracting from financial statements. Evolution AI supports all required financial concepts such as reporting periods; and currency and multipliers.
Custom robustness checks and anomaly detection make sure your data quality is always maintained.
Evolution AI was funded by one of the largest AI R&D grants ever awarded by Innovate UK, the UK funding body for innovation.
A full audit trail is stored so the provenance of each data point can be tracked throughout its life cycle.
ISO/IEC 27001 is the international standard for managing information security. Evolution AI is certified ISO 27001 compliant, ensuring that we securely hold information assets.
