Although Agiloft AI is powerful, convenient, and designed to work with many common contract types, some functions are limited. There are five main limitations:

Handwritten Data

The Agiloft AI Core models prefer documents in the format of either .docx or PDF that have been typed out. However, if a written contract has been digitally uploaded, the AI Core will not be able to process this contract. It will not be able to extract or classify any information and the Contract record for the document will essentially appear blank, and the Contract Clause table will not contain any new records.

You need to run on OCR for all scanned documents that contain text. However, when you OCR handwriting, it doesn't produce anything. 

New Labels

The ATHENA models are extraction models that can recognize data in a contract, and then pull and store that data in specific areas in Agiloft. The model can recognize certain data because it has been trained to do so; all of the data that these models can extract by default are represented as records in the Clause Type and Key Terms tables. These individual pieces of data are called labels.

It's very important to be aware that simply adding new records to the Clause Type and Key Terms tables will not cause the models to recognize and extract these labels from the contract; the model needs to be trained to recognize labels before it can be asked to extract them. If you need to add new Clause Types or new Key Terms to their respective tables for extraction purposes, you need to customize the model and then train it to be able to carry out what is specified in the customization.

Language

Contract documents must be in English. Agiloft's extraction models are only trained by default to analyze and extract contracts that have been written in English. 

Table Data

Out-of-the-box extraction models have difficulty analyzing and extracting information from within the tables or charts of a contract document. When running extraction, it’s nearly impossible to tell what data will be extracted from a table, if any data is extracted at all. Due to the unpredictable nature, comprehensive table data extraction isn't possible without first developing a model customized to do so.

Confidence

AI, in general, cannot be expected to extract data with 100% confidence. While the ATHENA models are very convenient, they are not perfect. After a contract is analyzed using AI, it is always recommended to check that there are no clauses and key terms that have gone untagged, and that those clauses and key terms were classified correctly and completely. This is easily done by using the Agiloft Contract Assistant for Word to tag data automatically with AI, and then have the user parse through the document to ensure everything relevant has been tagged.