Smart Extract is Revver's AI-powered feature that automatically pulls and saves metadata you care about from your documents. You will be able to easily automate filing, improve searches, drive reports, and power workflows, all triggered by metadata that Smart Extract finds. All without having to manually work on the file at all.
This document covers some helpful tips, tricks, and use case examples to help your business get the most out of Smart Extract.
Access & Permissions Required
In order to view profile information and information collected by Smart Extract, your user will require the 'Profiles' feature to be enabled.
In order to enable Smart Extract on locations and files, your user will require the 'Profile Management' feature to be enabled.
To use automation features with the metadata that Smart Extract collects, your user will require the 'Workflow Management' feature enabled.
Behind the Scenes - The Smart Extract Process
When a document is uploaded into Revver in a location that has Smart Extract enabled, it will go through a multi-step process of, indexing, optical character recognition (OCR), then smart data extraction.
Indexing occurs first, making sure the file can be searched, it will then check if there is any OCR text already attached to the document, and run our own OCR process on it if none is found. After that, Smart Extract will try and identify specified metadata fields by looking for them in the OCR or extracted text.
This process is lightning fast, only taking a few minutes or less on average.
Getting the most out of OCR Data
Smart Extract works entirely within the OCR data on your documents. If no OCR data or extracted text data is present, our file processing queue will take care of that for you. Here are some tips and tricks to get the most out of the OCR data:
Tip #1 - Getting the Right Data
If Smart Extract isn't pulling the correct or required information, you can make sure that the OCR data contains what you need it to look for:
Click on the 'Smart Extract Feedback' button on a file, it will allow you to view the OCR data that Smart Extract is using.
If text is missing, the existing OCR data could be poor.
Try re-running the OCR process through Revver by right clicking the file and selecting 'Make Searchable (OCR)'.
If you have set up Smart Extract on a location that already has files in it, use the 'Make Searchable (OCR)' option in the right click menu on that location to force Smart Extract to run on all documents in that location. This will use existing extracted text or OCR data on the document. If any individual documents need additional OCR processing, right click them and use the 'Make Searchable (OCR)' option on it directly.
Tip #2 - Tables and Complex Layouts
Smart Extract currently doesn't benefit from what the document visually looks like. This can have an impact on how it reads information with tables. We are working hard to optimize this in the future, but in the short term we recommend experimenting with documents that include simplified table layouts.
Tip #3 - Smart Extract & Metadata
When you select metadata fields for Smart Extract to look for, those field names are passed directly to the AI to try and find matching data. This gives you a key way to manipulate and adjust what you get back from Smart Extract simply by adjusting the names of the fields.
Clear labels for each field that closely resemble matching fields in the document is a key to success. For example, if the documents have an invoice number listed as "Invoice #", you will get better results by making your field name match that, rather than using "Invoice Number".
Get around inconsistencies in how the field is labeled in your document by broadening the field name in Smart Extract. Including the "Or" statement will help Smart Extract broaden its search; for example, "Client or customer name" may produce better results than just "client name".
Experiment with what you are naming your fields to try and match is on the document, be as broad or as specific enough for your use case. For example, if you're using a preset values list, experiment with the name of the values on that list.
Metadata Groups that mirror the fields found in your documents can produce great results. This helps maintain consistency, but also makes it easier to find and select the right fields when configuring Smart Extract to locate metadata on specific locations. One example could be creating a metadata group for W2 documents that includes fields that correspond to the forms.
Be sure you are including any fields you want extracted, this is especially important when considering fields needed for Workflows.
If your use case includes consistent file types being processed, such as customer documents or invoices, include all of the relevant fields for that type of document.
If your use case includes mixed sets of documents, you may find that you want to just include 1-3 fields for each documents, cutting down on the number of fields that could introduce ambiguity.
Take special care with similarly labeled fields if they aren't all necessary. Lack of clear labeling in the document can produce ambiguous results and reduce the overall accuracy of Smart Extract.
Tip #4 - Divide and Conquer
When working with mixed document sets, a great way of ensuring Smart Extract is pulling the correct data from each is by creating an initial set of Smart Extract logic and using that in conjunction with Revver Workflows to separate the documents by type.
This could be based on file names, or specific types of identifying data in the documents. Workflows then move the different document types to different locations that have Smart Extract rules tailored to those document types, allowing for a broader set of fields specific to the document type.
Tip #5 - File Naming
The name of the file is also passed to Smart Extract, giving it extract context and clues on what it should be looking for.
For example, if the file has 'Check' in the name, Smart Extract may have greater success in looking or deducing a check number
A metadata field for 'File Name' can also get Smart Extract to return that as metadata, allowing you to leverage that in Workflows.
Using this method, you can add keywords to your file naming convention that will help Smart Extract and Workflows sort the files into separate locations with additional Smart Extract and Workflows tailored to documents using that key word.
Tip #6 - Field-by-Field Prompting
A recently launched feature in Smart Extract is field-by-field prompting. This helps Smart Extract more easily find fields you are looking for and sets the foundation for future improvements to Smart Extract.
Instead of passing just the field name to Smart Extract, information like the preset value lists is given to the AI to see if it can locate one of the items on that list that is specific to that field name.
This can aid in situations where information isn't clearly labeled in a document, or is ambiguous.
For example, if different document types have different names at the beginning of the document, but isn't labeled as a document name, try creating a "document name" field with preset values of all of the potential document names. Smart Extract will be then be able to produce more accurate results, because it is matching from a list.
Share your feedback
While we continue to develop this feature, we actively look for customer feedback to help us improve it. Let us know what is working well and where we can improve, and any other ideas you would like to see for the future of AI in Revver.
Security overview
With the advent of this exciting new AI technology, new questions and concerns about the privacy of your files and data naturally arise. At Revver, we understand these concerns and prioritize the stewardship of our customers' data above all else. First, you can choose whether you want to enable and use Smart Extract or any future feature powered by AI. We also want to assure you that your data is never used to train any AI models. It remains securely stored within Revver's systems, only accessible to authorized users of your account and is never exposed to outside parties or resources not controlled by Revver. Additionally, while we do use customer feedback to refine our code and improve our services, this feedback is strictly utilized by our engineers and is not employed in training AI models. Our commitment to your privacy and data security is unwavering, ensuring that your information is always handled with the utmost care and confidentiality.
Need Help?
Contact Technical Services
To get in contact with our technical services team for assistance select the chat bubble in the lower right corner. Initially you will connect with our AI bot Fin. If Fin is unable to answer your question please select the talk to a person button below his answer to speak with a member of our technical services department. Fin can also transfer you to a live representative at any time, just ask Fin to "transfer me to a live technician".