How Veryfi Works

How does the training of Veryfi's Machine Learning models work?

At Veryfi, we prioritize continuous improvement and accuracy in our machine-learning models. To achieve this, our models undergo regular training sessions.

During these training sessions, all the changes made by users to processed documents are incorporated. This means that any corrections, updates, or feedback provided by users directly contribute to the refinement and enhancement.

By including user-generated data and feedback in our training process, we ensure that our models adapt to various scenarios, evolving to meet the specific needs and requirements of our users.

Rest assured that your active involvement and contributions play a vital role in the ongoing development and optimization of Veryfi's data extraction accuracy, enabling us to deliver more accurate and reliable data extraction results for you.

Understanding the issue

First, it is important to understand what causing the mistakes in reading your documents. Please use this article as self-service guide for debugging data extraction issues. What factors affect accuracy.

How to help Veryfi improve the accuracy of data extraction?

Correct Issues: Whenever you encounter any inaccuracies or errors in the extracted data, you can correct them using the PUT, POST, and DELETE operations provided by Veryfi's API. By making these corrections, you provide valuable feedback that helps train and refine the Machine Learning models.
Retain Updated Documents: It is important to keep the updated documents within your Veryfi account for at least the next 2-3 weeks after making the corrections. This timeframe allows the models to learn from the changes and incorporate the improvements into future iterations.
Escalate repetitive cases to Veryfi: By collaborating and exchanging data from both ends, we can expedite the process of enhancing accuracy.

Veryfi regularly releases new models approximately every 2 weeks. These updates integrate the feedback and corrections provided by users, leading to enhanced accuracy and performance in data extraction.

👨🏻‍🔧 Volume of edits for repetitive cases/vendors matters.

What users can update?

Update Document level fields (vendor, total, tax, bill_to, etc.)
Update Line item level fields (total, tax, price, category, qty, etc.)
Add a line item to an existing document
Delete a line item from the existing document
and many more*..
*Please check API schema for fields that can be updated here

How to provide model training?

Update extracted values in Veryfi API Portal Document details section
Update extracted values via direct PUT API call

Provide Model Training via Veryfi Portal

Inside Document's details, you will find visualized JSON fields, that can be updated easily by tapping on them and typing a new value. Updates in Portal call Model Training API

👩🏼‍💻 Veryfi supports more than 180+ fields. Not all the fields that can be updated are visualized to Web UI. For fields that are missing from Veryfi Portal UI, please use Interactive API or pass the update via a direct PUT API call.

Provide Model Training via API directly

Find detailed instructions and query samples inside Veryfi API Portal API Docs

Using Veryfi Data Extraction Engine to power your product?

When considering whether to pass all end-user edited values to Veryfi without any sanity checks, it is crucial to exercise caution.

Here's why:

Data Accuracy: Your user edits may not always guarantee accurate information. Depending on the context or user input, there is a possibility of introducing errors or inconsistencies in the edited values. Therefore, blindly passing all user edits without validation can potentially lead to incorrect data being processed.
Data Validation: Implementing sanity checks allows you to validate the edited values for consistency, format, and logical correctness. By performing these checks, you can ensure that the data being passed to Veryfi aligns with the expected standards and requirements.
Data Integrity: Sanity checks help maintain the integrity of the extracted data. By verifying the edited values against predefined rules or criteria, you can prevent unintended or malicious alterations that may compromise the accuracy or reliability of the data.
Model Training Improvement: Sanity checks provide an opportunity to gather valuable insights for model training. By analyzing your user edits and identifying patterns or recurring corrections, you can refine the training process and improve the model's performance in handling similar cases in the future.

By verifying and validating user edits before passing them to Veryfi, you can minimize potential errors, ensuring that users receive accurate and trustworthy results.

Experiencing persistent errors?

If you continue to encounter issues with reading specific parts of your documents, we are here to assist you.

Please reach out to our support team at support@veryfi.com. Provide us with comprehensive details and context regarding the problem.

Collect examples of issue (Document ID and source document files)
Provide expected results
Provide background on Severity and Priority

Please note that depending on the cause of inaccurate extraction Veryfi may not be able to fix it immediately, but will use your report as a reference for future improvements.

Before reporting data extraction cases to support, please exclude indirect causes like document type, image quality, invoice layout, structure, language, and vendor specifics.

Use this article as a self-service debugging guide What Affects ML Accuracy.

Our goal is to ensure that Veryfi consistently delivers accurate results, and we appreciate your proactive engagement in helping us achieve this objective.

If you continue to encounter issues with the new model failing to read specific parts of your documents, we are here to assist you.