Transforming Tax Bill Processing

Client

A state and local tax firm

Tags
Industrial
Challenge
The Client sought to decrease manual processing involved in extracting data from various formats of tax bill forms. Partnering with technology experts at Dialexa, a solution using Generative AI through a proof-of-concept (POC) project was sought.
Roles
  • Architect
  • Senior Engineer
THINKING
  • GenAI
  • POC
WORK

Replaced ineffective manual processes

Their existing system involved using Azure Form Recognizer combined with a unique Python script for each tax bill format. This resulted in an intricate and time-consuming data extraction process that was only utilized for the highest volume tax bill formats. The goal? To replace this with a more efficient, generalizable, and cost-effective system.

A single LLM was deployed utilizing Azure OpenAI.  The team implemented a single prompt workflow across all format types provided by the client and saved the output into the desired JSON format that could be utilized by their downstream processing.

In the short, 2 week engagement, Dialexa demonstrated how Generative AI and LLMs could revolutionize tax bill data processing in a more efficient and cost-effective manner.

OUTCOMES

This POC showed that maintenance needs would be significantly reduced, run costs would be cut in half, and the overhead associated with bringing on a new tax bill format would be removed.

50% Reduction in Run Costs

The proof-of-concept saw a notable reduction in the cost per bill across all tax file formats. While the exact savings will differ depending on the specific tax bill format, a general indication was that the new approach could reduce processing costs by up to 50% per bill.

95% Decrease in Development Costs

The LLM prompt developed in this project was able to cover all tax bill formats.  This will create a significant reduction in the number of python scripts and models needed for data extraction across multiple tax bill formats; from dozens of scripts down to 1 prompt.

~99% Accuracy

The POC demonstrated a higher percentage of successful and accurate data extraction across multiple fields in various tax bill formats compared with prior methods. For instance, most fields showed a scraping accuracy as high as 100% and others with markedly high precision such as 96.33% for TotalTaxAmount and 97% for TaxableValue.

Connect

Let’s make it happen.

The path to your next great product, invention or software application starts here. The first step is starting the conversation. Simply fill out this form, and we’ll reach out immediately.

"*" indicates required fields

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.