📊 Change Logs

We offer high-level descriptions of what's new and any bug fixes we developed in our platform over time.

Dec 2023

Increased Language and Region-Specific Support
- Successfully released a major change in our models to extract receipt data from more languages and regions. Our tests show consistently higher accuracy across all data points and different languages and countries.
- We have also streamlined the way we build new additions of data points and support for different countries. So, talk to us about your country or regional requirements.
Improved Long Receipts Processing
- Long receipts, especially supermarket receipts are important for our customers. So, we have improved the way we handle long receipts with better accuracy and stable results.
Bug fix: Fixed an issue when the currency code is not upper case.
Bug fix: Fixed merchantName to return an empty result when the receipt picture is not readable.

Sep 2023

New Feature: Line Item Extraction for All!
- In response to the growing demands for detailed data extraction, we’re delighted to announce the universal access to our refined Line Item extraction feature. With just a tweak in the parameter settings, every user can now retrieve detailed information with enhanced accuracy and speed. Explore the updated feature here.
New Feature: Line Item Extraction for All! Enjoy a Faster Processing Speed
- 30-50& reduced processing time on lengthy receipts or intricate invoices.
New Feature: Line Item Extraction for All! New Data Points Avaliable
- From offering tailored customer insights to catering to regional specificities like the intricate details required for Brazil Receipts, our AI is now equipped to deliver more, enhancing your business's global reach and adaptability.
Bug Fix: We have fixed an intermittent 502 error.
- We have implemented a faster model to prevent this issue with an additional benefit of fallback to the original model when required.

April 2023

New Product: Receipt Validation API
- Revolutionise your marketing campaigns with our receipt validation technology that cuts costs, detects receipt fraud, supports scale, and enables seamless customer experiences.
New Language:
- Our cutting-edge technology is now capable of extracting Japanese receipts.
More breakthroughs in accuracy when it comes to merchant names.

September 2022

Improved accuracy of the Total Amount field as we started seeing more diverse receipt formats coming through.
We've re-adjusted the confidence level threshold for OCR text to boost the precision and robustness of the model.
After some delay due to many iterations of experimenting and testing, the line items feature has recently seen a major breakthrough. We're excited to launch the feature publically in the coming weeks.
As we and our clients take on more global clients, we've made improvements to the currency code and location extraction.

July 2022

We’ve made some amazing improvements with merchant name and location accuracy by taking advantage of a new Named Entity Recognition (NER) model.
We also improve the way we process and extract information from invoices. This is to allow the engine to tackle some of the more complex invoice formats.
Recently, the OCR is noticeably robust in recognising handwriting. This is great news for handwritten receipts, especially for restaurant receipts in the US, where the tips and total amount are commonly written by hand.

May 2022

We’ve released a classification engine to categorise transactions. This is special because the labels are customisable and are based on each client.
We have made progress to improve line items - especially the accuracy. It's particularly to reduce any duplicates and improve the detectability of the line items for longer receipts. E.g., The grocery receipts in the USA.
Location data on the receipt is important for our customers. We have improved the way we detect and verify the location accurately.
Some of our customers use ID to identify and map the location accurately. We have also improved this feature so that it chooses the right location.
Detection of correct date format has also been improved specifically for receipts from the USA, where the date format is primarily month first.
Internally, we built a context classifier to differentiate receipts, invoices, and mobile screenshots. This is to fine-tune the results to achieve a better accuracy rate. We have improved the context classifier accuracy with new training data sets.

March 2022

Product Line Items performance has been improved - the response time has been reduced by 10-15%.
The Australian Business Number verification process is more robust - as Taggun is now hosting the database in-house and no longer relies on the 3rd Party API.
As requested by a customer, Business Type Classification development is underway and is targeted to be released in June. This will help Taggun customers in Expense Management Industry to automatically classify their customer's spend.

February 2022

Last cycle was a very short one and the focus for this one is Line Items. We will have a lot more to report on that soon!
Get in touch if you want us to enable line items for you - and please give us feedback.

December 2021

Made further accuracy and performance improvements for line items extraction for beta customers.
Improved the way we perform OCR on PDF files, especially for the non-standard PDF format files.
Investigated the threat of Log4j security vulnerabilities in our system to ensure it doesn't impact our customers.

October 2021

For our cloud customers, we upgraded to use AWS Global Accelerator. We measured over 30% faster time-to-first-byte in some regions. On top of that, cross-region failover is now instantaneous, (without having to wait for DNS to propagate) to minimise API downtime.
Improved accuracy for the total amount, especially for receipts in the Europe region.
Improved payment type to detect new payments types in the Asia Pacific region, and better detection for Visa, MasterCard and American Express.
Performance just got better. We reduce the average API response time by improving our parallel processing to squeeze more out of the few seconds we have for the receipts.
Available for Beta Testing ✌️ - Line item details (name, quantity, unit price, and total price) of products detected on a receipt. Please respond to this email if you would like to try it out now.

August 2021

Migrated our cloud customers to use Microsoft for the underlying image-to-text OCR, offering better accuracy in all languages, including better support for handwriting.
Our API monitoring just got better. We set up better APM monitoring tools for our API so we can better respond to any delay or spike of traffic for our system.
We improved the merchant name model to avoid picking up some generic stop words.

July 2021

Fixed a security vulnerability. We now only accept HTTPS scheme when domains are passed to the URL parameter. This is to avoid DNS rebinding attacks.
We did the groundwork to seamlessly migrate the underlying image-to-text OCR to use Microsoft OCR service.
We also improved the detectability Geonames location for city names with accent.
We fixed a few regression issues with Invoice number extraction.

June 2021

Moved a cloud-hosted server from the UK to the France region to serve our customers in the EU since Brexit. The endpoint will be removed.
For our Australian customers, we fine-tuned the ABN number extraction algorithm to improve the accuracy rate.
Cities with shorter names (Eg: Lyon) were sometimes missed, we fixed a weakness in our GeoNames city detection to improve the detectability of these cities.
We improved the accuracy for mobile screenshots of receipts from ride-sharing apps in the Asia Pacific.
We've tinkered with the receipt number detection to improve the accuracy for receipts in Hong Kong.

April 2021

For UK receipts, we have implemented HMRC services for merchantVerification, since UK VAT validation service no longer works directly with VIES (EU initiative).
We doubled the size of our cloud-hosted servers to improve faster response times and stability under load.
Continued to make further improvement on merchant name accuracy, which we are happy to have received good feedback from our customers.
Continued to make further improvement on merchant name accuracy, which we are happy to have received good feedback from our customers.
Improved the handling of receipts from Singapore, especially on the currency code and amounts.

March 2021

For receipts from the UK, we have implemented HMRC services for merchantVerification, since UK VAT validation service no longer works directly with VIES (EU initiative).
We doubled the size of our cloud hosted servers to improve faster response times and stability under load.
We continued to make further improvement on merchant name accuracy, which we are happy to have received good feedback from our customers.
We improved the handling of receipts from Singapore, especially on the currency code and amounts.

February 2021

We've introduced a multi-tax feature. We can now detect and extract multiple tax rates and tax amounts (if available) on a receipt. Please reach out if you wish to enable this feature for your account.
We fixed a peculiar and intermittent bug with PDF file processing that causes 504 error response code to our customers.
We resolved reported degradation with the ignoreMerchantName parameter. We found and fixed a weakness in our model when handling merchant names with lower confidence level.
We improved our Swagger API documentation, so that it is easier for our new users to learn and use the endpoints.
We've introduced a multi-tax feature. We can now detect and extract multiple tax rates and tax amounts (if available) on a receipt. Please reach out if you wish to enable this feature for your account.

January 2021

We made major improvements on merchant name extraction to allow better detection of shop names on receipts, especially for French receipts.
Fixed a minor weakness in our detection for date format.
We've also improved the quality of the OCR text content.

More logs

Recent updates before December 2020