How we Deliver Value?
On the surface, Taggun provides a set of super accurate and easy-to-integrate API to perform receipt OCR scanning, turning any paper receipts to machine-actionable data. We want to make receipt OCR scanning so easy, that even a caveman could do it. 🙂
When you sign up with Taggun, it is easy to assume what you see is all you get: a fixed sets of receipt OCR scanning API. But it is the improvements we make behind the scenes that really counts and delivers value to our customers. We released 24 times last month to improve the robustness and accuracy of the Taggun‘s engine. We do all the hard work, so it works like magic for you.
Taggun ❤️ Regression Testing
Day in, day out, we continuously measure and improve the accuracy of Taggun‘s engine scanning the receipts in your account. We use statistical methods to obtain the baseline accuracy rate for each field: total amount, tax amount, date and merchant name.
We then retrain the Machine Learning model, refine the algorithm and improve any tweaks in Taggun‘s engine. After all the tweaks and refinement, we run a suite of regression tests to ensure the accuracy rate is indeed better than the baseline. We only release if we are happy with the improved results. And then, we start all over again.
A Sneak Peak into the Magic Behind the Scenes
Here is a screenshot of Taggun’s self-benchmark app used to measure the accuracy of Taggun’s engine.
- $ = Total amount
- X = Tax amount
- D = Date
- M = Merchant name
- Each row (on the left) represents a receipt with the timestamp when the receipt is scanned and the format of the file
- Green colour box shows a correct scan of the field
- Red colour box shows an incorrect scan of the field
- Empty box shows that the field is not applicable on the receipt
- At the centre top of the screen, $:86.14% (166) shows that out of 166 scans of amount field, we achieved 86.14% accuracy rate.