Q: I have 2 companies, can I buy 10 codes and split myself 3 for company A and another 7 for company B?
Integrate with Xero?
How accurate is the OCR?
Does it also detect the registered company name by searching the local registra?
Antti_SparkReceipt
May 14, 2024A: Hi Noah,
Yes, you can split codes between your companies. However, at this initial stage we do not have a workflow for registering two company accounts with the same email address, but it is supported via our admin panel and coming to the apps (really) soon. In case that is not yet implemented upon redeeming the codes, contact us at sparkreceipt@valorbyte.com and we'll setup the login credentials to have access to both accounts.
There will be integrations with at least Xero and Quickbooks in a future version (and I promise they will be included in the AppSumo deal!), but the priority of the integration implementations are being based on feedback. Using Plaid to integrate with banks is also a consideration, but it is expensive and needs some though about pricing.
The OCR is pretty accurate as long as the source material is sufficient. Paper receipts are notoriously difficult for OCR, since they can be crumbled, twisted, faded and depending on the camera's quality, blurry. However, as we heavily utilize ChatGPT as the language model, it can usually do pretty good job at understanding the receipt text despite OCR errors.
And we do not, at least on this stage, support searching registers for official business names or identifiers. The merchant name is detected by the AI like usually a human would, by scanning the contents and understanding what is the merchant from the receipt's context. This is actually also the reason why the merchant detection works well on international vendors and small chains since it is not based on a registry. This approach has the weakness that the AI might sometimes detect e.g. "McDonald's" and sometimes "McDonald's Restaurants". In the future, we could implement a database of companies, their official names etc. And then try to match the AI's guess to the verified database data.
This would be immensely useful..... I currently use Scannable to email into Evernote then split to appropriate company. Your software will definitely save some time. In terms of OCR, the one with Scannable is probably the best on the market, no issues with clarity for receipts and paper documents. It sucks though on ID documents, hope your software can rectify that issue.
>>>> This would be immensely useful..... I currently use Scannable to email into Evernote then split to appropriate company. Your software will definitely save some time. In terms of OCR, the one with Scannable is probably the best on the market, no issues with clarity for receipts and paper documents. It sucks though on ID documents, hope your software can rectify that issue.
--
Yes, as an industry leader, Evernote Scannable is very good at OCR. And they probably use an AI model too to clean the output. We rely on Google MLKit (on-device OCR) and Tesseract 5 (web OCR) for the optical character recognition, then use ChatGPT to clean the receipt. And to be honest, there probably is room for improvement there, mostly regarding image preprocessing steps. We have an implementation for Amazon Textract too, but that is a bit slow (30-45s) for receipt scanning purposes.
And please let us know how the OCR & document management works in your case, and what if anything is lacking. I am very curious about hearing about all needs & use cases.