will update when they update
Everyone has chatGPT 4o now, QuickBooks online just advertised yesterday to me that they make snapshots of receipts super easy. To be valuable, SparkReceipt needs to update their expense categories to match the IRS categories, so we can use this for something practical like taxes. Not asking you to compete directly with the accounting softwares, just simply help us upload our expenses in proper format for the I.R.S. tax season like how QBO automatically provides these options. Maybe have meta categories so each category can be put into one of the IRS expense type buckets.
Otherwise it's just creating extra work by using this tool. Another thing we need to see prioritized on roadmap is multiple-expense screenshots being converted to multiple expense, just prompt the ai "This image may contain a list of multiple expenses, it is critical that you accurately capture each one individually". Such an easy update, and you can tell user "this feature may not work perfectly each time so double check it", that is fine but at least TRY to add value please. Everyone was nice since we got 20% off but if this is really for freelancers and small businesses we need ability to autocategorize for our simple tax-ready records, and to upload statements of multiple expenses and at least have the a.i. attempt to scan them all in as separate expenses. I'll update this review to 5 stars and make it sound super nice once these 2 simple features (which you can build in a few hours I'm sure, if not 30 minutes) are done. Please confirm if you understand/agree with my feedback; Thanks!
Antti_SparkReceipt
Oct 24, 2024Hi,
Thanks for the comprehensive feedback.
Country-specific default accounting categories (and tax codes - like QB) are planned, and US / IRS is for sure among the first ones to be supported, but I'll be wary on challenging e.g. TurboTax on foreign soil. Our US customer base makes about 28% of the total, followed by e.g. Canada, UK, Australia, Germany, Spain, France and India which I would also like to support on this.
I am sure Intuit will manage to make a globally available receipt scanner eventually (seems to take long though). But it will not make dedicated preaccounting software such as Dext, Autoentry, Hubdoc - or SparkReceipt go out of business by any means. But it will raise the bar, which is always a good thing for everyone.
Overall globally the small business audience is around 400M in size, with QBO taking up around 7-10M of that. And there are a lot of niches to fit these types of software into, and shaping the direction is an ongoing process for us. Right now we are attempting to make a tool that combines the best aspects of currently available software, with affordable pricing, out-of-the-box multi-workspace setups, and with our own spin on it - and of course our bigger vision does not end there. But right now we are here on AppSumo, on our early stages, selling cheap deals to get the ball rolling to bootstrap our way to the bigger arenas. And the current product, especially combined as a life-time deal seems already to be enough for many user's needs. The monthly pricing is also very lucrative and is selling very well so I think there is demand for this kind of thing, but for what exactly - we will see. I would say patience is the key here.
About your other issue - it could surely be a separate function that attempts to split the image into segments based on where the text boxes are, then do some analysis and feed the splits into a GPT. Not a few hours of work, but maybe 3 days to a week. GPT4 Vision could possibly help here, but it is super expensive, performs poorly on spatial reasoning, and it is not very good at OCR compared to e.g. Google's MLKit Vision.
As a first remedy, in the next mobile app version I will change the scanner to a ML based tool which should allow for better edge detection, better batch scanning and more accurate OCR. Unless of course you already have sheets with 4 expenses on them and would need an auto-split function when uploading? The feature does not need AI, since we can do computer vision and OCR-based segmenting faster and more reliably if needed. But overall it is not a trivial feature.
But thanks, this made me think about the bigger picture.
Best Regards,
Antti