Regardless of how you make purchases — cash, cards, or with your phone — the trusty paper receipt is still king. Paper receipts (and their close cousin, the emailed receipt) still power business purchases worldwide. And whether you’re a fledgling startup or a multinational company with operations in the United States, the IRS requires that all business purchases over $75 are associated with a receipt. Other countries have similar, or even more onerous, requirements.
At Emburse, our mission is to humanize work. And manually entering information off of receipts is nobody’s idea of fun. Fortunately this is an area where technology can help — but it’s a complex software engineering challenge. At Emburse, we receive receipts from all over the world, in all shapes and sizes. They are in multiple languages and currencies, and the receipts themselves can be wrinkled, blurry, or incomplete. Not to mention that the tip on your dinner check is likely scrawled in nearly illegible handwriting.
This is why we are excited to announce Emburse Receipt Transcription (ERT). The moment you snap a photo of a receipt (or forward an email receipt), ERT kicks into gear. This service uses a variety of strategies to automatically extract the merchant, date, amount, and VAT from your uploaded receipt. ERT powers Chrome River and Abacus today, and is coming soon to our entire family of products, supporting 4.5 million business travelers at over 14,000 organizations worldwide.
The all-star team behind ERT is composed of engineers from across the Emburse family. Each Emburse product previously had invested years of development in independent solutions to this problem. While each unique solution had strengths, we knew we could increase the overall accuracy of receipt transcription by forming this new team, bringing together the best ideas from across the organization.
We’re combining a variety of tricks and techniques from across our family of products. From Nexonia and Tallie we’ve incorporated a home-built machine learning classifier which determines whether or not a receipt contains handwriting (an indicator that the total may include a handwritten tip). From Chrome River we’ve incorporated internationalized auto-categorization logic (the service understands that coffee, kaffee, kaffe and café are all the same thing). From Certify we’ve baked in a strategy that identifies the business phone number in the scanned receipt, and performs a merchant lookup based on that number. We’ll be enhancing our European capabilities by leveraging Captio’s expertise in Spain and Italy. And by bringing these teams together, we’re also able to reduce costs, allowing us to scale and accelerate innovation across Emburse.
We’re excited to continue to improve ERT, delivering the fastest, most accurate, machine learning-driven receipt processing on the market today. And ERT is just the first of several new Emburse shared services that we’re working on, leveraging our combined scale to accelerate innovation in expense management and AP automation. We look forward to sharing more on these initiatives over the course of the year.
And if you’re an engineer interested in ML challenges like this one, we’re hiring!