SnipSnap & Wizard of Oz Engineering
I still get asked every week how hard was it to build the machine that powers SnipSnap? How long did it take you to master OCR and image recognition and machine learning, which appear to make the instantaneous, magical scanning of coupons possible.
The truth is: a) I didn’t build squat — that was all Kostas and our dev team. b) We actually didn’t master any of these up front. We instead became masters of what is known to some startups as Wizard of Oz engineering. Peel back the curtain and, voila, you’ve got people pulling the levers.
We realized early on that it would take us months (maybe years) to get any kind of funtional auto-detection and parsing working. So we cobbled together a solution that leaned heavily on Amazon’s mechnical turk. We’ve since gotten better and better at automating some pieces of the process, and have actually developed some innovative tech in the OCR space in the process. But the story of how we got our first prototype and V1 of the app to market is a useful tale of trial and error … and more error, that I hope at least one fellow entrepreneur out there will appreciate. I wrote it up for Fast Company. Here’s the opening excerpt.
When my iOS app SnipSnap was accepted to DreamIt Venturestwo years ago, it was little more than a screencast and a high-fidelity prototype built on Keynote templates. Vaporwear. We were planning to build a fairly sophisticated OCR app for coupons and had zero technology. But we learned you can overpromise like this with early features–if you know your way around Amazon’sMechanical Turk service.
Check out the full SnipSnap Coupon App story on Co.Labs.