Don’t only deliver expectations! Exceed them!
All target goals confirmed or exceeded!
- Proper channeling of snippets – CONFIRMED
- Automated document identification accuracy (expected: 85%) – EXCEEDED (PoC: 96.9%)
- Acceptable snippet error rate (sent to QC): acceptable ceiling 5% – EXCEEDED (PoC: only 0.73%)
We recently finished an IntelliVector PoC for a large international bank, the goal of which was to prove that the implementation of IntelliVector will significantly improve the processing (speed/accuracy/cost) of scanned paper invoices (typed text) and paper cheques (handwritten text), compared to the currently used, fully manual processing solution. The current solution requires each document to be manually processed by at least 2 users (1 maker, 1 checker) but in many cases by 4 users (2 makers, 2 checkers). Contrary to that, IntelliVector provided an approach, that allowed to effectively combine automated recognition (OCR and ICR) with a microtask-based data entry, that meant that in the majority of cases the automated recognition matched with the first data entry (91% in case of invoices (using OCR), and 57% in case of cheques (using ICR)), so no further interaction was needed. The rest of the cases, were the two results didn’t match, we almost fully covered by a second data entry, and only 0,7-0,9% of the cases went to Quality Control. And these results were achieved with no OCR/ICR engine fine-tuning, and with half the data entry personnel being fully new to data entry whatsoever.
Moreover, since IntelliVector’s microtask-based data entry guarantees that the data entry users have no access to the original documents (no PII) only to small snippets of information, the bank could even use crowdsourced users during the PoC (for the fraction of the price of full time employees) and results showed that the accuracy and speed of the inexperienced crowdsourced data entry users caught up with the novice and expert users in a matter of days. This opens the door for the bank to crowdsource large parts of the processing, saving massively on the operational costs, without risking the confidentiality of the initial data.
The PoC team consisted of 2 employee stuff (1 expert and 1 novice) and 3 contract stuff (crowdsourced stuff with no typing experience) and took only 3 weeks (including system introduction, training, functional testing, load and stress tests, optimization, feedback sessions) during which IntelliVector was used to process 8000+ structured and semi-structured document (total of 35381 snippets)
Now seeing these numbers, consider, would it be worth to invest a little time and effort and allow us to prove you, that MPS IntelliVector in fact can and will revolutionize your company’s data capture and processing!