Cheques Data extraction: With 97.2% automation, the client was able to cut the time spent on manual labour from n hours to just under 4 minutes.

Who wouldn’t want to reduce the amount of time they spend on tedious, repetitive tasks while also increasing accuracy? With this client, we were successful in achieving that.

We concentrated on the transformation by supporting growth and lowering service costs while optimising the end-to-end document processing service.

The goal here was to improve process accuracy and reduce time spent in Cheques Data extraction. That is to Control the accuracy level of the document process through automated fields and reduce the time spent on documents that are manually indexed, also reducing the number of documents going for manual indexing in order to enhance document processing.

Goal

The corporate objective here was to create a simpler, better capitalized and provide balanced framework for the Bank. Additionally, to build a superior experience for our people and customers in order to compete in the rapidly evolving digital age. We hence wanted to focus our efforts on where we can carve out a winning position and drive a purpose and values-led transformation of the Bank.

Solution offered

Hyperscience automates the classification, extraction and validation of data from customer documents such as Cheques Data extraction. This is done with extremely high accuracy, and produces a structured, machine-readable output. This is easily integrated with your upstream ingestion methods (e.g. email, folder, scan, photo etc) and downstream systems of record, to feed the relevant data into the process.

We promised

Significant cost reduction with 90%+ Automated fields, with 0 compromises with accuracy.

Significant processing time reduction and manual work enabling a 100% increment in the number of documents processed per day.

Better and faster customer service is offered with one of the fastest document turnaround times providing the client with a competitive advantage in terms of generating new business.

Utilising modern technology to directly unlock customer data in machine-readable format to enable deeper insights and further scope for automation.

The case at Hand

The client’s current automation solution does not support Handwriting and Poor Quality documents and also it takes long hours to value and match data with the current OCR solution. Hence it is time-consuming and cumbersome to onboard KYC Documents and it also has significant errors. This results in an increase in time to realise Cheque deposits in branches.

Consequence

Due to the aforementioned reasons, the client was facing the loss of new customers due to a lengthy approval process. Moreover, a loss of new business revenue with customers using trading accounts and hedging FX was seen due to a slow onboarding process.

Agreed success criteria

1. To act on Data Extraction from Account Opening Form, Aadhaar ID Card, & GST Registration Certificate, where Hyperscience achieves a 90% Accuracy level and 80% Automation for validation fields on 3 layouts.
2. To act on Document & Data Validation where Hyperscience validates with at least 90% accuracy of the below key points:

a. Check that 3 required documents have been received
b. Promo Code – If “PRO” then match to list of valid promo codes
c. Nature of Business – Mandatory checkbox
d. Industry Risk Profile
e. Match Industry to list of valid industries
f. If Industry is valid then confirm the appropriate risk checkbox marked H, M, L

3. To Reduce Manual Handling Time, where Hyperscience reduces the average handling time which was earlier taking anywhere from 2 days to 3 weeks depending on location, employees and errors and other variables.
4. To create a Path to Increased throughput, where Hyperscience demonstrates an increase in document throughput by automatically classifying documents to eliminate manual document routing.
5. To work with Low-quality, handwritten extraction where Hyperscience successfully extracts from low-quality, handwritten scanned documents including printed texts.

What Did We Do?

Account Opening Submissions (Structured / Semi-structured / Validation Use Case):

• Our team built new structured layouts within Hyperscience and built the Validation workflow as per Client requirements. We manually identified the fields on all documents in order to train the machine for further operations and kicked off training for a new field locator model within Hyperscience.
• We Completed Model Validation Tasks to improve the first model and finished the second round of training on the field locator model.
• We provided secure access to branch sites to upload documents and had sites submit whole Account Opening submission packs into Hyperscience.

Cheques Data extraction: Hyperscience then extracted 100% of the fields which were about 2400 in total and supervised on low confidence fields. Lastly, we reviewed accuracy reports presenting findings.

What did we achieve?

1. Acted on Data Extraction from Account Opening Form, Aadhaar ID Card, & GST Registration Certificate where Hyperscience achieved a 97.7% Accuracy with 97.1% STP Automation for validation fields on 3 layouts.

Due to this, now X% of fields no longer need human work at all.

2. Document & Data Validation where Hyperscience validated with 99.3% Accuracy Validation across 6 Key metrics.

Validation summary

• 104 Submissions with at least 35 pages per submission.
• 2,388 total validation fields.
• Achieved overall 99.3% average validation accuracy.
• Expect to be near 100% as manual supervision brings overall accuracy to 98%+.

3. Reduced manual time handling to 3 minutes 51 seconds per submission which earlier took hours, days even weeks.
4. After training just 11 samples, Hyperscience achieved 100% machine classification.
5. To work with Low-quality, handwritten extraction, hyperscience Trained and validated machine with 100% Client sample documents.
6. Hyperscience demonstrated how a new template can be created in less than one hour by a business user where no code will be required. This will increase the speed at which the Client can begin classifying and extracting information from new projects without the need to code.