The BFSI industry is increasingly considering intelligent automation (IA) to address key business challenges. NelsonHall recently conducted a survey in which the participants (50 BFSI executives) observed that IA was able to replace manual processing and human decision-making, allowing Operations to decouple processing volumes from headcount. The respondents expect some key benefits from IA including:
- Improved handling of low value transactions (100% of respondents)
- Reduced cost (98%)
- Improved service fulfillment (96%)
- Reduced error rates (92%)
- Improved customer experience (88%).
However, achieving these benefits have been challenging because:
- Use cases are domain intensive, allowing only up to 50% reuse of similar processes.
- Implementation is labor intensive, often requiring 4 to 6 months and costing up to $1m.
- Unmanaged bots can learn the wrong things and go off the rails, destroying value.
- IA is not one technology but multiple technologies working together, which makes effective system coordination critical to delivering a useable solution.
These challenges require IA vendors to create a roadmap and deliver an offering which targets a narrow range of problems. Typically, there needs to be three types of participants in the development of an IA offering:
- Systems integrator: who will identify client needs, implement the customized solution, and manage the ongoing operation of the IA solution to assure continued effectiveness.
- Academic partner: who will provide advanced cognitive technological tools and insights to develop a differentiated offering.
- Product vendors: to deliver COTS RPA and basic IA tools to the offering.
Let’s look at an example of how a vendor developed an IA offering and how it was deployed.
HCLTech decided to create an IA offering in 2016, with focus on processing of unstructured data. They began their process by partnering with a leading U.S. based university, which had developed analytic solutions for processing radiology images. The image processing capability was useful for processing physical paper documents used in banking contracts, procurement, and handwritten documents.
HCLTech launched its IA product, EXACTO, in 2017. The underlying platform is built on open source machine learning libraries. The product uses servers with GPUs to run Deep Neural network algorithms. EXACTO can be integrated into an existing workflow application with a single API. It has four differentiating capabilities:
- Image processing: fixes distortions, removes noise, and sharpens images.
- Text recognition: extracts text from heterogeneous fonts/handwriting. Detects multi-languages.
- Domain ontology: provides data correction and search-and-sort fields from data streams.
- Deep learning and natural language processing: extracts localized characters and classifies documents.
HCLTech claims EXACTO achieves 85% to 95% accuracy on typewritten documents and 60% to 65% accuracy on handwriting. It processes documents in three stages:
- Digitization of document images.
- Classification of the output into categories.
- Extraction of data from digital output, which is then fed into the target application.
Let’s look at how EXACTO was applied to a very manual and paper-based processing environment.
Trade processing at a global bank
Trade processing is well known for using faxes across large number of parties to conduct business, and frustrates attempts of intelligent process automation. The process has a T+1 reconciliation window, which drives up costs as labor needs to be applied in volume to meet the deadline. Initial errors from manual processing are enhanced by errors introduced by manual processes in the audit trail. Typical errors include:
- Trades erroneously marked as processed in the source system, when they have not been entered into the target system.
- Trades marked as duplicate when two similar but different trades are received from two different brokers.
- Data entry discrepancies at the time of trade booking which go unnoticed until reconciliation.
EXACTO was applied to the processing of faxes and paper documents. It classifies documents using a domain ontology and extracts text into a digital form which is then processed.
Benefits of this IA solution include:
- Reduction in average handle time by 60%.
- Enhanced accuracy due to machine learning.
- Improved audit trails and compliance.
- Identification and correct tagging of duplicates.
- Continuous automated evaluation and improvement of process.
- Capacity to run multiple instances to process large volumes.
- Ability to process 24/7, unlike humans.
- Auto-capture of data without the need to configure each input template.
As described above, IA is enabling institutions to read documents and pictures faster and more accurately than humans. It is then able to process the resulting information with greater accuracy and speed. The traditional cascade of errors is mitigated, improving regulatory compliance and customer satisfaction. The key to success in this endeavor is finding the appropriate technical IP (in this case, in a partnership between domain experts at HCLTech and image reading technology at a leading U.S. based university) to solve complex interpretation challenges. The resulting solution can process low-value transactions at high volume and with high accuracy rate. This type of solution will increase the number of high-volume, low-value transactions which banks will be able to deliver profitably, expanding the range of possible products they can introduce to the marketplace.
EXACTO is trademarked by HCLTech.