Enterprise AI — To Do or Not To Do? (Part II)

“Is this AI/ML potential real?”

“Will it be able to solve my business problems? And if yes, where should I start?”

If you are a business executive (and not a machine learning expert or an artificial intelligence enthusiast); I am sure the current wide stream coverage of advances in AI/ML must have crossed your eye. It is really hard to miss the stories of Tesla Autopilot crashes or the AlphaGo win or how the machines are going to take the human race over.

Part I of this series addressed the AI/ML potential issue. In this part, I am going to present a use case which may potentially be a good starting point for many businesses. The idea is to help an executive gauge an area where AI/ML may be leveraged efficiently and smartly.

Is your problem worthy for an AI solution?

BCG in an article have suggested an approach to identify areas fit for leveraging AI. They have also covered a list of evaluation criteria for the application of AI. The overall idea is to drive the AI initiatives with the strategic focus right from C-level executives to all the way down in the organization.

Let us take an example of document processing — this could be invoice processing, claim settlement or document verification process. These kind of processes are almost embedded in most companies and I will use this to build my AI use case.

These processes are either manual or at most automated using tolerance rules. In some industries e.g. logistics invoice processing is also linked to performance linked SLAs and require additional steps of verification against transactions. These processes are human intensive and often primary candidates for outsourcing to BPOs.

In summary — an ideal use case for AI. Additionally “document processing” qualifies well in terms of the assessment framework suggested above.

A typical invoice matching process involves following steps:

The new approach offers following advantages:

  • Smart matching — POs to match by PO Number, Release, Line, Shipment and PO Receipt
  • Smart search — match an invoice for “plumbing” against “kitchen sink repair” order
  • Handle differentiation due to aberration or spelling mistakes, example — match “Application Maintenance Service” vs “Application Maintain Serv” or “App Maintenance Service”.
  • Handles many PO to many invoices
  • Hybrid Engine (Rules (tolerances) + auto-learning)
  • Complete audit-trail and compliance
  • Manpower savings and error reduction
  • A new smarter approach to handle a “traditional problem”

AI Solution Components

From solution perspective an AI assisted invoice matching will require a document reader, validator, transaction processor and transaction creator.

From a skill matrix perspective; this requires a combination of Computer Vision, Text Analytics and ERP knowledge to create an end to end solution. Computer vision will help read a document and text analytics will help identify and extract the information present in the invoice. ERP knowledge is a must to seamlessly embed this information into the validation and transaction posting processes which are linked to ERP.

Following additional information/data will be required:

  • Identification of physical/electronic invoices (training data for document reader — to read and extract data from the document)
  • Identification and location of corresponding transactional data (PO, GR, Invoices, Payments and so on) from ERP and/or DW
  • Requirements for invoice matching.
  • Interface requirements for transaction posting into ERP

With this you have more than what is needed to launch your first AI initiative.

Summary

I hope as a business executive this helps put into perspective how AI/ML may be used in your organization to solve real world problems. I hope this also shatters some of the most common myths surrounding the topic. If your organization has been collecting data; have a routine, repetitive or mundane task or probably have something which was earlier automated using rules — you have a potential opportunity to run an AI/ML pilot.

Andrew Ng (chief Scientist of Baidu; Chairman and Co-founder of Coursera; Adjunct Professor of Stanford ) recently said “AI Is the new electricity“. I am sure you don’t want your business to operate off the grid forever.

Still reading :) Thanks for your time. Please share your experience/comments and feedback and if you are still thinking how to move forward or have any more questions/doubts give us a shout.


Originally published at crispanalytics.com on April 5, 2017.