The Low-Hanging AI Fruit You Can Pluck With RPA And Other Relevant Technologies
Every time we go out to our mailboxes, we go through a familiar exercise. We classify the mail into a few buckets. Junk goes into the “junk” bucket, bills go into the “sad” bucket, greeting cards go into the “happy” bucket — and so on. We then extract the relevant information from the appropriate mail pieces and act on them. Junk goes to the trash, the “total due” amount and “due date” are extracted from the bills, the words “happy birthday” and “from X relative/friend” are extracted from greeting cards — and so on.
Just like us, most businesses go through this same exercise, just with different scenarios and at much higher volumes. Customers communicate with businesses using physical (mail and phone) methods and digital (email, chat, and social media) channels. Customer service representatives and customer-facing staff members classify all the information they receive from customers based on rules and subjective judgment. The information is then routed to the appropriate group within the organization for further action.
There are many solutions that can aid this process, including the following:
1. Robotic Process Automation
Robotic process automation (RPA) is great at doing rules-based, front-facing tasks like receiving emails, extracting information from those emails and routing the emails internally as determined by certain rules. However, if emails are not structured (with free-floating paragraphs, attached documents, etc.), then additional layers of natural language processing (NLP) and machine learning technology may be required to classify and extract information from the unstructured sources. Although this sounds simple enough, in reality, it requires the machine learning models to be trained on different types of unstructured content. For example, if an incoming email is set up to monitor customer complaints, then the machine learning model has to be trained and fine-tuned with a large amount of data made up of saved complaints from past months and years to be able to classify them. The good news is that as the machine learning technology evolves, many firms already have developed “pre-trained” models that can solve many of these problems right out of the box.
Let’s look back at my first example. My imaginary robot can fetch me the mail from my mailbox, or my RPA robot can fetch my emails from various email accounts and then hand those off to its machine learning buddy robot. The machine learning (ML) robot will sort through the mail (physical or digital), classify it according to the pre-defined buckets and extract the pertinent information that I need. My RPA bot can then continue on and log into my bank’s website to pay the “total amount due” to my utility company and then mark the case as closed or complete. I may decide to keep myself in the loop initially to make sure my robot does not go rogue, but then as I start to trust my robot, my daily mundane process could be fully automated without manual assistance.
2. Traditional Application Development And OCR/IVR
The same problem can be solved by using traditional application development with various combinations of optical character recognition (OCR) and interactive voice response (IVR) tools for scanning documents and routing customer requests. These custom applications can be integrated with all front-facing customer channels and can classify information based on whether it is structured or unstructured. Structured information can be handled by rules-based logic, but unstructured information must be classified and extracted using the machine learning algorithms — preferably pre-trained algorithms specifically trained on specific customer service functions.
3. Pre-Packaged Software With Built-In Machine Learning Capabilities
Many firms either have their own homegrown customer relationship and customer experience systems, they invest in vendor-based systems — or they use a combination of both. Many of these vendors are increasingly building machine learning capabilities into their workflows to handle unstructured customer information. The vendor-based approach could be more attractive for some businesses, especially if a vendor addresses a specific industry segment and has built-in functionalities and workflows for that need.
Whether you decide to use an ad-hoc RPA and machine learning approach on top of your existing application stack or an integrated vendor-based solution approach will depend on your time-to-market constraints and overall maintenance and support costs. Either way, if you understand how to implement RPA and artificial intelligence creatively, the problems these technologies can solve are endless.