A fractal solution to automation

Asheesh Mehra, Group CEO and Co-Founder of AntWorks, talks about their novel approach to crack all four types of data—structured, unstructured, inferred, image—and leverage RPA to power their Integrated Automation Platform.

What RPA is and what it isn’t…

If I define it, RPA is a data transport engine or tool—where it picks up data from Point A and deposits it at Point B. So RPA has been around in the back office for a few decades. It wasn't called RPA, but a data transcoding engine or an Excel Macro. The whole industry came into being around 201-11, when a governance layer and a reporting layer was put on top of it and it was packaged beautifully as RPA. So it’s not what it promises to be: Robotic Process Automation. The key word there being: “process”. RPA on its own is not a tool that can be a process. It is a tool that can automate certain tasks of a business process.

Let’s say data from an invoice needs to be extracted and put into the ERP (Enterprise Resource Planning) platform that is being used by an enterprise, be it Tally or SAP or Oracle… So RPA is the tool that is used after data has been ingested or extracted using a data ingestion platform like OCR (Optical Character Recognition) or cognitive machine reading. When you have data available in an Excel, RPA uses business rules to pick up data from that and put it into the ERP that then leads to an invoice getting paid. The same is for claims. But can RPA take some decisions and apply business rules, provide certainty for the information that is being put into the claim system, and then ask the claim system to pay that particular insurance claim? RPA cannot do that. The technology is not built to do that.

On the difference in approach to RPA…

AntWorks was born in 2015. Both I and my co-founder (Govind Sandhu) come from the BPO (Business Process Outsourcing) industry. We are not technologists. We are business operators. We understand the back office, we understand the business process that is required and the tools needed to automate. So while the Gen 1 RPA companies came with a quick fix tool, called it RPA, their biggest challenge today has been data.

In the back office of an enterprise, you get four broad data types: structured, unstructured, inferred data and image data. That applies to a bank insurance company, retailer, telco, mortgage company or government. While everybody was focused on transporting data, they forgot that OCR can only deal with structured data and that is going to be the biggest challenge of the industry. That’s where our Integrated Automation Platform (IAP) comes in, which at the front end has data ingestion, replacing OCR with our own cognitive machine reading engine, our own RPA and our own Machine Learning and intelligence to do all the work together. We call it straight through processing. A business owner wants a transaction to be processed without human touch—straight through.

Of fractal versus neural…

We built our data ingestion engine using a fundamentally different science called fractal science, whereas everybody else uses neural. OCR is all about character recognition and bouncing off a font library. So it is very restricted in its capabilities, and that is where today's RPA business cases are failing because they use OCR. It cannot deal with unstructured data. We felt from the very beginning that we should be able to deal with all four data types in the back office. Take a bank application form for example, OCR can't read it—because it has a signature, it has handwriting, it has a photograph.

Our platform ANTstein can read and understand pictures, signatures and handwritten script. This is why we are able to deliver an end to end straight through process. Our platform can do multi-format data ingestion. It doesn't matter whether it's a JPEG, a Word or Excel file, a WhatsApp message, an email, a text message or even scraping from the web. We can do all that while OCR can only deal with structured documents like a Word or a PDF file.

Now fractal science is the science of patterns and self-similarity, whereas neural is about absolute character recognition. For example, if I had to train a neural engine to recognise an apple, it would need to be trained in recognising a small apple, an extra small apple, a medium apple, a large apple, an extra-large apple. Whereas with fractal science, you need to train it only on one apple, because fractal uses the principle of pattern recognition and the pattern of an extra small apple is the same as that of an extra-large apple. NASA, Boeing, Rolls Royce all use fractals. My CTO is a fractal scientist for the last 40 years and has spent 10 years at Boeing designing wings using fractal science. The fractal science is very very powerful and it’s like using a tanker to kill an ant.

Now if you had to deploy an automation programme today, you would have to contract with five or six different service providers and product providers (say one for RPA, one to perfect OCR, one for NLP, one for intelligence etc) and then build connectors and APIs to stitch together a solution. But ANTstein gives you all the tools under one roof.

On the automation journey itself…

A client reaches out to us and says: I want to automate my claims processing department or back office. The beginning of the journey is when we deploy what we call our process discovery bot. This is installed on the desktop or the laptop of the users. This particular bot understands the patterns of the user: How much time the user spends on a screen system, how much time is spent on the approval systems, what they are doing in the system… So they are capturing by keystroke and designing a workflow as to what is needed to automate that business process. Once that is done, the platform is deployed and the documents passed through our data ingestion engine or what we call Cognitive Machine Reading (CMR).

Once they deploy this CMR engine or platform, it is then trained to extract the right data fields because the more representative the data is, the higher the accuracy and precision of expansion of that data. Now, representativeness plays a very very big role in enhancing learning and that's where Machine Learning really kicks in.

Once we've trained the engine to recognise the patterns of data that needs to be extracted (all four types), then the data is picked up by our RPA engine that we call Queen Bot and transported into the claims system. That's where certain business rules are applied, certain influences are made, and certain decisions are taken with the cognitive ability and intelligence built into the platform as to whether a claim should be paid or not paid or whether we should write back to the person who has filed the claim, asking for more information. So that, in a nutshell is the journey of a when a client comes to us to delivering a fully automated business process.

How all this has impacted the back office and BPO industry…

One impact is that the transaction is being processed a lot quicker with a far higher level of accuracy and lower human error. Two, those processes that had regulatory barriers from moving data from one country to another has been demolished because data does not need to move anymore. Automation can be deployed within the environment and behind the firewalls of enterprises on shore. The business process can be managed remotely or managed in house by the enterprise's staff without having to move data. Thirdly it’s really increasing the efficiency of the business process, creating customer delight, creating customer success. Lastly, it is moving the human talent and the human brain to really more tasks, more processes and more value added work.

Of bots (digital robots) and the digital economy…

A bot is the data transport engine. Gen 1 RPA service providers provided a dumb bot: All they did was pick up data from Point A to Point B. Actually I don't like to call it a bot anymore. I think it's more of a digital workforce economy that can be very large. On defining that, the jury's still out. Some organisations define a bot that picks up one alphabet and transports it to an ERP. Another definition is a bot that can pick up information from one document and transport it to an ERP. Another definition is that you need one bot to deliver the full business process. But I don't think that should be the driving factor to measure success, but what percentage of my business process can be done straight through by automation, or by a digital worker. What does that digital worker do to increase customer delight, to reduce time to market of my product, to reduce costs of transaction? These are the true measurement metrics that can impact an enterprise, either under a stock exchange, or its valuation, or on the top line.

On digital workers totally replacing human workers…

I don't see that ever happening. The intelligence of the human mind cannot be compared to a digital worker, who will be built by a human being. A digital worker will be configured and given commands by a human being. On the manufacturing side it's happening today. Cars in assembly lines, the packaging of ketchup and jam bottles are totally digitised and managed by robots. It's happening in the automobile industry and retail. Will that happen to the enterprise where human capital is needed? Will that happen on the healthcare side? While there will be large amounts of progress made, I don't see the human element or the human factor going away. I like to put it in quotes: “The digital worker is like the carbon fibre vest that a human being wears to become a superhero”. It is really both of them coming together and using their powers together. Either one without the other is not as strong. Rather than Artificial intelligence or Augmented Intelligence, I think AI should be called Acquired Intelligence.

Let’s say I give you a software robot and if you do nothing to it and if you don't train it, then it’s as dumb as a stone. So you need to train it. AI is not a magic wand. It does not just wake up one morning and say: I am more intelligent than you. I could solve the problem smarter, faster and cheaper than you can. It needs to be trained not different than when you and me started going to school. We learnt ABC first, and then we learnt DEF. In the same way, technology needs to be taught. Yes, there is a self-learning mechanism that takes place. That takes place after it has been trained to a certain point. It cannot start self-learning from Day Zero with a zero knowledge base.

***

Note: What are fractals?

According to the Fractal Foundation, “A fractal is a never-ending pattern. Fractals are infinitely complex patterns that are self-similar across different scales. They are created by repeating a simple process over and over in an ongoing feedback loop. Driven by recursion, fractals are images of dynamic systems… Fractal patterns are extremely familiar, since nature is full of fractals. For instance: trees, rivers, coastlines, mountains, clouds, seashells, hurricanes…”

Fractals may be used to analyse different types of processes and even make predictions. They are used in diverse fields such as signal and image compression, seismology, neuroscience, studying urban growth, electrical engineering etc.