Try : Insurtech, Application Development










Dev Ops(2)

Enterprise Solution(20)




AI in Insurance(26)


Product Innovation(34)


Augmented Reality(7)

Customer Journey(7)


User Experience(21)

Artificial Intelligence(95)



Cognitive Computing(7)

Computer Vision(6)

Data Science(14)


Intelligent Automation(25)

Machine Learning(43)

Natural Language Processing(10)




Telehealth Care(1)

InsurTalks Podcast with KV Dipu: Protecting the Demand-side in the New Normal

7 minutes, 26 seconds read

The outbreak of Covid-19 pandemic has deeply impacted the global economy. Industries such as healthcare, travel and hospitality among others are still reeling from the immediate fallout of the crisis. The Pandemic has exposed the cracks in the Indian healthcare system, and the exposure of India’s masses to a multitude of personal risks who are largely uninsured to stave off financial ruin. At the same time, Insurance has had to adapt their processes to the fast changing climate. Core insurance functions like claims processing and customer support operations have had to accelerate transition to the cloud in order to ensure operational continuity during the crisis and adapt to the new normal. 

In this special podcast, we talk to Mr. KV Dipu about how Insurance is coping with this crisis. Before joining Insurance, he worked at GE Capital for 19 years, where he has built a career in retail finance operations. He is a certified Lean Six Sigma Black Belt and a member of the Harvard Business Review Advisory Council. Today he drives digital transformation as the President of Operations, Communities, and Customer Experience at Bajaj Allianz.

During our conversation Mr. Dipu shared valuable insights on the state of insurance, how insurers need to gear up for the challenges in the New Normal and the initiatives undertaken by Bajaj Allianz to meet their customer’s expectations.

You can watch the full podcast here: 

Interview Excerpts from Insurance in the New Normal

Potential Insurance Frauds amidst COVID-19

Insurance, at least in India, is not strange to the experience of dealing with outbreaks even though at a smaller scale – with virus outbreaks like Ebola & Zika in the past. However there aren’t too many reliable historical models to learn from and you’ve stated in the past that fraud triggers can only work if there are strong flags sitting on top of really good data. In the absence of really good data and unreliable historical models, how does this affect dealing with fraud?

KV Dipu: That’s a good question and I think this is exactly what a lot of players today across industries are grappling with because no PCP or model ever envisaged this. And if you do not have passed precedents then you have to learn as you go. So I think that is clearly what we have seen. In terms of COVID-19, you can see a series of potential fraud possibilities. 

I’m using the word ‘potential fraud possibilities’ because we have to see how they play themselves out. One is you could find a lot of people who possibly could get into scams, that they can maybe influence the entire ecosystem, especially in terms of helping customers who are seeking benefits from the insurance company or various entities. And whenever there are losses you typically will find that there are people out there who are going to try to to make a fast buck. So I think that’s one area we need to watch out for. 

The other is you will actually find that as business models emerge there are some people who’ll be quick to jump into the game. For example, today everybody feels that health insurance is one thing we should focus on and that’s typically when you could have both type A and type B errors. You have middlemen who basically promise health insurance saying ‘I can get you this.. I can have my way through various insurance companies’. You may have people trying to forge various checkups through the entire process. 

So these are some areas which we are very off, right, and the good thing is even if a model from the past is not going to help us with the specific input I think our own experience of various scenarios will come into play.I think as long as we are smart on that front it will help us. Now this is where it’s a classic combination of technology and expertise technology can enable the process but you need years of experience to figure out the fraudulent ones from the good ones. Which is where I think established companies like ours which are technically and technologically savvy, as well as years of deep expertise will be really able to figure out who the fraudsters are.

Change in the Nature of Risks & Its Impact on Underwriting

From an underwriting perspective it’s usually said that poor underwriting leads to poor financial performance, so the ‘not knowing what to expect’ will definitely have an impact on underwriting losses. Going forward, how does this change the nature of risk from perhaps the actuaries point of view? 

KV Dipu: If you look at actuarial science, what they do with every event is they learn, right. The learning adds to their kitty, so to speak. So, today you have various players globally trying to figure out what the models are, what are the potential scenarios and we can also learn from the experiences of different countries. You see while it’s still a global pandemic, the scenarios across various countries are different. Some countries for example have had a very sharp recovery, where they’ve shown a v-shaped recovery. Now there are some countries which are in a u-shape recovery pattern, and  there are some where there is a recovery-outbreak-and then a recovery which would be a W pattern. 

So I think as we see the scenarios play themselves out in various countries, we draw learnings very quickly and then basically recalibrate our models accordingly, that’s point number one. Point number two – I think once the lockdown is lifted and then when you start to see people back on the roads, when you start to see cars back on the roads, and when you start to see hospitals functioning again – that is when I think the rubber will start hitting the road and that is when our extreme vigilance will help. I think as long as we’re prepared with data it will really help us get through this.

[Related: New Product Development in Insurance: The Actuary]

Product Innovations in the New Normal

Today a lot of companies are ‘investing in digital’. They’re making sure they have digital assets, capabilities and tools not just for employees internally in the business but for outward facing agents as well. And that has been  the trend even before the Pandemic had broken out. Most sales teams and channel partners are equipped digitally with mobile apps to generate quotes, issues policies even remotely. 

Given that the physical act of selling itself has been severely affected due to lockdown restrictions and social distancing norms, How can insurance build and protect the demand side?

KV Dipu: Okay, so there is one famous whatsapp forward doing the rounds nowadays. it basically says “Guess who’s responsible for digital transformation in a company? Answer number one: CEO. Answer number two: the relevant CXO. Answer number three: COVID-19.” No prizes for guessing, right? Now what COVID-19 has done is to the point that you made everybody believe that in a push product like insurance in-person meetings, relationship building  is all important and rightly so. And that is the reason this business is intermediated and it’s been that way for a while now. New normal is where people will have to learn how to do contactless selling. That is where COVID-19 helps because if let’s say COVID-19 had been restricted to let’s say one particular city or one particular sector you would not have had a change in universal behavior. 

But the fact of the matter right now is globally right I think there are more people under lockdown than at any previous point of time in history. We have so many people on lockdown and everybody realizes the need for social distancing and the need to go digital. That is when people are also more amenable to being sold to digitally. Which is why now the smarter companies who figure out that in the new normal we have to build relationships while being physically away, and manage to sell from remote or contactless sales as i call it – are the ones who will be able to make a difference going forward. 

The good thing is from a process perspective we have enabled them like you rightly said they have the tools to generate quotes, they have the tools to issue policies, they have the tools to even raise claims. It’s about willingness and that willingness has been accelerated and fast tracked by COVID-19. So what could have potentially taken a long time has now been fast-tracked now in the last 60 days – which is why once the lockdown is lifted and we go back into the world you’ll realize that some parts of this contactless selling or even large parts of it are here to stay.

Click on the below link to watch the full episode of InsurTalks with KV Dipu –

Mantra Labs is an InsurTech100 firm building products and solutions for fast evolving enterprises. To connect with us for interviews, drop us a line at hello@mantralabsglobal.com 

Podcasts in this series:


Knowledge thats worth delivered in your inbox

Tabular Data Extraction from Invoice Documents

5 minutes, 12 seconds read

The task of extracting information from tables is a long-running problem statement in the world of machine learning and image processing. Although the latest accomplishments in the field of deep learning have seen a lot of success, tabular data extraction still remains a challenge due to the vast amount of ways in which tables are represented both visually and structurally. Below are some of the examples: 

Fig. 1

Fig. 2

Fig. 3

Fig. 4

Fig. 5

Invoice Documents

Many companies process their bills in the form of invoices which contain tables that hold information about the items along with their prices and quantities. This information is generally required to be stored in databases while these invoices get processed.

Traditionally, this information is required to be hand filled into a database software however, this approach has some drawbacks:

1. The whole process is time consuming.

2. Certain errors might get induced during the data entry process.

3. Extra cost of manual data entry.

 An invoice automation system can be deployed to address these shortcomings. The idea is to upload the invoice document and the system will read and generate the tabular information in the digital format making the whole process faster and more cost-effective for companies.

Fig. 6

Fig. 6 shows a sample invoice that contains some regular invoice details such as Invoice No, Invoice Date, Company details, and two tables holding transaction information. Now, our goal is to extract the information present in the two tables.

Tabular Information

The problem of extracting tables from invoices can be condensed into 2 main subtasks.

1. Table Detection

2. Tabular Structure Extraction.

 What is Table Detection?

 Table Detection is the process of identifying and locating tables that are present in a document, usually an image. There are multiple ways to detect tables in an image. Some of the approaches make use of image processing toolkits like OpenCV while some of the other approaches use statistical models on features extracted from the documents such as Text Position and Text Characteristics. Recently more deep learning approaches have been used to detect tables using trained neural networks similar to the ones used in Object Detection.

What is Table Structure Extraction?

Table Structure Extraction is the process of extracting the tabular information once the boundaries of the table are detected through Table Detection. The information within the rows and columns is then extracted and transferred to the desired format, usually CSV or Excel file.

Table Detection using Faster RCNN

Faster RCNN is a neural network model that comes from the RCNN family. It is the successor of Fast RCNN created by Ross Girshick in 2015. The name Faster RCNN is to signify an improvement over the previous model both in terms of training speed and detection speed. 

To read more about the model framework, one can access the paper Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks.

 There are many other object detection model architectures that are available for use today. Each model comes with certain advantages and disadvantages in terms of prediction accuracy, model parameter size, inference speed, etc.

For the task of detecting tables in invoice documents, we will select the Faster RCNN model with FPN(Feature Pyramid Network) as a feature extraction network. The model is pre-trained on the ImageNet corpus using ResNET 101 architecture. The ImageNet corpus is a public dataset that consists of more than 20,000 image categories of everyday objects.  We will therefore make use of a Pytorch framework to train and test the model.

The above mentioned model gives us a fast inference time and a high Mean Average Precision. It is preferred for cases where a quick real time detection is desired.

First, the model is to be trained using public datasets for Table Detection such as Marmot and UNLV datasets. Next, we further fine-tune the model with our custom labeled dataset. For the purpose of labeling, we will follow the COCO annotation format.

Once trained, the model displayed an accuracy close to 86% on our custom dataset. There are certain scenarios where the model fails to locate the tables such as cases containing watermarks and/or overlapping texts. Tables without borders are also missed in a few instances. However, the model has shown its ability to learn from examples and detect tables in multiple different invoice documents. 

Fig. 7

After running inference on the sample invoice from Fig 6, we can see two table boundaries being detected by the model in Fig 7. The first table gets detected with 100% accuracy and the second table is detected with 99% accuracy.

Table Structure Extraction

Once the boundaries of the table are detected by the model, an OCR (Optical Character Reader) mechanism is used to extract the text within the boundaries. The text is then processed using the information that is part of a unique table.

We were able to extract the correct structure of the table, including its headers and line items using logics derived from the invoices. The difficulty of this process depends on the type of invoice format at hand.

There are multiple challenges that one may encounter while building an algorithm to extract structure. Some of them are:

  1. The span of some table columns may overlap making it difficult to determine the boundaries between columns.
  2. The fonts and sizes present within tables may vary from one table to another. The algorithm should be able to accomodate for this variation.
  3. The tables might get split into two pages and detecting the continuation of a table might be challenging.

Certain deep learning approaches have also been published recently to determine the structure of a table. However, training them on custom datasets still remains a challenge. 

Fig 8

The final result is then stored in a CSV file and can be edited or stored according to one’s convenience as shown in Fig 8 which displays the first table information.


The deep learning approach to extracting information from structured documents is a step in the right direction. With high accuracy and low running time, the systems can only learn to perform better with more data. The recent and upcoming advancements in computer vision approaches have made processes such as invoice automation significantly accessible and robust.

About the author:

Prateek Sethi is a Data Scientist working at Mantra Labs. His work involves leveraging Artificial Intelligence to create data-driven solutions. Apart from his work he takes a keen interest in football and exploring the outdoors.

Further Reading:


Knowledge thats worth delivered in your inbox

Loading More Posts ...
Go Top

May i help you?

bot shadow

Our Website is
Best Experienced on
Chrome & Safari

safari icon