Hi Ushur Community,
Do you have a question related to AI and insurance that has stumped you or maybe you just want some expert feedback on? Now is your chance to get all of your questions answered during Ushur’s inaugural AMA event!
We’ll have VP of Ushur AI Lab Viju Shamanna, PM Vishnu Rajkumar and Sr. PMM Will Roberts on hand to answer all of your questions.
How to participate:
- Drop a question or comment in this topic any time from October 25-29.
- Ensure you’re ‘Watching’ this topic, so you’ll be notified of updates. Just click the watching button below this post.
Here’s a bit more about our featured experts:
Viju is VP and head of Ushur AI lab, the research arm of Ushur that drives all the AI/ML innovations of the Ushur platform. He brings almost twenty-five years of experience in software and systems engineering, much of it spent in leadership positions spanning domains such as data center infrastructure, cloud-native applications and applied machine learning. Prior to Ushur, Shamanna bootstrapped the India engineering team for Vexata, a disruptive enterprise storage startup acquired by StorCentric. At Sandisk India, he led the emerging systems group that developed hyper-scale, disaggregated flash storage systems optimized for big data.
Vishnu is a Senior Product Manager at Ushur driving product strategy, roadmap and execution.
Vishnu has over 12 years of experience is Marketing and Product Management. Loves to make an impact in peoples lives by solving problems with technology. Prior to Ushur he has gained a ton of experience working at different verticals such as telecom, e-commerce, shipping and retail serving both B2B and B2C customers.
Will is the Senior product marketing manager at Ushur. Before Ushur he was a product marketing manager and data science evangelist at IBM. He’s also consulted with Red Hat with a focus on financial service clients.
3 Likes
Do you see claims adjudication being the biggest AI affect in the insurance industry vs. other area like customer experience? And how do you think this will change in the next 3-4 years?
1 Like
Yes of course! A typical claims adjudication process goes through the following stages:
-
An Initial quick triage (mismatched patient names, service/diagnostic codes, incorrect plan numbers, etc)
-
A more detailed review (for duplicate claims, no authorizations, bill review etc)
-
A detailed review by examiner, cross checking claim against multitude of docs, requesting additional information etc
-
Payment Determination
-
Finally, the payment
I think that steps #1 and #2 above will be automated using AI. There will be interesting AI/ML-powered “assist” mechanisms that can augment human decision-making in Step #3. However, for any AI/ML application to help in Step #3, strong explainability and trust will be the primary concerns.
Finally, I am not sure if claims adjudication will be the biggest, but I can see that AI is going to make a significant impact in that area. In fact, any workflow which involves unstructured information processing is going to benefit greatly from AI.
2 Likes
AI projects typically begin with a “start small”, experimental project and successful projects need to scale fast to deliver on business expectations. This approach results in organically grown algorithms, business rules and AI models. I often wonder about sustainability … if we’re not careful we’ll build a house of cards and my fear is that if we don’t protect ourselves that some day business value will deteriorate. This is a philosophical question but might be more related to technical debt than AI. However, do you have any views on what the future holds in the “technical debt” , “refactoring”, “re-work” space for AI solutions … for example, do you see a future where the AI solution can just “fix” itself so that the notion of technical debt goes away.
1 Like
How effective is sentiment analysis? It’s something I often wonder about … if we delegate front line customer engagement to robots and we use sentiment analysis to tackle the irate customer first will all of our customers start SHOUTING or USING CAPS LOCK in all their engagements because they know that it’ll get the attention of a robot?
Conversational AI in the future? I’ve been using Amazon Alexa for a few years. In my opinion it hasn’t really gotten more intelligent in the past four years. If anything it’s getting less effective because it’s feature set has grown. One aspect where it’s useless is in conversations – I can’t have a back-and-forth conversation with Alexa beyond a dumb shopping list conversation. I’m not talking about the non-Amazon features here either – I feel that the core of Alexa has a lot of space to improve … so my question – if Amazon can’t get it right is there any hope for us to build great solutions for our customers? I’m sure the answer here lies in the fact that our own solutions will be very specific to our business (much narrower focus) so we have a fighting chance of creating something good.
2 Likes
As our customer needs evolve our models and solutions would evolve and even become smarter. Today, our customer expectations have changed drastically - there is a market stat which says “89% of customers would prefer to communicate with their business over text”. More and more customers are preferring digital channels as opposed to traditional portals and calling into call centers. As you had mentioned a narrow industry focus is required towards building a strong understanding of our customer needs which will aid in building better models. Leveraging analytics to better understand customer behavior and friction points will help deliver tailored solutions to meet or exceed customer expectations & solve the majority of their pain points.
3 Likes
You are right on the money here Barry! Classical Software engineering stipulates abstractions, isolated, loosely coupled systems with a modular design to support future enhancements, reliability, etc. But, AI/ML tends to bring together disparate sources of data, thus creating tight coupling and making independent, isolated improvements very difficult. In fact, this is called as CACE principle (Changing anything changes everything!!). Some of the typical sources of tech debt in ML world are feedback loops (model output ends up getting fed back into model!), feature sprawl (too many features get added over time for little increase in value), override/correction cascades (some overrides might have been added to improvise on model output but over time, they get too hairy!!), visibility debt in the form of output of a model being fed into some other system without any kind of governance etc. I am not sure if AI can just “fix” itself as some of it is more cross-systemic in nature. But I can see a lot of innovations coming our way in terms of model observability, better collaborations tools to avoid pipeline jungles, automated feature management tools etc
4 Likes
thanks Viju, that’s a great insight, lots of food for thought there and it’s good for us to be aware of these pitfalls. I guess if a robot could solve all of these problems then I wouldn’t have a job 
3 Likes
Thank you all for participating in our first Ask Me Anything! Stay tuned for more AMAs coming in November.