Rise of AI Conference 2019 · Berlin

Cognitive Claims Handling with AI

At Rise of AI 2019 in Berlin, Sofie Quidenus-Wahlforss, CEO and founder of omni:us, describes her company's journey applying narrow AI to insurance claims handling and argues that general intelligence remains far off, so startups must focus on a specific, measurable use case. She traces her path from a document-scanning robot she built in 2004, which digitized over 980 million documents, through the realization that most digital documents like PDFs are not machine-readable, to founding omni:us and eventually narrowing its ambition from a broad, horizontal document-understanding platform serving 5-10 industries to a vertical focus on a single process: insurance claim handling. She explains that while standard approaches extract 7 to 14 data fields from a claim, omni:us technology extracts around 200, enabling a shift from process-driven to data-driven insurance. She describes the technical architecture of a reusable, scalable training engine combined with an industry-specific natural language processing layer, and closes by describing the cultural difficulty, for both her company and its insurance customers, of pushing large-scale transformation.

Key takeaways

  • Quidenus-Wahlforss's early document-scanning robot digitized over 980 million documents, becoming a key data asset for later AI work.
  • omni:us narrowed its ambition from a horizontal, multi-industry document platform to a vertical focus on a single process: insurance claim handling.
  • Standard claims processing extracts 7 to 14 data fields; omni:us's combined vision and NLP approach extracts around 200 data fields per claim.
  • The company's reusable technical engine is about 95% transferable across domains, but achieving high confidence scores requires a domain-specific language layer built for each industry.
  • omni:us can offer pre-trained models needing only a few hundred documents to fine-tune per customer, rather than requiring millions of new documents.

Detailed notes

From a document-scanning robot to narrow AI

Quidenus-Wahlforss recounts starting her first company 14 years before this talk, building a robot that automatically found the right pressure to turn a book page and used a camera and inter-robot communication to digitize documents; over 160 units were sold worldwide, and the robot eventually digitized over 980 million documents. This work taught her two lessons that shaped her career: access to unique training data is the real bottleneck for AI startups, and most digital documents, including PDFs, are not machine-readable despite appearing intelligent to humans. Reflecting on general AI, she notes that a system beating the world's best Go player still cannot play chess, and argues genuine general intelligence, applying one learning to an unrelated situation the way a toddler can, remains distant, while narrow AI trained specifically for one task is what is achievable today.

From horizontal ambition to vertical focus on claims

Quidenus-Wahlforss describes omni:us's early temptation to stay horizontal, claiming to understand documents across 5 to 10 industries and a 200-billion market, arguing the same underlying document types like invoices recur everywhere. She says this was a painful choice for a founder unwilling to let go of opportunity, but the company ultimately decided to vertically focus on the insurance industry and, within it, on a single process: claim handling, aiming to become the leading technology for that specific transformation. She raised 4.2 million on an early pitch built mainly on credibility and access to data, but notes investors now expect concrete numbers and process specifics rather than general AI claims.

Why insurance and why claims

She explains the company chose insurance because it affects everyone, yet claims processes are slow, opaque and frustrating for customers who expect the same digital speed they get from their smartphones. She cites insurtech peers and customers such as Wefox, Lemonade and Coya, competing alongside larger incumbents like Allianz and AXA, and describes the industry as moving from skepticism toward the vision omni:us proposed two years earlier to active demand. Standard document processing extracts 7 to 14 data fields from a claim; omni:us's combined computer vision and natural language processing approach extracts around 200 data fields, enabling a shift from a process-driven to a data-driven claims operation, with benefits in speed, transparency and product design, not just automation.

Technical platform: a scalable engine plus a domain layer

Quidenus-Wahlforss describes two components any AI startup product needs: a scalable, reusable technical engine for fast experimentation, training, retraining and ground-truth annotation, and an end-to-end product layer built on deep domain-language understanding, since natural language processing depends on the specific vocabulary of an industry. omni:us started with a dataset of over 30 million documents, letting it offer customers pre-trained models that need only a few hundred documents to fine-tune, rather than requiring new customers to supply millions of documents themselves. She mentions collaborating with the Computer Vision Center in Barcelona as an example of the research partnerships a startup needs since it cannot build everything like a research center.

Q&A highlights

Asked how much of the omni:us architecture would be reusable in a different domain such as electronic healthcare records, Quidenus-Wahlforss says the technical platform, the engine enabling fast experimentation and ground-truth annotation, is about 95% reusable, but achieving a confidence score above 95% requires training a domain-specific language layer for each new industry, which is why the company focused so narrowly rather than trying to serve multiple domains at once.

Quotes

Neural nets are out there since the 60s, and the bottleneck, especially for startups, is the access to unique training data.

Sofie Quidenus-Wahlforss

We give back time because we automate the process of claim handling.

Sofie Quidenus-Wahlforss

Welcome to the uncomfort zone at omni:us.

Sofie Quidenus-Wahlforss

Questions answered in this talk

Why did omni:us focus on insurance claims instead of a broader document-understanding market?

According to Sofie Quidenus-Wahlforss at Rise of AI 2019, omni:us initially considered staying horizontal across many industries but found that achieving high confidence scores requires a domain-specific language layer, so it chose to vertically focus on one process, insurance claim handling, within one industry.

How much more data can AI extract from an insurance claim compared to standard processing?

Sofie Quidenus-Wahlforss states at Rise of AI 2019 that while typical approaches extract 7 to 14 data fields from a claim, omni:us's combined computer vision and natural language processing technology extracts around 200 data fields.

Is an AI document-processing platform reusable across industries like healthcare?

Sofie Quidenus-Wahlforss says at Rise of AI 2019 that about 95% of the underlying technical engine is reusable across domains, but reaching a high confidence score in a new domain such as electronic healthcare records requires training a new, domain-specific language layer.

Mentioned

narrow AI, general AI, document understanding, computer vision, natural language processing, confidence score, process-driven vs. data-driven

Keep exploring

More from Rise of AI Conference 2019

More on Enterprise AI & Adoption

Transcript

Auto-generated transcript, not edited. · 3,668 words

Show the full transcript

I'm Sophie and the founder and CEO of Omni us and we are applying artificial intelligence to the insurance base and I will share with you our learnings and our journey and how I think about general AI the big hope and narrow AI and where we are in this journey and yeah we share with you some insights of what we did in the last couple of years so we are in a I actually already since 2004 so yeah I am always fascinated about the way people think about artificial intelligence where we are sitting in our little office trying to crack the nut of deep learning algorithms people think about AI as even this robot killing all of us or robots which will replace us whatever we do playing thinking working talking and then and if you are really into that space and you try to just understand us a document once in class of document you understand how wonderful our human brain is you start appreciating how intelligent we are it's just crazy a two-year-old child

every one-year-old baby is actually more intelligent than the algorithms we are building currently at our company at Omni and it's not that I'm not proud what we do because for the current state where we are I think we're doing an amazing job but still so just from my point of view the human brain is just it's undescribable what we can learn and how much of that we can apply to a totally different situation so long time ago 14 years ago or even longer I started my first company in deep tech so it was a robot I invented a robot which was able to automatically find the perfect pressure to turn a page of a book and then it activated the camera and they communicated with each other and all the robots we sold over the world over 160 they communicated with each other they exchanged notes so it was artificial intelligence and robotics in combination and during this whole journey I have learned something very important for the rest of my career in AI and

that is how important data are of course because neural nets are out there since the 60s and the bottleneck especially for startups is the access to unique training data so this was one important learning ahead back then the other important learning was we are creating more and more digital documents which are non machine readable we think that a PDF is intelligent it's not it's covering letters for a machine it has no content so that's the reason why with all the new devices like an iPhone and we are sending documents back and forth and creating millions of new digital documents we're creating more and more manual work and I found that super fascinating and yeah and that also led to the understanding that because of Google's initiatives around open sourcing all their technologies tensorflow it was also clear to me back then eight years ago that there will be a wave around artificial intelligence and that there and I had two hypotheses this could be a space where a lot of money is going in and

it will attract a lot of investment so I thought ok all this together could be a good reason to start a new company which is tackling exactly this problem understanding digital documents in debt in an intelligent way with neural nets so we were part of the digital transformation at least the first wave so now 15 years later I'm still in this phase or in this yeah it's really digital transformation everybody's talking about it and the great thing is now it's super hip to be in that space so back then people looked at me it's like what is this girl 21 years old doing with robots everybody was totally into B to C and this is the new thing but now I mean everybody wants to be part of the journey everybody's talking about artificial intelligence so much hope so much protections hey I will do everything yeah solve all our problems here we have this big problem yeah I was awful so the big question mark I I had when I started this company was

where are we when we think about AI how far away is it generally I generally eye means that we can apply one learning to a different situation so a child two years old you can show the little girl boy how to jump on a chair and it will have no problem to jump on the stage to jump on a sofa if you show it to a robot it will totally fail just because Google goal can win against the best Google Go Go player in the world it cannot win against a chess player so we are still still struggling where are we so in this discussion we thought it's interesting to talk to a lot of people in that space and to try trying to have a reality check so what are the arguments argument one for narrow AI and narrow AI means that we have to Train algorithms very specifically on exactly the task and the requirement we want to achieve neural nets are nothing new they are out there since the 60s and generally

is for dreamers it will take something like hundred fifty years to really apply learnings from one use case to another use case the other party is arguing no everything is different today we have computing power which is enabling all of this wonderful experiments in parallel distributed we have access to so much data we never had before the neural Nets models architecture way more flexible so we were in this struggle and why as a start-up do you even have to ask you yourself this question it's about focus and use case and the value you want to ship to your customer because the world doesn't need a second deep Mandic anymore so as a start-up you confront it with this very pragmatic question if you are in the eye space what's what's the use case what's the real business value what is the confidence call what is the outcome what is in numbers why should someone even think about implementing a new technology and as you all know it's totally different game to upgrade buzzers and rule-based

systems are thinking about replant forming everything to AI driven decision making support systems if you want it's like really a big challenge for every company to implement something like that so that's the reason why as a start-up you need to be very precise in the edit value' you cannot just talk about AI this was this was true three years ago so this was my first pitch and I raised 4.2 million with this pitch it was a slight bag I said yeah I have some data and I mean I and I have credibility and he raised a lot of money it was awesome today the world is one step ahead and people realize okay what exactly are you doing which which process is it exactly you lift up and show me the numbers so horizontal approach versus a vertically approach what does it mean so in our case we struck we were struggling with the question okay is it better to stay horizontal and to claim that we can understand documents from 510 industries and to

kind of take it all this huge opportunity 200 billion as a market it's awesome why not why should an invoice from the farmer space be so different from healthcare in the insurance space or logistics from manufacturing I mean it's kind of the same it's an inverse so we were struggling a lot and the more we kept diving into into the real use cases the more we understood that company who is really in front of this question of replate forming this is what we're talking about if you're talking about huge volume not just a nitty-gritty innovation project every platforming they want to understand what the confidence core is in real time and for us and I'm just talking about our company we we struggle a lot by focusing and narrowing down our space but in parallel as I said before we we invested a lot of time in the research space to understand where is the technology where is the perfect match horizontal approach more generic technology more narrow way to train algorithms more vertical accession

needed so that's the reason why I'm sharing this with you because it's a very important running we had and at some point we decided to focus and it's always a very painful for start-up to focus because as you can imagine as a founder you have as part of your nature that you want to catch the door you don't want to let go any opportunity of course not yeah and if the market is 200 billion I go for a market of 200 billion why should a focus on a market of 24 billion how boring is that so it was very painful for us but as at some point we said okay let's do this bad let's focus on insurance and let's not only vertical eyes on this industry let's vertical eyes on one process and let's become the number one for this single process claim handling and let's become the key driver for this huge transformation so this was the bet we made and why is it exciting for us so to be honest in the

beginning as I shared with you I wanted to crack the nut of the document understanding in general so I didn't start with the idea to transform the insurance base so I didn't see myself in the insurance base when I started with I 15 years ago so why am i excited about this space so we looked at different spaces and we looked at different industries and we tried to find the industry where we can have the most impact so the reasons why we decided to go for insurance all-in with all our energy and all the money we've raised in the mean time over 20 billion so it's because insurance affects everybody everybody here in the room has an insurance who knows exactly what he or she is covered for no one but in the moment we have an accident it's pretty good to have an insurance and everybody who sent in the claim already knows how painful it is it takes a lot of time it's super annoying you don't get a status report you don't

get your money immediately and the products sorry but they are not great they should be personalized they should in the ideally prevent us from having accidents and diseases fraud is a huge topic by a huge topic so there are a lot of things you can improve and in parallel we see that customers are more and more impatient of course I have in my I have my iPhone I myself how do I send in the claim and make five photos from my invoices I send it to Universal my insurance company and I expect them to process it automatically of course I already sent them in my claim up in a digital form what's the problem so they don't get it right so the customer doesn't understand why why'd I have to wait for eight weeks and then they get a decline with any explanation so customers are more and more impatient technology is driving the amount of digital documents making it more complex making pressure on the insurance companies they're in a good position today they

have a lot of money but they understand okay there is something going on and you have to insure tech space you have companies like we Fox lemonade Koya Chetty by the way customers of ours and they are pushing with us there are these big dragons like aliens and AXA and the little startups the intra tech world but we are in line fighting from with each other little little annoying cameras on there on the toes of the big dragons but together we are moving something and that's the beautiful thing of this base and that's why it's so exciting something happened in the last two years so when we started to talk about our vision how we think about insurance and how we want to rethink insurance people were very skeptical in the meantime we got more and more confirmation that's exactly what this industry is is that and they will go through this transformation and we are here and we are waiting for them to start so we have a clear mission and it's we want

to make insurance better and we know exactly how to do that so we are starting with the claim why because here we have a ton of data so in comparison with using AI you can extract something like 7 to 14 data fields with our technology you can extract something like 200 data fields so suddenly we cannot only process the claim automatically the whole process is fueled by intelligent data so it's not only about speed it's about transparency it's about the potential to build new products so we say it's something it's really a different way to think about insurance and it's from process to data-driven and it's the first time we can unlock all this data so I will jump over some slides so in short we give back time because we automate the process of claim handling the claim hits the insurance company in a digital form we take the claim we process it collect all the data and fill the whole process and give recommendation of what should be done we combined approach of

visual and textual technologies so computer vision and natural language processing and we had a huge advantage we had access to a lot of data so with my robot which I invented in 2004 we digitized over time 980 million documents and it's a lot because we are still in the phase where we are talking about narrow AI it was a head of work to select the right documents to choose what exactly do we have to annotate to produce the perfect round truth but still yeah we had an amazing data set we started with over 30 million documents and that helped a lot so we can offer pre-trained models that's a huge difference imagine you would try to sell something to big insurance company and your pitches do you want to sell a product it's awesome but I need something like 10 million documents from you first we start up don't worry data protection it's it's gonna be okay so it helped a lot that we can now offer a product where we need only very few

data few hundred documents to fine-tune the algorithms and of course you need collaborations as a start-up you cannot work like a Research Center that's the reason why you need to collaborate with religious research Research Center to start up you have to act fast and everything must be oriented on building products so we collaborate for example with the computer vision Center in Barcelona and it's very very helpful and for us the most important part was I have to turn here 2 minutes 30 seconds Wow ok scalability of the technical engine so when you talk about a product as a start-up that's my big recommendation you have two things you probably should think about it's the engine it's the tech platform it's all about training retraining failing as fast as possible so you need to build a scalable platform that allows your team to experiments as fast as possible and then you have to end to end product with all the domain language with the understanding of the process because that other reason why we are focusing

on industry natural language processing is based on language as it already says and so the algorithms need to understand the domain language and that's also the reason why it's super tricky to train an algorithm to understand the language from the logistic farmer manufacturing space that's the reason they have to understand the language and we use different words and different ways to describe things differently and different industries so that's why we invested a lot of work and time in feedback loops and and it's like it's it's like like we learn something if we focus on cooking risotto at some point it would be super good in cooking risotto and it's that's the same with the feedback loop you get feedback result is getting worse not a feedback it was not okay so this is this is how it works with algorithms as well so it's happening and the insurance pace is moving that's the amazing thing we have realized and we have big customers who trust us so they are at the start and that's also

the exciting part it's not an industry where we are too late it's an industry where we are very early so we are waiting for our customers we are here to help them they need time to think about it they sometimes even skeptically that I hey I can help them because there is this confusion generally I so if we talk about generally I people are confused because we are not there yet so that's the reason why we have to educate people and also say hey look if we are narrowing down the task then it is wonderful only then in our point of view and there are a lot of different opinions here it takes a lot of energy to change things when I have my own boring sessions with my new piece I always say welcome to the uncomfort zone at omnious because I believe in the yeah I mean it's paid for - if you want to change the whole industry that's not gonna be fun for everyone destruction is painful and we try to

be as sophisticated elegant charming gentle as possible but it takes a lot of energy but not only from our side it also takes a lot of energy from the customer side and I'm very thankful to call these important insurers as our customers and they are helping like crazy they are pushing and they're really they understand although the pain we have and we can go together so we make insurance better but only together and so what would be different more human because we give back time speed transparency more intelligent products preventing us ideally from diseases and insurance we can really use an acceptor for us so join our ride we are hiring we are always searching for a smart and cool partners we want to scale our sales we want to scale our technological partnerships so this is where we are we are convinced that if we want to embrace technologies like AI there are others other great technologies as well we have to embrace change and vulnerability and it's as you can imagine sometimes very

uncomfortable for a big corporate to open up the process and to admit that this or that is not working very well so we appreciate if customers are opening up and yeah it's like a war well sometimes I feel like wow that's a big wall in front this stone is not a stone it's a wall but yeah with our customers together and that's the interesting thing and also with the whole team behind us we are we are pushing this change so I see a huge potential in AI to give that time to make processes more human and that's all for the end of my talk thank you very much for your time thank you you're so perfect in time that you will allow one question is there one question there it is the reason I was going to give you the microphone or she's coming with the microphone thank you so much I have multiple questions of Lila on the ask you've built a narrow AI system to deal with a very specific problem which is

the right way to do it I think to be a design thinking based but the architecture that you've built how much of that would be reusable in a different domain for example electronic healthcare records super good question that's also the reason why we struggle so much 95% the platform the technically platform so in order to achieve a high confidence car you need to add the domain layer so to so I'm talking about the tech platform the engine the engine means that we have a built-in engine which allows very fast experiments training's ground truth annotation so we will collaborate with consultancies with partners and we will hand over our tech platform and allow them to use our pre trained models so we understand the general and layout of an invoice but if you want that this engine to deliver a confidence score over 95% and that's what you need to make your customers happy you need to train to the specific language use case you need to understand exactly what the requirements are they differentiate from

industry to industry but we can help you but it's not what we drew because also for the marketing and sales approach and everything you have to understand the trends so it's it's way more than just to talk about technology yes it's possible that other reason why it's so painful to focus but especially the natural language processing after the NLP models they need on top training

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.