Insurance - How Luko leverages AI to outperform customers’ expectations
Quickly label the invoices received from clients, especially the veterinary invoices for insured dogs and cats.
Get its ML models in production faster and improve the accuracy of automation, quickly absorbing a growing volume of claims.
Improve annotation quality and identify annotation problems with external annotators
Review recurring questions from annotators and exchange daily with them to provide clarification on questions about how to annotate particular invoices.
Significantly enhance the accuracy of the data model by improving labeling quality and therefore achieve more efficient automation.
Externalizing annotations to enhance automation's accuracy faster, thus quickly absorbing an increasing volume of claims and limiting manual tasks performed internally.
Luko is Europe’s first insurtech. Founded in 2018, it has started from scratch and adopted a transparent remuneration model. Only 30% of the contributions go to Luko to cover the operating costs. The 70% left is entirely dedicated to reimbursing the policyholders' claims. Thus, its remuneration is utterly independent of the reimbursements.
With more than 350,000 policyholders in France, Germany, and Spain, Luko leverages AI to provide the best customer experience possible. AI helps Luko with:
estimating insurance risks
estimating damages in the context of claims;
automating claim payments.
Created with the ambition to reinvent home insurance through its transparent offers, Luko differentiates itself from traditional insurance companies through its direct-to-customer model. Its overall model provides a simple, transparent, and sustainable solution. Customers can lock in their insurance by signing up on the website or through the app and enjoy efficient services like live chat and quick reimbursements.
Besides their strong focus on delivering the best policy-holder experience possible, Luko also commits to being beneficial to society. To do so, it leverages:
its economic model: if there’s money left from the 70% after covering the previous year's claims, Luko doesn’t take it as its profits. Instead, Luko injects into its Giveback Program, donating this money to non-profits selected by Luko’s policyholders.
its B-Corp Certification: since 2019, Luko’s BCorp certified. It's a first for a European insurance company and validates its positive impact on society.
Following its success in providing a convenient and user-friendly way of securing housing insurance, Luko is taking its innovative DNA one step further. It shifts into a fully-integrated home care platform:
- preventing domestic accidents;
- eliminating the hassle of claim-filing;
- providing its customers with comprehensive policies – now also including Pet Health and Dog Liability insurances.
Luko’s ambition is built upon its mission to reinvent housing insurance through machine learning to boost coverage and customer satisfaction while offering low-cost subscriptions.
Customer satisfaction is one of Luko's core tenets. When Luko acquired a Pet and Dog Liability insurance company, the company decided to replicate its strategy to deliver a first-in-class customer experience with these new products.
To do so, it focused on the average processing time of claims, pertaining to 4 weeks before the company was acquired. In its mission to improve policy holders' customer experience, Luko's ML team faced a major challenge… There was little automation on the pet health portfolio they took over. As a result, most of the invoices had to be manually processed.
Knowing this, and bearing in mind the objective to drastically shorten the claim processing delay, the company wanted to up its model's processing capability to guarantee quick and efficient invoice management. To achieve this, the insurtech searched for a tool to train the model at scale so that Luko could operate to its full potential. and stop relying on a large human involvement to improve its model's performance. Willing to stop diffusing internal resources, resulting in higher wait times – of about 7 days for customers, Luko has implemented Kili.
Before annotation, about 55% of the invoices processing weren’t automated. This was not compatible with our ambitions. To perform significantly better, we needed to train our AI. The existing process to get this excellent-quality training data required manually labeling each invoice. It was particularly time-consuming and defocusing for the AI team. Consequently, we searched how to annotate invoices more accurately while saving our AI team from being defocused and performing repetitive tasks. We quickly identified Kili as the solution suited to our bandwidth constraints and specific needs.
The Luko tech team developed an AI algorithm to quickly process pet health bills in order to improve customer service. Once the invoices are uploaded to the platform, the AI model recognizes the type of invoice, the amount, the reason for the visit, and other details and processes the reimbursement immediately. This gives Luko a strong competitive edge as traditional insurance companies take up to 4 weeks to process those kind of claims. The AI model enables Luko to provide an outstanding service experience by speeding up the compensation procedure.
However, this innovative approach had a drawback. The model was performing below expectations: only 30-40% of the invoices were adequately processed by the model. This was attributed to the lack of high-quality data to train the AI. Hence, the existing process required a large human involvement, which was particularly time-consuming and distracted the AI team from focusing on other tasks.
We knew we needed to leverage our ML model to guarantee quick and efficient invoice management but still had to improve its accuracy. To do so, we needed a platform to train our data model.
Luko’s main challenge was to improve the performance of its model, thereby reducing the turnaround time for invoice processing and delivering on its primary goal of fostering customer satisfaction.
Kili’s training data platform allowed Luko to industrialize the production of excellent-quality training data. Faster annotation meant less time spent per invoice and hence access to a larger dataset to train the AI model, thereby decreasing project lead times.
Luko also benefitted from Kili’s partner network of outsourced labeling service providers to improve its productivity and quality of the annotation process. The outsourced workforce provided by Kili now handles the time-consuming task of labeling large volumes of invoices. Luko’s team can focus on:
the quality of the resulting dataset;
the project delivery monitoring;
the data workflow orchestration.
Annotation quality is crucial to us. Having someone from Kili in charge to point out potential annotation problems and follow up with the annotators played a role in improving our model’s performance. We also have weekly follow-ups with annotators to review specific recurring cases for annotation. Additionally, we have daily exchanges with annotators as well: as soon as they’ve questions or doubts on how to annotate certain invoices, we can answer them directly from Kili’s platform.
Project management was facilitated by Kili’s customer service teams, who worked closely with Luko teams to guarantee project delivery.
One of Kili’s capabilities that helped Luko was identifying which specific case the AI model was struggling to perform. For the insurtech, this has been done by:
generating results from the model;
and sending those to the annotators, who would identify where the model had failed and correct it.
The gap between what the AI did in the first place and what the annotators needed to fix allowed Luko to clearly mark what type of data class the model was having trouble with. This enabled them to select with more intelligence the data required by the model to improve its performance and therefore reduce the number of needed iterations.
With excellent-quality training data and quick annotation and iteration, Luko improved the model performance from 45% to 70%! This meant shorter wait times for the customers and a more efficient and effective AI model.
Luko’s AI venture also showed drastic improvements in performance thanks to the excellent-quality training data generated and was able to process up to 70% of the invoices – versus 45% before Kili.
Following this boost in performance, Luko’s team is confident that Kili’s platform will help them to reach 85% processing capability within 2 months.
Kili allows us to label every invoice we receive from our customers, including the veterinary invoices for our insured dogs and cats. This labeling has resulted in a significant improvement in the accuracy of our data model. Therefore, we’re more efficient in our automation. Not having to do this annotation ourselves – internally – made our improvement of automation accuracy much faster. Thanks to this, we can now quickly absorb a growing volume of claims.
The swift annotation and iteration process offered by Kili comes with additional benefits. It fastens the development and execution of AI projects, eases collaboration, and empowers Luko to take on additional artificial intelligence projects.
With the increase in automation – and automation accuracy fueled by Kili, Luko has improved its organizational processes. Employees can now focus on other tasks rather than the labor-intensive work of manually handling invoice processing.
Kili’s training data platform, alongside its network of trained labeling workforce service providers, allowed Luko to boost the productivity of its AI resources.
Get started! Build better data now.