Insolver library changes pricing for an insurance company. In 2018 we as a insover team decided to create a product that helps insurance companies to solve certain calculation and deployment problems. As analysts and participants of different projects, we thought about which problems are insurance companies meet very often? One way or another, insurers are trying to look for innovative approaches for their typical tasks. Many small-sized insurance companies and medium-sized insurance companies also often require standard solutions for pricing systems and risk assessment systems. Potential clients may have issues such as fraud rings. Such solutions as the analysis of internal processes, the search for the so-called bottlenecks, and the acceleration of the process are also in demand. The problem of building communications with intermediaries and agents is an example of a bottleneck in the interaction of a person with a machine, and we would also like to draw attention to this problem.
In addition to typical business problems, insurers are faced with fairly typical problems related to dirty data, incomprehensibly filled fields, failures resulting from failures in systems. Incomplete description of data and fields in their data. In particular, data analysts require a complete description of table attributes and their contents.
One of the problems faced by analysts in insurance companies is the problem of too much data and a lack of resources to process it. Companies are struggling with these problems using various methods. We also want to offer a solution related to resources for computing. Insurers are also faced with the problem of introducing modern solutions to a modern library, and it is often unnecessary to spend time implementing these technologies.
It’s profitable for commercial companies to use a finished product that solves implementation problems. In the process of implementation, the company faces many difficulties and asks many questions such as who will be involved in changing the models when the data changes and how this will happen, the problem of the computation point where exactly the computation takes place, on which servers in the Cloud or in the perimeter of the companies. One of the complex problems is the flow of data between environments when it is not entirely clear which data is flowing and how.
Over time, the use of the model poses the problem of degradation of the irrelevant services of model models, which requires updating and operational implementation in the usage circuit, and these tasks are close to the concept of automatic machine learning autoML; other companies that also study this type of insurers more often have to deal with.
And all these processes are seasoned with security issues of data processing and data transfer storage of these data, taking into account personal data.
Pricing is the basis for these processes of building a system of rating apart, the operation of seams and problems, in particular, simplification of pricing, reduction of time for risk assessment during coding, an overall reduction in the time spent by a client for purchasing a policy, has been decided. The actual problem of the quality of risk assessment is too much data for risk assessment, which leads to a probably incorrect risk assessment.
One of the problems is the cost of the formation of pricing, another is understanding by the insurer of the correctness of this pricing of its validation relative to the market.
Insurers are faced with the problems of reducing the time we finish at various stages of the sale of settlement payments, operating activities, applying the policy of the system, and so on.
All these tasks the company is trying to solve independently within itself and the time spent on solving these problems depends, among other things, on the software that the company chooses to use. We all know that in the end, people perform the given task and the problem of selecting the right employees is the right staff structure in the process of the company is an important part of solving all these problems.
The popularity of the profession of machine learning analytics has led to the emergence of a lot of haha specialists who are not fully familiar with machine learning and the data for such people is extremely convenient to use ready-made solutions that can reduce the cost of errors that they can make.
Even professional analysts face peculiarities of data behavior, in particular, the use of factors or the use of various error functions that are not typical for solving this problem can lead to errors and inaccuracies in the work process and the value can affect the company’s business.
Traditionally, We are used to the fact that companies have a division into the data processing departments of the analytics department and the implementation department, as a rule, data processing departments are specialized in running databases.
Often, analysts within companies do not want or are not able to code complex models to create code, and accordingly, one of the main advantages of products is low code solutions for typical problem-solving.
To solve new challenges we offer analysts the product Insolver, which is designed to these problems. The first block is open source, which allows you to include several basic components of a universal analyzer, recalls the approach to autoML.
The second large block of our product is service modules.
The 3rd module is what the proprietary module is called that allow you to solve highly specialized tasks that can be built for a specific customer and includes, among other things, the implementation of modern state-of-the-art algorithms.
To build a service system that could be installed both in the perimeter of the customer and work in the Cloud according to the SAAS principle, the service could ensure the ease and speed of data processing, building models. Of the key components, we plan to implement and support modularity, flexibility, speed and simplicity, and we plan to support the RPC protocol for interacting with heavy requests.