top of page

Artificial Intelligence
for item valuation in 
the insurance industry

An essential but challenging task for insurance companies is determining the value of various items. The value determination itself can play a role in deciding the amount of insurance costs or be used in the event of loss or damage to an item.

The Challenge

Insurance is an industry that is highly dependent on accuracy and precision. Trust in a factual and accurate valuation of the item is of paramount importance to a customer seeking to insure a valuable item. Evaluations, however, can be fraught with difficulty when there are many parameters to consider or when information is lacking. By using Deep Learning in valuation processes, insurance companies can efficiently and instantly analyze images and provided details and combine them with historical data from previous valuations.

The opportunity for deep learning

With Deep Learning, it is possible to summarize data and find patterns in it that are difficult or impossible to see with the human naked eye. It also opens up the possibility of combining and drawing conclusions from different types of information, such as combining images and tabular data. Once a deep learning model is trained and ready to use, it can identify exactly what it should focus on in the incoming images and tabular data to provide the right result.

Platform model to use

Based on what needs to be evaluated and how to perform the evaluation, we recommend either table data regression, image regression, or a combination of both.

How does the model work?

Like any modern Deep Learning, it takes a bit of experimentation to find the optimal solution for the problem at hand. We will focus on using both data sources. To understand and extract information from the images, a conventional neural network is used, and to find relevant features from the table data, a fully connected neural network is used. Both models are available as pre-built models at


Normally, combining different networks for different data types is a challenge only for advanced Deep Learning users. Thanks to's intuitive interface, this can be done in a few seconds without any programming knowledge.

Data requirements

To train the model, historical data with images and their actual rating are required. 

How much data is needed is hard to say without trying it out, because it depends on the task and the variety of data points. A good starting point is a table with at least 100 data points.

Model performance and success

See the performance of the model in the evaluation view on the platform after you run your first experiment. For this type of regression experiment, it is important to look at both the average loss and the standard deviation. 

Where to learn more

If you want to build this yourself with, a suggestion would be to follow our tutorial in the platform.

You need some help to discuss your business case?

Contact our experienced
Sales Team now


bottom of page