We help companies automate
processes using methods
of artificial intelligence (AI)
Methods of artificial intelligence in practice
A whole gamut of administrative tasks in companies can be carried out more efficiently or become completely automated by applying artificial intelligence methods, enabling considerable cost savings. The BBS news agency has even created a special website on the theme of „Will a robot take your job?” which presents an overview of employment positions and their likelihood of being automated.
Our projects aim to help clients with implementing AI methods in their current IT environment. From the numerous options available in AI, we especially concentrate on the methods outlined below.
Decision tree learning
Decision tree algorithms are based on decision-making diagrams or models in the form of a tree. These models can take various forms, from simple types, or more complex ones incorporating probabilities and costs, through to the involvement of machine learning. Machine learning makes it possible for such diagrams and models to adjust themselves on the basis of changes in entry parameters. A typical example of its utilization is a set of tools developed by our company called the BizRules Studio, which models and calculates business rules in an organization.
Cluster analysis
Clustering is used to discern clusters, i.e. groups of objects, which bear greater similarity to each other than objects within other clusters. Unlike classification, this is referred to as “learning without a teacher”. In practice, that means that cluster analysis is ideally suited in instances where it is not possible to state which objects or components under study are similar to each other prior to applying AI.
Rule-based machine learning
Programming sequence algorithms for various cases of artificial intelligence is rather demanding. This is especially true when insufficient data exists to utilize standard machine learning methods. However, the basis of RBML is that a system always contains only relations and rules that jointly describe the knowledge of the given system. The system actually identifies these rules and relations itself, so there is no need to create them manually. These properties are useful in the increasingly progressive area of Big Data and data-mining.
Pattern Recognition
This subsection of machine learning deals with recognizing repetitive patterns in data. In order to identify such patterns it is possible to apply classification algorithms (with a teacher) as well as cluster analysis (without a teacher). ICR (Intelligent Character Recognition) algorithms developed by our company employ classification algorithms based on complex multilayer neural networks.
Neural Networks
This approach to algorithmization imitates nature, as neural networks work in the same way as the human brain. They typically consist of several layers of neurons which carry signals throughout the network, including those of backpropagation, which helps the network to learn through finding its balance. The utilization of this technology is varied, as is true for other machine learning methods, our projects having investigated it for specifically sound and image analysis purposes.
To determine which processes should be in your company could be automated using AI?
Write us!
Selected references
Analysis of sound waves using neural networks
Analysis of sound waves with the use of neural networks. Classifying waves is carried out with the use of so-called Fourier transformation algorithms, signal convolution for a selected window function, spectrogram analysis, a two-dimensional mutual correlation function and feed-forward multi-layer neural networks.
VOP – 026 Šternberk, s.p.
Design and implementation of algorithms for recognizing and classifying the sound of weapon systems.
Digitizing handwriting (ICR, IWR)
We are developing a system to recognize handwritten text, which combines OCR (Optical Character Recognition), ICR (Intelligent Character Recognition) and IWR (Intelligent Word Recognition). This self-learning system uses neural networks and is designated for back office companies for data-mining from hand-written data on forms. Our company is coming up with a complete solution for this process, i.e. digitizing documents, image processing allowing utilization of ICR to achieve the best possible results and implementing a hybrid method based on ICR and IWR.
Decision trees, Machine learning
As a part of our internal projects, and in cooperation with specialists from Tomas Bata University in Zlín, we are developing tools based on decision tree principles that apply advanced machine learning methods. The fresh approaches developed are also employed alongside other methods in the BizRules Studio.
BizRules Studio
is a suite of software tools dedicated to defining, modelling and calculating business rules applied in the decision-making processes of an organization or company. It is worth noting that we have been refining the product for a number of years now, starting out in 2010. The new version of BizRules Studio enables, in addition to calculation patterns, interconnection to company processes (workflow), hence represents the complete automation of certain business decisions. It also constitutes an automated solution for minor operational issues.
Read about us in the magazine Bankovnictví
Artificial Intelligence will never replace a financial consultant
This is because consultancy relies so much on personal contact. Financial products are complex, and clients want to gain a thorough understanding of them. They need someone who is patient enough to explain everything and set up matters correctly. However, most other processes in financial consultancy can be fully automated or significantly accelerated using AI methods.
Fincentrum a.s., one of the most significant companies involved in financial consultancy in the Czech and Slovak markets, generates an annual turnover of about 1.5 billion CZK. The company has hundreds of financial consultants on their books, who close dozens of contracts with clients on a daily basis. As a result, the contracts - in paper form - need to be processed and then sent to financial institutions. That demands a huge amount of manpower, so the intention is that these tasks shall eventually become fully automated by implementing AI methods, or more precisely by applying machine learning models. As a part of our cooperation with Fincentrum we have become their main software supplier. It is our job to develop complex information systems, and carry out research and development into algorithms in the sphere of Artificial Intelligence. In fact, some of these algorithms have been employed in applications created for Fincentrum.
How can AI be utilized?
A standard algorithm is a set sequence of instructions for processing inputs. AI methods are used in cases where the input and required output are known, but constructing an exact algorithm would be too time-consuming or completely impossible. With the use of machine learning methods, the computer gradually creates an adequate procedure of computation. In practice, that can relate to tasks such as automating company processes (workflow), by employing large-scale decision trees or pattern recognition applied to handwritten text.
Challenges faced in automation
It is always necessary to take into consideration the abilities of modern intelligent algorithms and the computing capacity of the equipment. For instance, the widely-held opinion that recognition of handwritten text has been solved is not actually true, and that is even more true with regard to the Czech language. Nevertheless, printed text poses no problem to AI. For example, through our cooperation with Fincentrum, we often come up with suggestions for improvements or propose ideas for entirely new applications. As a part of implementing analysis in a project we carried out a feasibility study with employees from the back office department. The aim was to find out if it was possible to speed up the manual transfer of scanned contracts into digital form by using methods for digital recognition of handwritten texts. The aim was simple, instead
of typing all the data in from a contract, part of the data is recognized digitally and the back office employee would only then be required to check the digitized contract. Thanks to the instigating an intelligent algorithm we achieved a 95% success rate in recognizing handwritten text. To our surprise, it took much longer to check the document than to enter the data manually.
The volume of data is rising as the number of consultants increases
The numerous interactions between financial consultants and clients generate a huge volume of data, not just contracts but also information gained through meetings with potential clients. In the past, such data was considered a side product of algorithmization, but today data is effectively an asset, and literally a gold mine for the future decision-making of the company. We need not talk only of big data in telecommunication companies – consultancy companies also create tons of data to be analysed via AI. Employment of machine learning methods will make it possible to carry out greater analysis and divide clients into specialized groups. By developing systems for making recommendations, it will ultimately be possible for consultants to offer clients products tailored purely for them. This shall lead to more conversion in the system and increased efficiency in concluding contracts in the future.
A robot can never replace a consultant
Financial consultancy primarily relies on the relationship with the client and that is not easily replaced by a robot or artificial intelligence methods. However, all the steps leading up to the time of personal contact with the client shall gradually become automated. Some companies are already experimenting with or have introduced special talking robots, known as chatbots. In effect, these are a simple text editor that handles clients’ questions, where the answers are not supplied by a company employee but a digital assistant. Thanks to the advanced AI methods in place, the answers actually look as if they were written by a human. When the digital assistant fails to answer a more complicated question, it will forward it automatically to the appropriate member of staff in the customer support department.
Jan Pálka graduated in Artificial Intelligence from the Faculty of Applied Informatics at Tomas Bata University in Zlín and Universidade de Vigo in Spain. The topic of his thesis and other scientific works comprised the application of neural networks in practice, including the sphere of ICR pertaining to the specific nature of the Czech and Slovak languages. Since 2005 he has been developing desktop, web-based and mobile applications. He has authored several scientific works on artificial intelligence and data-mining.
Read about us in the magazine Bankovnictví
All of your questions help us to improve. Please don’t hesitate to ask.
Or you can follow us on our social media sites
© 2019 Business Logic s.r.o.
Custom software development according to your needs.