AI Integration with 1C: Enhancing ERP with Machine Learning

Artificial Intelligence (AI) is no longer a futuristic concept but a practical tool for enhancing business processes. This article explores how AI technologies can be integrated into 1C, a popular enterprise resource planning (ERP) system, to improve efficiency and decision-making. We'll delve into various AI applications relevant to 1C developers and users, moving beyond the hype to focus on real-world implementations.

AI encompasses a range of technologies, each with its unique capabilities. Key areas include Machine Learning (ML), which involves algorithms that learn from data; Deep Learning, a subset of ML using neural networks with multiple layers; Natural Language Processing (NLP), enabling computers to understand and process human language; and Computer Vision, allowing systems to 'see' and interpret images. Understanding these distinctions is crucial for identifying the right AI solution for specific 1C integration needs. Machine learning is not artificial intelligence, and artificial intelligence is not a neural network. Neural networks are a relatively small and narrow field. For example, they can offer recommendations – when you download a movie/buy a product, and the system offers you similar ones based on your preferences. XGBoost, CatBoost – these are also neural networks, but trained a little differently. There is no artificial intelligence, the Turing test has not yet been passed. Classical linear regression is also machine learning, and in many cases this is quite enough.

Several tools and frameworks facilitate AI development and integration. For Python enthusiasts, Scikit-learn offers a user-friendly entry point, while TensorFlow and Keras are powerful options for more complex tasks like image recognition. .NET developers can leverage Accord.NET. These tools can be used to create microservices that 1C can interact with via HTTP requests, allowing for seamless integration of AI functionalities. Scikit-learn is the easiest, most popular and well-known if you want to learn something in Python. In it you can do something classic - say, train a perceptron. This is three lines of code, and 90% of the tasks will be solved. If you want something heavier - play with pictures, learn to recognize something, write your own OCR, then there is TensorFlow. Moreover, TensorFlow should most often be considered together with Keras, because TensorFlow itself is quite difficult to work with, and Keras is a wrapper that slightly simplifies working with TensorFlow, it is much more pleasant to work with it. Accord.NET will be useful for dot-net developers - a rather old thing, but it solves a lot of issues specifically for .NET.

Computer Vision enables 1C systems to process and analyze images. OpenCV, a popular library, provides tools for tasks like Optical Character Recognition (OCR) and object detection. By integrating Computer Vision, 1C applications can automate document processing, extract data from images, and improve data entry accuracy. For standard tasks, you can try Yandex Vision or OpenCV. Yandex Vision is a great thing, it has a good REST API, recognizes faces and recognizes handwritten text very well. It costs a penny per document, everyone can try it. You subscribe, get a key. In the REST API there is one method, batchAnalyze. With its help, you can transfer a picture, and Yandex Vision returns a JSON structure to you, in which, depending on what you want, either the coordinates of faces or texts in which all the text. Moreover, all this is written literally in three lines. OpenCV is a little heavier, but knows how to do much more. Here, too, the feature detector and face detector are a completely separate story, because most Computer Vision systems are focused specifically on highlighting some specific objects in the picture. And this thing works well. And if you want to somehow classify the text, then the classic Computer Vision will not help so much.

OCR technology allows 1C to automatically extract text from scanned documents and images. While various OCR solutions exist, Abbyy FlexiCapture is a robust option for 1C integration. Tesseract OCR, a free library, offers a cost-effective alternative for basic OCR tasks. By implementing OCR, 1C users can streamline invoice processing, automate data entry, and reduce manual labor. There are people who have already done this for quite a long time and who have also attached knowledge of the Russian language, phonetics, spelling to this. This is Abbyy. And, of course, Abbyy FlexiCapture is still better than all the startups combined. Moreover, it is quite easy to attach to 1C. And Tesseract is great because it's free. With its help, you can make a recognizer in an hour. It will not be of very good quality, it will not be able to fully recognize all the primary documents for entering into 1C. But you don’t need to do this, our future is in electronic document management. But you can recognize passports, you can recognize certain details in the document (for example, a department), and classify this primary document. I’ll show this further in the case studies.

Deep Learning and NLP offer advanced capabilities for 1C systems. Deep Learning can be used for complex tasks like image classification and predictive modeling. NLP enables 1C applications to understand and respond to natural language queries, facilitating chatbot development and automated customer service. For NLP, I would recommend two libraries (maximum, three). Firstly, you can contact the Yandex SpeechKit service - it is made quite well, there is even a good example on Infostart, which implements an appeal to it. For most tasks, such as recognizing phone calls, this will be enough. There is a metaparser, if you, for example, want to extract some specific data from the purpose of payment, but you will have to mess around with grammars there. But it will be difficult for those who have not worked with this directly. There is an open library in Python from http://ipavlov.ai - now this is probably the best solution that works with the Russian language. They use the standard word2vec, but they have improved it. It's simple, convenient. In principle, if you want to make a chatbot that is based on AI, so that it is free, and you train it yourself, then - welcome, ipavlov - the best choice to make a consulting system.

AI can significantly enhance data analysis and forecasting within 1C. Machine learning algorithms can identify patterns in data, predict future trends, and improve decision-making. 1C offers built-in mechanisms for data analysis and forecasting, allowing users to create predictive models and gain valuable insights from their data. For those who are completely 1C-shnik, but want to make some kind of predictive model so that your 1C recommends some products or does some training. Even from 8.0 in 1C there is a wonderful mechanism “Data analysis and forecasting”. Note that these are not neural networks in 1C, these are networks inside the platform, they are written in C++. And you have a decision tree - there is a fairly simple forecast model, and it works quickly, and it also learns quickly. But the pleasant thing is that you just take it - open ITS, look at the example. There is a regular data request. The request is transferred to the analysis model, where there is a set of columns. Then AnalizDannyh.Vypolnit() is written, and at the output you get a model that can be saved somewhere in the value store. This is, in principle, everything you need if you have some kind of simple classification. Moreover, 1C also has cluster analysis, and predictive analytics, sequence search, and building event chains. The toolkit has been around for a long time, the tool is working, but for some reason it is not used.

AI is already being used in various 1C solutions. Examples include forecasting purchases in 1C:ERP, classifying requests in 1C:ITIL, and performing OCR in 1C:Document Management. 1C:Fresh even offers advanced OCR capabilities for creating documents with automatically filled values. These applications demonstrate the practical benefits of integrating AI into 1C systems. AI in 1C is used quite actively. In 1C:ERP for forecasting purchases. I haven’t seen it live, I read it in the announcement. In 1C:ITIL - in the classification of appeals. God himself ordered AI to be used in this way. I saw it live - it works cool, correctly. It is written in 15 minutes. There is nothing terrible here. 1C:Document Management has a simple OCR. Previously, it simply recognized PDF, now it also recognizes images. 1C:Fresh has already appeared not a simple OCR, which not only recognizes text, but also creates documents based on it with automatic filling of values. Moreover, this recognition service seems to have been implemented by the platform developers. There is also the 1C:Products service, a recommendation system for assortment management. The form from this service is shown here on the slide. How this service actually recommends - I don’t know, but according to the description - a cool thing. At a minimum, it clearly did not do without gradient boosting in it.

While it's possible to train neural networks within 1C, it's generally not recommended due to performance limitations. Instead, it's more efficient to leverage external AI services and integrate them with 1C. This approach allows you to take advantage of specialized AI platforms and frameworks while still benefiting from the capabilities of 1C. Is it possible to train a neural network inside 1C? It is possible, but not necessary. It is better to train it somewhere outside, and then simply call it from 1C. The main problem is not the technology, but the fact that the platform is not very well suited for this. There is no normal GPU support, there is no normal support for vector operations, there is no normal support for parallel calculations. Therefore, if you want to train a neural network, it is better to do it somewhere outside, and then simply call it from 1C.

AI offers significant opportunities to enhance 1C systems, from automating document processing to improving data analysis and forecasting. By understanding the different AI technologies and leveraging available tools and frameworks, 1C developers and users can unlock the full potential of AI and drive business innovation. The key is to identify specific use cases and implement AI solutions that address real-world challenges within the 1C environment.