ChatECM: The Impact of Artificial Intelligence and Machine Learning on Enterprise Content Management
Artificial Intelligence (AI) and Machine Learning (ML) have the potential to revolutionise the way businesses operate. The Enterprise Content Management (ECM) industry is surely no exception. The integration of AI and ML into ECM systems allow organisations to automate routine tasks, enhance search and gain deeper insights from their content. An intelligent content management system aims to make the document management process more efficient while saving time and reducing errors.
One of the primary benefits of AI and ML in ECM is the automation of routine tasks. AI and ML can classify documents, extract key information (metadata values) and automate workflows. For instance, new documents are added to an inbox where the ML models will then later extract the needed metadata, document type and place the document in the correct location. This improves the overall efficiency of the organisation while allowing employees to focus on more strategic tasks. These models can not only be used on new documents but also on existing documents to validate the correctness of their metadata.
Improved search and discovery is another benefit of AI and ML in ECM. With these technologies, ECM systems can provide more accurate and relevant search results, making it easier for employees to find the content they need. Intelligent search can even suggest recommended content based on an individual’s search history, further enhancing the search experience. Furthermore, ML models can identify and tag document types that are frequently occurring (even when they are not defined in the typical “content model”). For instance, search queries such as “Please give me all the invoices of this folder” and much more become possible.
Integration of intelligent insight models into ECM systems also has the possibility to provide organisations with deeper insights into their content. ML models can analyse large amounts of data, identifying patterns, trends, and insights that may be difficult to detect manually. This information can be used to make informed decisions about content and structure, improving the overall management and governance in ECM systems.
As an experiment, we have used an unsupervised LDA (Latent Dirichlet Allocation) model that could find similarities, identify patterns between all the organisation’s documents. In short, the LDA model works by finding hidden latent groupings within the content, the so called topics. Each document is then later represented by a distribution of topics which are in turn represented by a distribution of words. This allows us to create an alternative view / grouping of documents with similar topics, potentially giving the users new insights about alternative ways to structure documents. An additional bonus to these distributions of topics is the possibility to quantitatively measure the similarity between two given documents.
Chat GPT as an ECM integration
To illustrate the impact of AI and ML on ECM, we experimented with the possibility of using ChatGPT alongside an ECM system. In this example we provided the content of an invoice for ChatGPT to analyse. As you can see, the model was able to understand that the provided content was an invoice and able to find the key information that might be critical to your business i.e.: the unique identifier VKF-2023-010. Another possibility to use ChatGPT is to ask for a short summary of the document so that it could be quickly interpreted.
However, there are some limitations of what can be done with ChatGPT. The model only has information/knowledge of what is publicly available. Our vision at Xenit, is to train these state of the art model architectures (Transformers) on your organisation’s data so that you can have the same feeling as working with the magic of ChatGPT but with the knowledge of your own data. This way, the model can answer specific questions aimed towards your own organisation data. An additional benefit to training your own models is that your data never has to leave our secure network.
In the context of metadata extraction we have already trained a transformer (LayoutLmv3 model developed by Microsoft) on data of one of our clients in the health insurance business. The model is able to understand the content and layout of documents and can extract the defined metadata very accurately. This same model architecture can also be used to answer queries written in natural language (such as chatGPT) or be used for document classification. The results are very promising and we will for sure keep on investigating the possibilities with transformers.
In conclusion, AI and ML are going to change the way we manage content, and the benefits of these technologies in ECM are clear. By automating routine tasks, enhancing search and discovery, and providing deeper insights into content, AI and ML are improving organisational efficiency and driving better decision-making. As these technologies continue to evolve, we will evolve alongside them and create even more impactful applications in the ECM industry.
Additional Resources