In this blog we have been talking about the facets of artificial intelligence for years now. But the most requested use, after years of data collection, has absolutely nothing to do with the main activity of Enterprise Statistics.
If you probably follow me on instagram, and you pay attention, you will have noticed it, but the most frequent use request for AI has nothing to do with data analysis, for example for preventive maintenance purposes or more generally for predictive statistical models, but with obtaining data. This means that AI does not necessarily have to be used but the boring but still effective IT technology of a few decades ago can be used.
Specifically the relative majority of the demand asks to recover data from documents, especially PDFs, and not from websites. These PDFs, when things go well, come from Word documents, Excel, etc., in short, something that is still part of the digital world. When it goes wrong it involves scans of handwritten documents.
By relative majority I mean that if I collect, for example, 10 thousand requests and I can group them into 100 categories, the first category, the subject of this article, involves 10% of the total cases. Therefore it is a distribution strongly skewed to the left and with a very long tail.
Clearly this is my experience, therefore anecdotal evidence.
This demand certainly arises from a market that has an entrepreneurial cohort of a certain age, especially in certain countries, which consequently has business processes that are not yet entirely digital. This article may have the side effect of suggesting a business strategy to some IT company, making it hyperspecialize on that type of question. But I return to the demographic consideration: that relative majority certainly attracts profits, but what sustainability do they have? 5 years? Maybe 10?
An example covered on Instagram:
Gen AI can be applied to an archive of tenancy documents, which can be dense and full of specific terminology, making it difficult for owners of many properties to sift through and find information at scale. An AI-powered tool can summarize key issues in leases, such as the amount of expected monthly rent or market forces (such as local environmental, social and governance compliance laws) that may affect leases. Additionally, the tool can scan leases for a particular parameter (for example, all leases with a rent price per square foot below a certain level) and generate tables of information. At that point, professionals can review the information gathered by the AI tool.
Tools:
longchain, lambda, openAI
Build the stateless chatbot using openAI lambda langchain deployed on AWS
In this case, artificial vision, i.e. the multimodality of generative artificial intelligence, allows you to obtain information from documents for renting a house. There are clearly various types of computer vision, which also means less potential costs for maintaining the solution.
Want to chat about your business processes dealing with unstructured data like documents? Let’s talk in a free first call.