How Keboola uses AI to automate over corporate data
The trend of artificial intelligence (AI for short) has affected not only the field of information technology, but also online marketing. That is not only why TRITON IT is a member of the Czech Association of Artificial Intelligence, where we are constantly trying to gain new knowledge and information in cooperation with other companies.
In order to broaden our horizons and make our services more efficient for our clients, we attended a lecture by Keboola. The lecture was divided into three blocks, spread over several hours, and we took away some important considerations that are definitely worth thinking about, which we have compiled into separate chapters below.
The lack of good data and its value
Quality data is key for AI. As it now turns out, the set of sufficiently high-quality data is not infinite, which can be problematic for effective AI training. Good quality data is in high demand at the moment and companies are, quite understandably, starting to protect it more from being stolen by AI. Quality data can be considered as any internal company resources, documentation of company procedures, or other documents containing company know-how. In contrast, poor quality data can be defined as a variety of easily accessible discussions on the Internet that are not verified and may contain inaccurate or outright false information.
What does this imply? In the long run, it may not be worthwhile to have a 100% transparent company with publicly documented processes, because it exposes the company to the risk that sensitive data will be used by competitors’ artificial intelligence and subsequently used to improve them. What used to be a hallmark of honesty, i.e. transparency, may now be a weakness for many companies and it is certainly not a bad idea to ensure the necessary security measures.
Another risk may be the various private cloud storage systems that smaller companies often use. Their information can be stolen by the companies running the storage without the owners of the stored data noticing. All of this can happen to enrich the AI model.
Data maturity and digital twin
This is basically the concept of how ready a company is for automation. If it owns “mature data”, then it has it available in databases that can be queried by machines. The biggest investment initially is creating the first process to get the company’s data into the repositories. Most companies, in order to achieve the desired result, started data collection with marketing. If a company can store all the key data from marketing, sales, or operations in a data warehouse, it is then able to create all sorts of simulations and predictions of how the company is performing without wasting much time doing this activity, such as creating a digital copy of the company.
In the context of this insight, AI can be used in three basic ways. The first is AI Assist, which is the engagement of AI in existing work, for example, using an AI-generated image to create a web page. The second way is Augment, which is a way of using AI so that an entire section of human work is replaced by AI that independently creates the entire design of a web page. The third and final way is Enhance, which is the use of AI in a way that goes far beyond the scope of human work. As an example, consider a model situation where AI delivers a web page design directly split into template HTML and CSS files.
A quick way to create a corporate AI chatbot
Keboola has introduced a process to extract data from a corporate database using Atlassian’s Confluence tool. Keboola uses its proprietary software through which indexed data is exported to the database for AI, selecting only the relevant information. It then connects the Chat GPT API as a chatbot to Slack. The result is a chatbot that can answer anything documented to junior staff and can even extrapolate recommendations to cases that have not yet been addressed.
Keboola SQL bot as a tool available directly in Keboola
This tool works by having its user write a query for data and its context, referring to a database. A SQL query and text response is then generated. If the user has a hypothesis for marketing, there is no need for him to spend long minutes filtering files and Google Analytics, but he can have an answer almost instantly. Using Keboola’s SQL bot tool, it is also possible to generate highly personalized content, which has proven very useful for mailing campaigns, for example. If a company has all the data from client purchases and marketing linked together, it can have the AI generate an excel file with a completely personalised email to each customer and apply the result in a mailing.