Get Your Data Ready for Generative AI
Generative AI - a term in the field of new technologies that we are currently encountering in many places. It refers to artificial intelligence (AI) that can create or generate new ideas and content - e.g. text, images, videos, but also conversations and even code. In order to create such content with the help of generative AI, the AI is provided with a large amount of existing data. The AI then learns the structure, pattern, style or design of the data and can thus formulate new texts with similar content.
Imagine you want to create a drawing. By inputting a large number of different drawings, which serve as "training material" for the AI, the AI learns how to draw and draws conclusions about styles, types of work or colors. Based on this knowledge, the AI is then able to create a completely new drawing (as shown in the very simple example below):
(on the left: three templates from stock photos, on the right: new variant generated by Gen AI)
This very simple example illustrates how new content is generated from the input of old content. In reality, this goes far beyond simply drawing a tree - large amounts of data can be processed and analyzed in seconds. At the same time, reports and recommendations can be generated. And that's not all!
But why are many companies not yet ready to apply generative AI to their data management, despite the promise of the possibilities?
The idea of letting artificial intelligence take care of tedious and time-consuming data management sounds appealing at first.
For example, let's say you want to create a text from various features of your product. The AI will do an excellent job of putting the available information into context and shaping it into a good-sounding description.
Let's look at this again using a very simplified example:
If we ask the AI to generate a short description from existing attributes for a product detail page, we are initially helped in that we receive the requested components in a matter of seconds:
Done. Right?
In reality, any product manager or marketing manager would probably start with key attributes of the product and its individual benefits. Or other information would be cut out to create a shorter text for certain print or online channels. And this rule may not apply when technical documentation is required that contains all relevant information but is less promotional in nature.
In this example, what ultimately remains is the problem of specific rules and hierarchies for product communication, which cannot be solved directly by AI and requires preliminary work in the company's own data.
So what (at least) needs to be done before the use of Gen AI in data management is even considered by companies?
1) GRANULAR DATABASE:
As already mentioned, the creation of texts first requires a clean database. The first step is to have a granular database that contains the key product features. This basis allows the AI to formulate fully-fledged and correct product descriptions in the first place.
2) CORPORATE SPEECH:
Every company has its own brand DNA, which must also be reflected in the product texts. Creating texts that match the "corporate language" is only possible if the terminology database is tidy and clean first. Otherwise a screwdriver will end up being a screwdriver again.
3) INVENTORY DATA:
For example, in order for images to be automatically tagged with information such as metadata based on certain rules, the asset data (or at least part of it) must first contain the correct information. This could be the asset category, for example. This information must also be entered manually by humans.
In concrete terms, this means: Before companies can effectively use generative AI, they must have well-structured data that contains additional information and rules for further processing by an AI. In addition, generative AI models require large amounts of high-quality training data to produce reliable and accurate results. However, many organizations struggle with data structure and quality issues, such as incomplete or inconsistent data and rules.
The data that Generative AI uses must be of high quality in terms of accuracy, timeliness, categorization, and other properties for the resulting models to be effective and useful. Poor internal data leads to poor results in generative AI models.
To realize the full potential of Generative AI, organizations must recognize the importance of well-structured data as a prerequisite. By investing in data governance, data quality assurance, and efficient data management practices, organizations can significantly increase the success rate of their generative AI initiatives.