AI Guidelines for Businesses: Using AI in Your Own Company
- Determine and Define the Success Criteria
Once the correct use case has been identified, it is also easier to determine the next steps, such as defining the success criteria. Not all available solutions are suitable for every use case. For this reason, a few parameters, so-called “success criteria,” need to be defined in advance.
These criteria are:
♦ Business needs
♦ Data sources and data quality
♦ Semantic relationships and extraction
♦ ROI calculation
Business needs refer to the discrepancy between the actual situation (real-world practice) and the target situation when the solution has been optimally implemented. After the specific discrepancies have been formulated and the implementation requirements have been defined, companies need to know what data they need from which data sources in order to achieve the desired objectives. After all, businesses possess a tremendous wealth of data that is stored in countless different sources—or perhaps has even been forgotten. Sensor/machine data, big data, documents, Internet, SharePoint, and historical company data are just a few examples. In order to extract information from this existing data and to make it usable in the form of knowledge, the relationships between the various pieces of information must be extracted and models need to be constructed so that further information can then be correctly interpreted and interlinked.
In terms of impact monitoring, it is necessary to define meaningful KPIs (Key Performance Indicators) in advance that can be understood by both employees and other stakeholders. These provide the hard facts used to measure success and serve as the basis for the ROI calculation.
- Hands-on Testing With Company Data
A proof of concept (PoC) is an important milestone for implementing AI solutions. It provides the foundation for further decisions and should help separate the wheat from the chaff among the vendors. In our experience, it doesn’t make much sense to try to establish the PoC simply by presenting a set of slides. The customer should use their own data to test whether the identified requirements can actually be met using the solution. At the same time, a PoC using a company’s own data makes it possible to identify problems at an early stage, and the results can then be integrated seamlessly into live operation.
One factor that should not be underestimated when it comes to successfully implementing AI-based solutions is the quality of the existing company data. Many challenges can be solved with the help of AI, as long as the required data is of reasonable quality. “Data garbage” can significantly complicate the use of machine learning and AI. For example, if the data is incomplete, present in multiple versions in different data sources, contains inconsistencies, spelling, punctuation or other general errors, it will be difficult to obtain correct and relevant results (garbage-in-garbage-out principle).
To avoid this, the existing data in the various data sources needs to undergo a more detailed examination beforehand:
View and understand the data: It is necessary at this point to consider which data sources are available in the company and whether all these data sources are relevant for the planned PoC. It may prove useful to first work with and learn from the most relevant data sources. When selecting the solution, care should be taken to ensure that the pool of data sources can be expanded as required.
Data cleansing: The existing data has to be subjected to a critical examination to determine whether all existing file shares are still needed.
Process and link the data appropriately: The better the chosen product is, the less manual effort this step will entail. The ability to automatically recognize essential correlations and, most importantly, to render them visible to the user together with the possibility for providing feedback is a key factor to consider when selecting the right tool.
The steps described above can be carried out in a very short time but are immensely important for the quality of the results.