AI implementation: Getting more out of corporate data

Business man using a tablet computer

Outside of the digital economy, many companies are still in the early stages of adopting Artificial Intelligence (AI). Here’s an example of how AI can unlock the potential of existing data, enable better decision-making, and eliminate repetitive work. There are two keys to success: a method which is appropriate for the application, and a consistent data structure.

AI promises to simplify, accelerate and improve processes in general. In everyday life, we constantly interact with intelligent applications. Many people use them intuitively without giving them much thought. It’s the principle behind chatbots, i.e. automated service dialogs. It enables banking institutions to capture handwritten forms at the touch of a button. Or it simplifies the search for similar images in large databases – Google is a good example for this. Intelligent tools are able to detect and evaluate patterns in a large amount of numerical or semantic information. Another characteristic of AI is that it’s capable of learning from experience and delivering better and better results over time.

Automating reporting routines

Our case study focuses on the project portfolio of an international company. For quite some time, management has already been using a reporting tool to keep an eye on the project landscape, track progress, control budgets, and identify potential synergies. Initially, AI was not applied in this process. Employees had to review and evaluate the reports manually in order to identify trends and correlations that were relevant for decision-making. This procedure required more resources than necessary – and it omitted potentially interesting insights that may be beyond human observation. For example, the sheer number and variety of projects, some organized centrally, others decentralized with external partners, made it difficult to identify bottlenecks early on.

 

Find “green hydrogen” where it’s not even mentioned

This is where AI-based analytics can help, because they cross-read all data, including natural language input. Using statistical measures such as tf-idf (term frequency and inverse document frequency) or neural models such as BERT (bidirectional encoder representations from transformers), the tool can recognize and assess different relevant terms. This makes it possible to assign cases with related content to a category, even if they are described with different terms. Example: Project A deals with “chemical energy sources”, project B with “power-to-x”. If the AI is properly trained, it can derive the common topic “green hydrogen” from such input.

This kind of natural language processing is standard in commercially available software. However, AI tools are still a long way from being plug-and-play. Getting them ready for use requires a lot of upfront work – one of the main reasons why companies are still reluctant to use the intelligent helpers. Two essential prerequisites must be fulfilled:

 

  • Data consistency – All entries must be complete, in a defined format, and comparable in terms of content: from dates to currencies to keywords. Among other things, it must be clarified whether you want to stick to one national language or whether several languages should be processed.

  • A rational purchase-decision – For non-experts, it’s not easy to distinguish between different types of AI. E.g., people often talk about deep learning, a very complex subtype of machine learning: Deep learning is based on neural networks and delivers results that are no longer comprehensible to humans. Instead, an AI based on decision trees might be the better choice for many applications. Before purchasing an AI tool, companies should conduct a thorough cost-benefit analysis.

 

Experience shows that about 70 percent of the initial work consists of analyzing the data structure and standardizing data. Another aspect that shouldn’t be underestimated is change management. In our case study, the AI project must address all employees who are involved in reporting or who supply data. After the AI launch phase, it will be up to the users to ensure that all data is up-to-date and complete. The project team must convince all stakeholders of the benefits and provide them with appropriate training. If the company lacks internal AI expertise, now is the time to build up a team.

 

Scaling up the new technology

Companies that have not ventured into AI before can introduce the technology step by step. In an agile approach, for example, they might start with a reporting application to gain initial experiences. If the project delivers good results, more data can be incorporated, or the company deploys the technology in additional areas. Across different operational units, the manual effort in data preparation and analysis will be significantly reduced. Soon, the employees will use AI as effectively and naturally as we know it from other everyday applications.

 

2022-01-07, Grosse-Hornke

Weitere Artikel

2021 Grosse-Hornke Private Consult