Applied AI for Managers - Explained Simply

For Busy Readers

Generative AI helps with automation and cost reduction in the enterprise by applying GenAI models to proprietary company data. GenAI rightfully receives the most attention among available AI technologies today.
PDFs, scans, emails, and multimedia content can be processed automatically using GenAI. Order entry, invoicing, contract management, and numerous searches across company data and applications are common use cases with enormous productivity potential.
Employees and business partners can interact autonomously with applications and data via intelligent chatbots. This eliminates the need for manual typing, copy-pasting, and the tedious task of gathering data from multiple sources.
Like any technology, AI must be mastered and deployed strategically. There are several quality assurance measures and methods for the secure use of confidential company data.

A Tip to Try Out

To see what AI can do with language, check out fireflies.ai. Their software-as-a-service solution records video conferences, creates meeting summaries, extracts next steps, and makes the entire conversation searchable. Practical! How will the major providers react to this?

Influencers and Influence

When a new technology captures people's imagination, things can get eerie. In the best "carnival barker" fashion, technically unskilled individuals often emerge to spread exaggerated expectations and fantastic predictions about the blessings of this innovation. Social media is the perfect breeding ground for such hyperbole.

Competent technicians see through such fireworks of superficial naivety. For them, what matters is how the underlying technical "marvel" works, what tools it uses, where its strengths and weaknesses lie, under what conditions it performs well or poorly—and what can actually be achieved with it.

Far-sighted entrepreneurs examine whether a new technology has substance or will turn out to be a shooting star; what operational and potentially fundamental impacts it may have on their business; what economic value it can generate; and what adjustments to their business model might be necessary.

What is of no use to me is of no value. What harms me makes me think.

Performance

What do GPT models—or the platforms in which they are embedded—actually do? This is important to understand!

GPT models generate:

Using an input, usually a question or instruction known as a "prompt,"
And their "model"—the vast amount of previously trained data (primarily text from Wikipedia, internet content, or source code platforms like GitHub, etc.),
Predictions about data derived from "input" and "experience"—usually text again—known as a "completion."

In short: The AI platform understands the input and, using an algorithm pre-trained on similar tasks, generates a new output that appears to be the work of a human.

But what is AI not? AI—and specifically GPT here—is not "creative." It has no understanding of the content it interprets and generates. AI has nothing to do with "intelligence" in the human sense. GPT models generate continuations of something based on a pre-trained parameterization of the generative model.

Benefits

Analysis of Large Volumes of Information

Given the appropriate input ("prompt"), GPT models can summarize, evaluate, explain, rephrase, or interpret large amounts of textual—as well as visual and acoustic—information. This is their immense strength: they turn data into something else—shorter, differently phrased, or segmented text. This is why they are used for sentiment analysis, summaries, extraction of key points, and paraphrasing or translation.

GPT models are not a replacement for search engines.

Generating Source Code, Making Developers More Productive

With Copilot X for Visual Studio Code, a popular use case has emerged for generating software based on textually formulated requirements. The GPT model generates source code in the desired development language from precise requirements.

Furthermore, a GPT model can understand source code, explain it clearly, document it, create test cases, detect errors, suggest improvements, and translate it from one programming language to another.

Intelligent Bots Instead of Annoying Voice Menus

Customer service bots or helpdesk bots are another area of application for GPT in the enterprise. If you use available information about the customer—e.g., from the CRM—the AI bot can perform relevant and specific tasks. Furthermore, it can control processes—AI should always serve to automate manual work—so that, for example, missing invoices can be retrieved and created in a format specified by the customer.

Application in the Enterprise

Automated Document Processing

Whether PDF, Excel, emails, Word documents, scans, or photos: GPT-based solutions understand the content of these documents and can pass them on to downstream systems. Concrete application examples:

Automatic order entry from heterogeneous PDFs in retail
Transport order entry in logistics
Processing and correct posting of incoming invoices
Generation of outgoing invoices in the format requested by the recipient
Verification of certificates of analysis in the life science industry
Evaluation of delivery notes with subsequent posting in the warehouse management system
Automatic policy issuance for applications in the insurance industry
Creation of medical reports and repair reports based on recorded conversations

Enterprise Information Search

What works with selected documents also works with all company information. GPT models can be applied to blueprints, spare parts, inventory, CRM content, orders, and internet information.

Concrete example applications:

Providing field staff with step-by-step repair instructions
Providing lawyers with concrete answers to a specific case
Compiling sales proposals with all available know-how
Searching for information in seconds instead of laboriously searching through programs and databases

AI helps employees reach a solution step-by-step. The tedious task of gathering distributed information is a thing of the past.

Components

Answer 1: Nothing for now!

Before the technical requirement has been fully understood and broken down into its technical aspects, you should not buy anything. It is all about adapting existing platforms to the company's specific situation.

AI is not a technological niche, but an integral part of all major solutions.

Answer 2: Gain Competence

To achieve a viable solution, deep competence in the use of various AI approaches is required. A superficial understanding of a technology and the use of consumer services like ChatGPT is not enough.

Answer 3: Think and Develop in a Modular Way

Due to the concentration of power among internet giants, one must always assume unfavorable dependencies. No company should put all its eggs in one basket.

Influence on Results and Benefits

The following dependencies determine the quality of the results:

Model Selection - Technical parameters such as parameter count, token limit, and integration capability
Training Data Quality - Relevant, representative data for the problem to be solved
Model Usage - Prompt engineering, zero-shot, one-shot, few-shots, and fine-tuning
Result Optimization - Continuous improvement through error logs and clarification cases

AI in the Enterprise

How should an AI project be set up?

Define the task and the problem exhaustively. Use clear, operational, and concrete descriptions.
Check whether and which AI technology(ies) can handle the task.
Try out the crux of the matter. Money walks, BS talks.
Plan for exceptions. No AI solution delivers 100% perfect results.
Prepare yourself emotionally for difficulties. The technology is young, and the complexity is high.