Due to the reduction in effort and the simplicity of the application, companies can focus more on the business perspective. The technical complexity and risk of introducing these processes is significantly lower compared to previous AI projects.
The Foundation Solution Matrix (see figure) provides managers with an overview of possible foundation use cases for various company roles.
For companies, this offers a wide range of new areas of application with very low entry barriers at the same time. AI-based processes can now be used for the first time by classic management roles in marketing, finance, HR or legal to generate typical work results in an automated or semi-automated way:
In this way, companies can leverage high efficiency potentials and achieve quality improvements.
There are already a large number of use cases that can be successfully implemented with foundation models. Figure 3 shows examples of different use cases based on the dimensions of business value and feasibility. A high level of feasibility means, for example, that the AI can be used quickly and easily as an out-of-the-box model without company-specific adaptations. If, on the other hand, company-specific and domain-specific models have to be developed and trained, the feasibility is comparatively low. For example, social media posts and landing pages can already be written well with out-of-the-box solutions. A recent study “Man-Machine – Who writes the better content” shows the possibilities and limits of Transformer models for content marketing 6.2 SWOT-Bot for corporate strategy
Automatic creation of SWOT analyses to support management processes and decision-making. For this purpose, relevant sources are automatically crawled to create a knowledge base from a large number of articles and earning calls for each S&P 500 company. In addition, companies can also add specific sources.
The SWOT bot uses a tuned language model based on DeBERTa (Deco-ding-enhanced BERT with disentangled attention) model. It has been trained with tau-sends of examples of strategic questions. The result is a SWOT PowerPoint presentation.
Unlike ChatGPT, the SWOT bot uses a multi-step process based on the two models DeBERTa and GPT-3 to ensure that all insights are supported by evidence from a company’s news reports or earnings announcements. Using this knowledge, summaries are produced using GPT-3. The support of the language models saved an average of 310 minutes in the creation of the final SWOT analyses.
E-commerce is becoming increasingly important, but also more competitive. Companies are competing with numerous price comparison sites, e-shops and products. This makes it increasingly important for companies to quickly and scalably generate product and category texts for e-shops that have a high visibility and conversion probability. Texts can be generated automatically on the basis of GPT-3. When creating texts, customer and market data as well as company-specific data are taken into account and thus content is created that combines the customer and product view. A human-in-the-loop editor ensures quality assurance and systematic optimisation via feedback. The editorial effort could thus be demonstrably reduced by 83 %.
This means that companies can not only massively save costs in the creation of product texts, but also increase the conversion and thus sales potential – and all this in the language of the respective country.
Automatic content generation in marketing along the customer journey (cf. Figure). Texts and images were generated for the different phases of the customer journey using various foundation models (for texts: GPT-3 Playg-round and ChatGPT from OpenAI; for images: Dall-E 2 from OpenAI). These were compared with the editorially created content in a survey of 100 consumers. Both the consumer perception and the quality metrics put the AI-generated content ahead in all other phases except the awareness phase.
In all phases of the customer journey, AI-generated content is preferred over editorial content. The exception is texts in the awareness phase. Here, the editorial texts are preferred: Not only AI-generated texts but also AI-generated images are predominantly preferred.
There are no significant differences in terms of gender and age of the consumers surveyed.
Overall, AI content leads in the criteria readability, formulations, entertainment and credibility. In the Creativity criterion, editorial text is rated better overall.
The GPT-3 Playground performed slightly better overall than ChatGPT.
The Flesch Index (measures readability) is higher than average for AI-generated content.
The Perplexity Index (randomness of a text) varies unsystematically across all texts. Based on the Flesch Index and the Perplexity Index, no clear patterns can be derived with regard to AI-generated and editorial content. This means that AI-generated content cannot be systematically detected.
As the e-commerce shop of Teltec AG – market leader in the B-to-B sector for professional camera and studio equipment – becomes more and more important, the need to support digital consulting and purchasing processes in a digitally scalable way also increases. In addition, customers in the e-commerce sector expect more and more 24/7/365 services.
In order to provide interested parties and customers with a good customer experience, no FAQs or predefined question-answer categories of a simple chat bot will help. In addition, the expert bot is intended to relieve Teltec’s staff through AI process optimisation, thus also counteracting the acute shortage of skilled workers and increasing customer satisfaction in the long term.
The automated advice to customers is to be provided in as natural a language as possible, taking into account dialogues that have already taken place and the information from the 20,000 existing products. For this purpose, a corresponding Expert Bot was developed on the basis of GPT-3 and Conversational Module Lumi from Aleph Alpha along a maturity model.
Using OpenAI’s DALL∙E 2 AI system to generate personalised images to illustrate individual podcast episodes, rather than sticking to the same podcast cover.
RTL Germany is currently testing how DALL∙E 2 can be used to personalise content. One idea is to use DALL∙E 2 as a cover bot to generate unique images to illustrate individual podcast episodes instead of sticking to the same podcast cover. The challenges are particularly in the real-time quality assurance of the personalised content. It must be ensured that images appropriate to the child’s age are generated and played out.
Foundation model such as ChatGPT can support various aspects of risk management:
Particularly exciting is the combination of ChatGPT with a Business Knowledge Graph as an integrated lever for risk management.
On the other hand, ChatGPT can increase the readability and usability of a Business Knowledge Graph:
The integration of the Knowledge Graph also increases the accuracy and quality of the generated texts. Through the contextual and factual knowledge of the Knowledge Graph, a kind of quality management can be implemented.
© 2025 · Peter Gentsch
