At a glance
- Generative AI produces new output and thus goes beyond classic, purely evaluative AI.
- The basis are Large Language Models (LLMs): systems trained on enormous text corpora that can formulate, explain, and reason in a context-related way.
- Generative AI supports but does not replace entrepreneurial judgment. Responsibility for decisions remains with the human being.
- Great leverage in BI and controlling: automatic commentary on variances, the condensing of planning scenarios, or answering natural-language questions about data holdings are typical use cases.
What questions do you have about generative AI?
Your input is sent to an AI service to answer it and is stored to improve the answers. Please do not enter any personal data. Privacy policy
Generative AI – definition
Generative AI (also: Generative AI or GenAI) refers to a class of AI that independently generates new content based on large training datasets – including texts, images, audio, code, or structured data. In contrast to classic AI methods that analyze or classify existing data or make predictions, generative models synthesize original outputs: they write, formulate, design, and reason based on learned patterns.
The technological foundation consists predominantly of Large Language Models (LLMs), which have been trained on enormous text corpora and respond in a context-specific way to inputs – so-called prompts.
„With the automatic commenting of variances in DeltaMaster, we review and supplement management comments in minutes instead of hours. The judgment stays with us, the generative AI handles the wording.“Head of Controlling
Industrial company
How does generative AI work?
Generative AI systems are predominantly based on so-called transformer architectures. During training, these models learn statistical patterns in huge amounts of data – in the case of language models, typically billions of text documents. The result is a model that, for a given input (the so-called prompt), computes a probability-based continuation or answer.
The decisive difference compared to rule-based AI systems is that generative models do not merely classify or predict but independently synthesize new content. They can summarize, rephrase, or translate texts, write code, or also generate images and graphics.
For enterprise use, base models are often adapted to specific data sources and specialist domains through fine-tuning or retrieval-augmented generation (RAG). In doing so, the model does not only draw on its trained knowledge but also incorporates current, company-internal documents and database queries into its answers.
Trusted by market leaders and hidden champions
Generative AI vs. classic AI: What is the difference between AI and generative AI?
Classic AI methods – such as statistical forecasting models, clustering algorithms, or rule-based systems – are designed to recognize patterns in structured data or to make decisions based on defined rules. They deliver as output, for example, a number, a category, or a yes/no.
Generative AI, on the other hand, produces open, context-sensitive outputs. It can not only calculate a variance analysis but also comment on it in understandable language. It can not only compare several scenarios but also formulate their implications. In this way, generative AI bridges the gap between quantitative analysis and human-readable communication.
| Characteristic | Classical AI | Generative AI |
|---|---|---|
| Output type | Structured (number, class) | Open (text, code, image) |
| Training basis | Structured datasets | Many, heterogeneous information sources |
| Adaptability | Model retraining required | Prompt engineering, RAG, fine-tuning |
| Strength | Precise prediction, classification | Synthesis, explanation, communication |
| Risk | Overfitting, data bias | Hallucinations, loss of control |
| Strength | Precise prediction, classification | Synthesis, explanation, communication |
| Risk | Overfitting, data bias | Hallucinations, loss of control |
What are the opportunities and risks of generative AI for companies?
Opportunities
- Efficiency gains: Accelerates repetitive work such as the automatic summarization of meetings and documents, the creation of presentations, or code generation in software development. Employees gain time for other tasks.
- Lower barrier to accessing data: Instead of complex tools or specialist knowledge, a question in natural language is enough to access data, documents, or systems.
- Democratization of knowledge: Specialist knowledge that was previously concentrated in individual experts can be encoded in models and made accessible to broader groups of users.
- Creativity and idea generation: Can serve as a sparring partner in early phases of work – for brainstorming, drafting texts or concepts, or developing product ideas. It quickly delivers a first draft that can be developed further.
Risks
- Hallucinations: Generative models produce outputs that appear grammatically correct and plausible in content but can be factually wrong.
- Data security and compliance: If internal financial data is transferred to external cloud models, potential risks arise for trade secrets and personal data. Companies must clarify which data may be made accessible to which models.
- Uncritical adoption of outputs: When texts are generated automatically, there is a risk that they are passed on without reflection – with the loss of the professional responsibility that previously lay with the user.
What does generative AI mean for controlling?
Generative AI structurally changes the role of the controller in the company. The work shifts away from manual data aggregation and report production toward the interpretation, validation, and strategic classification of AI-generated results. Controllers increasingly become quality assurers and sparring partners of management. This is a development that many experts have been calling for over the years, but which is now becoming increasingly feasible through technological support.
At the same time, the demands on data competence and critical thinking are rising: those who are to assess AI outputs must understand how these come about, what data underlies them, and where models typically reach their limits. Controlling expertise and AI competence thus grow together inseparably.
Practical example: Generative AI in controlling with DeltaMaster
A controller opens the monthly report in DeltaMaster. Instead of commenting on the variances manually, DeltaMaster automatically generates a context-specific management comment: “The contribution margin in segment B is 11% below the previous year. The main driver is a price decline of 7% as a result of increased competitive pressure; the volume effect of -4% is attributable to a postponed major order.”
The controller reviews, supplements, and approves – in minutes instead of hours. Management receives a concise, audience-appropriate summary that shows not only what was but also what needs to be done. Generative AI takes over the formulation – judgment remains with the human.
What are examples of generative AI? – Use cases in companies
GenAI is no longer a topic of the future – companies across a wide range of industries are already using it productively. An overview of the most important fields of application:
Where would you like to use generative AI?
Your input is sent to an AI service to answer it and is stored to improve the answers. Please do not enter any personal data. Privacy policy
Bissantz and generative AI
Bissantz has long been concerned with the question of how numbers can be not only calculated correctly but also communicated clearly. Generative AI is a natural next step in this: With DeltaMaster, Bissantz integrates GenAI functionalities directly into the analytical workflow – for example, for the automatic commenting on variances or the linguistic preparation of reports. In this way, the technology serves decision support, among other things with the help of the following functionalities:
- Interpretation of anomalies: Variances between plan and actual can be automatically contextualized and interpreted by generative models.
- Natural Language Querying: Users ask questions in natural language – “Which cost centers exceeded their budget by more than 15% in Q3?” – and receive a structured answer directly, without SQL knowledge or navigation effort in the BI system.
- Simulation and planning support: Generative models can not only calculate scenarios but also describe their effects in prose, name risks, and formulate options for action.
- Management summaries: The most important findings are automatically condensed into a concise, audience-appropriate summary – as a reliable basis for management discussions and decisions.
In addition, Bissantz offers comprehensive consulting services on the use of generative AI in controlling.
| Function | Benefit |
|---|---|
| Interpretation of anomalies | plan-actual variances are automatically contextualized and interpreted |
| Natural Language Querying | questions in natural language deliver structured answers without SQL knowledge |
| Simulation and planning support | scenarios are calculated and their effects described in words, including risks and courses of action |
| Management summaries | key findings are condensed into audience-appropriate summaries |
| Automatic commentary on variances | contextual management comments directly within the analytical workflow |
„Today, our departments ask their questions in natural language and receive a structured answer directly, without any SQL knowledge. This has made access to our data significantly easier.“BI Manager
Retail company
Frequently asked questions
What is generative AI, explained simply?
It is a form of AI that does not just evaluate but creates. It reads a prompt – an input or question – and then independently generates new content: a text, an image, code, or an analysis. Like a very well-read assistant who formulates a fitting answer based on everything they have ever read.
What are AI hallucinations and how do you deal with them?
A hallucination refers to output that sounds grammatically correct and plausible but is factually wrong. Dealing with it requires: human review of the output, use of RAG systems that restrict models to reliable data sources, and clear governance that regulates which output may be passed on without review.
What risks arise when internal company data flows into GenAI systems?
If sensitive data is transferred to external cloud models, risks arise for trade secrets, personal data, and regulatory compliance. Companies must clarify which data is made accessible to which models – and whether on-premises solutions or strict data isolation are required.
How does Bissantz use generative AI in DeltaMaster?
DeltaMaster integrates generative AI directly into the analysis and reporting process: variances are automatically commented on, questions in natural language are answered, scenarios are prepared in words, and management summaries are generated automatically. Bissantz also supports companies with governance design and the productive integration of generative AI in controlling.
Summary
Generative AI marks a paradigm shift in how companies deal with data and knowledge: from pure analysis to active communication, from manual report production to automated preparation of insights. For controlling and business intelligence, this means: more time for interpretation and decision-making, less time for data maintenance and formatting. With DeltaMaster, Bissantz integrates generative AI as a natural part of the analytical workflow – transparently.