How to Select the Right AI Projects based on LLMs
There is currently a lot of discussion around how companies choose their AI initiatives – and just as often, how many of these initiatives are likely to fail. At sol4data, we follow a structured, hands-on approach grounded in experience with data-driven organizations and scenarios.
A) Focus on Business Context
The first and most critical step is evaluating the underlying business scenario. In practical terms, implementing AI is only worthwhile if it clearly adds value – whether by increasing appeal to internal or external customers, or by significantly reducing costs. Ultimately, it must contribute to revenue and/or profit.
What sounds obvious is rarely simple in reality. These considerations are similar to those of a product manager: Where is the value created? For whom? And with what impact?
B) Questions Around the Data Product
The next step involves evaluating the data situation and asking the same questions one would ask when building a data product. Here are a few examples:
- Does the company have the relevant data, and is it allowed to use it continuously?
- Is ongoing data provision and updating ensured?
- What is the quality of the data, and what framework conditions need to be observed?
At this stage, data architects, IT experts, and product owners come into play – people who are familiar with both the data landscape and the relevant processes.
Even when the questions sound simple (“Do we have the data?”), the reality is usually complex and nuanced. Topics like data quality, GDPR compliance, end-to-end architectures, or concepts like data mesh require thorough consideration. It’s not just about feasibility, but also long-term cost implications.
C) AI Fit & Scenario Evaluation
If the business and data foundations are solid, the next question is: Is AI the right solution for this problem? Here, AI architects take the lead.
We categorize AI scenarios into three fundamental types:
- Deterministic – Well-defined, rule-based tasks that are often solvable with traditional software, without the need for AI.
- Heuristic – Complex, often semi-structured problems where AI workflows or agentic AI offer clear advantages.
- Epistemological – Open-ended, context-dependent questions (e.g., “What is the next successful marketing campaign?”). AI can assist here, but is unlikely to deliver reliable decisions.
Based on the answers to the above questions, an initial system design can be created. These three dimensions – business context, data product, and AI scenario – are not a one-time checklist but an iterative process.
Success Factor: Stakeholder Alignment
A frequently underestimated factor: alignment between product owners and technologists is essential. Only when all relevant stakeholders – from business units to the data team to IT – work towards a shared goal can an AI project become a true business success.
sol4data offers a solutions portfolio for Agentic AI implementation.
