MIT’s The GenAI Divide: State of AI in Business 2025 report found that only five percent of enterprise AI projects succeed past the proof-of-concept stage and move into production. That means there are significant investments in AI but business benefits are hard to find. This is not because the technology is failing. It is because the approach companies take is.

In many organizations, AI is treated as an isolated innovation exercise rather than as part of a broader transformation. The result is predictable. Proofs-of-concept demonstrate what might be possible, but they do not change how decisions are made, processes run, or businesses operate. The gap between experimentation and enterprise value remains wide.

There are, however, identifiable success factors. With the right framework, it is possible to create scalable, valuable AI solutions that will make a real difference to the enterprise.

A data-first mindset

The first major stumbling block is data. Another study found 43 percent of AI projects fail because of poor data quality. Complete and consistent data will fulfill a business, not just a technical threshold. There are a few scenarios where precision is less critical but, for the majority of analytical use cases that SAP customers focus on, data quality is absolutely critical.

Poor data does not remain contained. It is amplified. When AI models are trained or prompted with inconsistent or incorrect data, the outcome is not just slightly off – it can be fundamentally misleading. Hallucinations, bias, and overconfidence are not abstract risks. They are direct consequences of weak data foundations. And perhaps more dangerously, these issues are often not immediately visible. They surface only when decisions have already been influenced.

This is why the move to SAP S/4HANA and modern data platforms should not be seen as a technical migration alone. It is a rare opportunity to correct structural data issues and establish what we would call business-ready, AI-ready data. That means data that is not only technically available, but also trusted, consistent, and aligned with how the business actually operates.

The real value of data

Ensuring data is business-ready is the foundation, but the real value starts when the right data set is implemented at the right point in time – when it is relevant to the influenced decision or prediction being made by AI, so the data cannot be outdated or inaccurate.

At the same time, it is important to challenge another common assumption: more data automatically leads to better outcomes. It does not. What matters is not volume, but relevance and timing. Data must be accurate at the moment it is used and relevant to the decision it is meant to support. A perfectly harmonized dataset that is outdated is of limited value. Equally, a large dataset filled with irrelevant attributes or manually-entered free text creates noise rather than insight.

Getting data business-ready is not a straight line. There are multiple elements that impact its quality and use. The goal is to combine trustworthiness, consistency, accuracy, and relevancy. When this is done properly, the outcomes are always better.

Human-centric experiences

Even with the right data foundation, however, enterprise AI does not become successful automatically. There is another dimension that is often underestimated: the human factor.

AI should not be positioned as a replacement for human decision-making. It should be understood as a system that augments it. A useful analogy is to think of AI as a junior colleague. It can process vast amounts of information, identify patterns, and make recommendations, but it still requires guidance, validation, and context. It learns from interaction, and it improves over time when it is embedded in real business processes.

This is where the concept often described as human-in-the-loop – or what we increasingly frame as cogniance – becomes critical. Trust is not established by accuracy metrics alone. It is built through continuous interaction between humans and AI systems. A proof-of-concept may demonstrate that a model works, but real value is only realized when business users trust the outcomes enough to act on them.

Building trust in the process

Proofs-of-concept provide an opportunity to experiment, but using AI for AI’s sake is a mistake. It is important to define a framework to make clear which use cases are the most appropriate for human-centric and AI-led processes. For example, if a KPI metric query asks about the revenue in a particular region, it is probably not a use case for AI. That is just standard reporting, so BI can handle the question. But if you ask for all the high-risk suppliers in that same region, then AI is a better choice because multi-step reasoning is required to answer the question.

Even with business-ready data and solid use cases, stakeholder buy-in and change management are still key factors. As enterprise AI becomes more complicated and harder to explain, teams need to build trust as a driver for adoption. Because even if you build a successful use case and prove value, it is not always good enough just to be right. Building trust and embedding it in the culture is part of the process.

Companies can help turn experiments into repeatable impact by leveraging SAP Advanced Data Migration and Management by Syniti, the SAP Business Data Cloud solution, and scalable architectures using the SAP Databricks capability and the SAP Snowflake solution. Take the opportunity to truly transform with the move to SAP S/4HANA.