With the rapid development of information technology, enterprises are facing increasing decision-making pressure in a complex and changing market environment. The traditional decision-making information analysis system can no longer meet the needs of modern enterprises for accurate analysis and rapid response. To address this problem, this paper constructs a data-driven enterprise decision-making information analysis system and proposes an optimization path. First, the Apache Spark platform is used for big data integration to comprehensively collect and integrate multi-source heterogeneous data from inside and outside the enterprise. Then, the Transformer model is used to preprocess the data and extract features to enhance the depth and breadth of data analysis. Finally, the BERT (Bidirectional Encoder Representations from Transformers) model in natural language processing is combined to achieve effective analysis and mining of unstructured data, thereby extracting valuable decision support information. In the decision accuracy evaluation, the MSE (Mean Squared Error) of the data-driven system was 0.014 and the R2 value was 0.91, showing strong predictive ability. In terms of response speed, the response time of the data-driven system was 2.40 s for 10,000 data items, which was more adaptable than the response time of the decision tree and logistic regression models. In terms of system stability, the availability of the data-driven system was 0.82 when the noise ratio was 30%, which was higher than the traditional model in terms of stability and availability. The experimental results verify the effectiveness and feasibility of the system in complex environments, providing strong support for the intelligentization of enterprise decision-making.