© 2020 A reliable benchmarking system is crucial for the contractors to evaluate the profitability performance of project tenders. Existing benchmarks are ineffective in the tender evaluation task for three reasons. Firstly, these benchmarks are mostly based on the profit margins as the only key performance indicator (KPI) while there are other KPIs fit to drive the evaluation process. Secondly, these benchmarks don't take project context into account, thereby restricts their predictive accuracy. And finally, these benchmarks are obtained from small subsets of data, making it hard to generalise. As a result, estimators cannot probe into tenders to judge the strengths and weaknesses of their bids. This advancement is critical for not only choosing more lucrative opportunities but also driving negotiations during the tendering process. This study aims to develop a benchmarking system for tender evaluation using Big Data of 1.2 terabytes, comprising 5.7 million cells. A holistic list of seventeen (17) KPIs is identified from the email data using Text Mining approaches. Besides, eight (8) key project attributes are chosen for ensuring context-aware benchmarking using Focused Group Interviews (FGIs). At the crux of this work lies the proposition of a deep ensemble learner based on the decomposition-integration methodology. In the decomposition stage, the model predicts several attribute-specific benchmarks for each KPI using our proposed context-aware algorithm. In the integration stage, deep neural network-based learners are trained to generate final project-sensitive KPI benchmark. The learner is deployed in the Spring tool to support the tender evaluation of power infrastructure projects. A tender of 60km underground cabling project is evaluated using the proposed learner. The system spontaneously identified KPIs in the tender that require further attention to achieve greater profitability performance.