An empirical study is described using data collected during 50 architecture audits of large industrial telecommunications systems performed over a period of two years. The goal of this study was to develop metrics that could be used to differentiate between projects that are at high risk of failure and those at low risk. These metrics would be computed following an architecture audit before coding has begun, allowing measures to be taken to correct identified problems very early in the development lifecycle. Detailed information is presented about the use of the proposed risk prediction metric for seven of the systems for which we provided architecture audits.