The role of a relatively small cadre of high-tech startup firms in driving innovation and economic growth has been well known and amply celebrated in recent history. At the same time, it is well recognized that, while the overall contribution of startups is crucial, the high-risk and high-reward strategy followed by these startups leads to significant failure rates and a low ratio of successful startups. So, it is curious to notice that literature tends to focus on successful startups and on quantitative studies looking for determinants of success while neglecting the numerous lessons that can be drawn by examining the stories of startups that failed. This paper aims to fill this gap and to contribute to the literature by providing a repeatable and scalable methodology that can be applied to databases of unstructured post-mortem documents deriving startup failure patterns. A further and related contribution is the analysis carried out with this methodology to a large database of 214 startup post-mortem reports. Descriptive statistics show how the lack of a structured Business Development strategy emerges as a key determinant of startup failure in the majority of cases.based on discriminant analysis [5] and multiple discriminant analysis [4], followed by more recent approaches exploiting regression [6][7][8]. Since the 80s, artificial intelligence methods started to be used as well to predict ventures success/failure. Suggested solutions relied on decision trees algorithms [9], artificial neural networks [10], clustering [11] and hybrid genetic algorithms [12]. Approaches based on financial data had the advantage of being potentially applied to a high number of companies, since data could be gathered from their annual reports. Nonetheless, company revenues were frequently consequences of other aspects, such as entrepreneur's ability, company's core competencies, market, etc. In this view, other research works investigated whether such aspects could contribute as well to the success or failure of a venture. For instance, analysis conducted on entrepreneurs examined the influence of their gender and ethnic origin on the likelihood to succeed or fail [13,14]. The work in [15] conducted a logistic regression analysis based on 15 independent variables success versus failure prediction model in Israel, including the management experience, the education, and the age of the owner. The logistic regression analysis was also adopted in [16] to model the relationship between small business mortality rates and the aggregate levels of internal and external risks (e.g., bankruptcy related to interest rates, discontinuance of business or of ownership, etc.). Other researchers focused on entrepreneurial attitudes and linked startup failure to dissonances between corporate goals and the goals of its founders [17]. Others presented failure as the result of entrepreneurs' overconfidence and hubris [18]. In contrast, other researchers argued that, without a reasonable level of positive perception of one's abilities, several successful...
