The term “deep-tech” has been used to refer to technologies related to 4th Industrial Revolution/4.0 Industry such as artificial intelligence (AI), big data, drones, quantum computing and robotics, among others. For example, digital-enabled unicorns are based on consumer-driven business models where new technologies are not critical to their success (Urbinati, Chiaroni, Chiesa, & Frattini, 2019) and receive much more attention from researchers and funding organizations (Aldrich & Ruef, 2018). Digital unicorns are feasible by a digital architecture, normally based on pre-established technologies (de Massis, Frattini, & Quillico, 2016). In contrast, businesses based on deep technologies did not receive attention from funding programs, venture capitalists and policymakers until recently (Different Funds, 2020; Gigler, 2018).

The term was introduced in 2015 by Swati Chaturvedi, CEO of venture capital company Propel(x). Chaturvedi (2015, p. 1) defines deep tech as “companies founded on a scientific discovery or meaningful engineering innovation.” Chaturvedi proposes a distinction of deeptech companies from digital-enabled unicorns, considering the role of technology and competitive advantage. Currently, most digital-enabled unicorns are innovative business models based on existing or pre-existing technologies. In contrast, the deep-tech companies’ business model creates value by proposing a technological solution to existing problems. As they are based on scientific-technological discovery, the business models of deep-tech companies are more difficult to copy (Chaturvedi, 2015).

In this sense, instead of business model innovation, deep-tech startups use deep technologies (e.g. AI, Big Data, robotics, etc.) as a source of competitive advantage. Thus, many deep-tech startups are spin-offs or collaborators in facilities and research infrastructure (Scarra & Piccaluga, 2020). However, as deep technology is difficult for many investors to understand, these entrepreneurs must find early supporters to secure funding for their innovative projects (Fisher, Kotha, & Lahiri, 2016; Vossen & Ihl, 2020).

In contrast to digital startups, deep technology requires complex integration between software and hardware (Siegel & Krishnan, 2020). This means that deep-tech startups can deliver unique innovative solutions, but also that finding compatible existing technology architectures is difficult (Adner & Kapoor, 2010; Thomas, Autio, & Gann, 2014).

Unlike many digital startups that use the lean approach, fast and iterative development cycles to improve their products according to consumer requirements, deep-tech startups require long/slow and sequential development cycles (Dealroom, 2021). Furthermore, unlike digital technologies that provide direct solutions to the market, deep technologies represent basic and intermediate components, i.e. enabling technologies that feed the creation of application or facilitate the delivery of solutions to end-users (Bresnahan & Trajtenberg, 1995). Thus, entrepreneurs’ role is to identify the uses of creating deep technologies for endusers (Garud, Gehman, & Giuliani, 2018). These characteristics lead to the assumption that these companies are associated with high risk and uncertainty.

Promoting entrepreneurship is associated with public policies and context (Autio et al., 2014). Entrepreneurship generates added value in the form of economic growth and jobs (Haltiwanger, Jarmin, & Miranda, 2013; Ordeñana, Vera-Gilces, Zambrano-Vera, & Amaya, 2019). Therefore, governments, whether by subsidies, creating a favorable regulatory framework, or implementing supportive policies, often encourage entrepreneurial activity (Autio & Rannikko, 2016).

At last, the country’s business environment, i.e. the costs, requirements and procedures to start a business, can represent an obstacle to business creation and a discouragement (Chowdhury, Audretsch, & Belitski, 2019; Dutta, Sobel, & Roy, 2013). However, excessively reduced costs and procedures can increase the number of noninnovative entrepreneurs (Bailey & Thomas, 2017). In this sense, establishing a regulatory framework that does not discourage innovators without encouraging noninnovative entrepreneurs’ entry is necessary (Darnihamedani, Block, Hessels, & Simonyan, 2018).

Human resources represent the knowledge, competencies and skills acquired by individuals (Schultz, 1961). Entrepreneurs who have received formal education, especially tertiary education, are more likely to create innovative ventures (Michelacci & Schivardi, 2020). Thus, education in science, technology, engineering and mathematics (STEM) facilitates the adoption of deep technologies (Delera, Pietrobelli, Calza, & Lavopa, 2022) and, consequently, entrepreneurship (Colombo & Piva, 2020). The STEM education is not restricted to the tertiary level, some countries have overhauled secondary education systems by implementing STEM education models (Higde & Aktam ıs, 2022; Kutnick, Lee, Chan, & Chan, 2020).

Knowledge is also generated in knowledge-intensive business (KIBS) and therefore incorporated by knowledge-intensive workers. The KIBS can provide solutions and support services for early-stage entrepreneurs, acting as an innovative entrepreneurship driver (Badulescu, Badulescu, Sipos-Gug, Herte, & Gavrilut, 2020).

Also, RD are essential for knowledge creation. The knowledge spillover theory of entrepreneurship assumes that knowledge generated by RD activities by universities and incumbent companies can create entrepreneurial opportunities (Tavassoli, Obschonka, & Audretsch, 2021). Entrepreneurs can interact with universities, research institutes and RD companies, using the research infrastructure to develop innovations. These interactions can represent a driving factor for developing innovative ventures (Malerba & McKelvey, 2020). Therefore, knowledge flows and the ability of individuals/entrepreneurs to absorb these flows and transform them into innovations is fundamental for creating innovative ventures (Ganotakis, D’Angelo, & Konara, 2021).

Physical infrastructure is fundamental for connecting the economic agents and, therefore, crucial for entrepreneurial activity (Audretsch, Heger, & Veith, 2015). Digital infrastructure, i.e. information and communication technologies (ICTs), also enables digitalization, promotes the growth of the digital economy and generates entrepreneurial opportunities (Ganotakis et al., 2021; Jafari-Sadeghi, Garcia-Perez, Candelo, & Couturier, 2021). Finally, physical infrastructure also is important for the environment and the need to implement sustainable corporate practices, as well as the creation of sustainability startups (Tiba, van Rijnsoever, & Hekkert, 2021).

Access to credit is one of the major obstacles to the venture creation, as most early-stage entrepreneurs deal with a lack of financial resources to make their respective businesses viable (Dutta & Meierrieks, 2021). Studies indicate that financing is fundamental for entrepreneurship, especially for innovative ventures, and the lack of investment funds is one of the main barriers to the EEs’ improvement (Ács et al., 2017; Spigel & Vinodrai, 2020). Besides, the demand is essential for entrepreneurs, as selling novel goods will only be viable if the population has the material conditions to acquire them (Leendertse et al., 2021). Researchers show that growing markets increase the firms’ entry (Eckhardt & Shane, 2003; Sato, Tabuchi, & Yamamoto, 2012). Entrepreneurs often operate in large markets far from their headquarters; thus, easy access to potential regional markets is critical for startups.

The NCA is an approach introduced by Dul (2016a, 2016b) that provides information about the necessity of an input for a certain desirable output. A necessary condition is a key driver of an output: without a certain condition, the output will not be achieved. This concept imposes only a necessary condition, thus differing from fuzzy-set qualitative comparative analysis (fsQCA) (Ragin, 1987; Ragin, 1989; Rihoux & Ragin, 2009), which deals with sufficient conditions [2]. For example, without certain input variables (e.g. RD) reaching an output variable will not be possible (e.g. a filled/granted patent), and this cannot be compensated by other critical factors. However, the presence of a necessary condition does not guarantee the outcome (this is the case of a sufficient condition). Figure 1 shows the main concepts and rationale of a necessary condition.

The NCA assumes that an X (input) condition constrains a Y (output) result by tracing a line on top of a set of values plotted on a scatter plot X vs Y graph (Figure 1). This ceiling line can be produced by two methods: ceiling regression-free disposal hull (CR-FDH) line

Figure 1.
Exemplification of NCA rational
Source: Elaborated by the authors based on Richter et al, (2020, p.2,246

(orange); ceiling envelopment-free disposal hull (CE-FDH) line (red). The area above this ceiling line – top left corner on Figure 1 – is named empty space, which suggests that high output levels (Y) cannot be achieved by low input levels (X). Computing the proportion of empty space size in relation to the total space (TS) gives us the effect size (d) statistic, a measure of necessity. Therefore, the greater the ES, the greater the restriction imposed by X to Y (Dul, 2020). To trace the ceiling line, we selected the CE-FDH, a method recommended when the sample is composed of a limited number of outputs. Figure 1 shows that the benefit of using this technique (CE-FDH) is its 100% accuracy, that is, no observations are within the empty space.

Additionally, we choose to apply NCA to investigate necessity conditions, as important theoretical and empirical evidence shows that NCA leads to better and robust results than fsQCA, concerning necessity analysis. In short, the evidence claims that:

NCA can make a statement “in degree,” whereas fsQCA can make only “in kind” statements.

fsQCA uses Boolean logic to set the pertinence of a condition to a given result, and doing that, the resultant analysis is sensitive to some extent to the calibration procedures (logistic or standardized algorithm) and chosen threshold parameters (raw or proportional reduction in inconsistency (PRI) make the results sensitive).

fsQCa can produce more false negative/positive (Baumgartner, 2015, 2022; Dul, 2016a, 2016b; Patala et al., 2021; Vis & Dul, 2018).

Finally, the NCA also displays a result named bottleneck (right side of Figure 1). The researcher can choose between percentage or original values of variables and then inform the exact value that X bounds Y (yellow lines). In this illustration, since the scale is from 0 to 10, we can easily interpret interchangeably that, if we intent obtain an output of Y = 7 (70%), we need at least an input of X = 5.5 (55%); otherwise, the output will be impossible. Obviously, if this condition was satisfied it still configures a nonsufficient condition to the output (Y) happening, but it is certain that, if not a certain amount of X, then not at all a certain amount of Y. This is a strong appealing to policy marking analysis and formulation as the bottleneck results provide an “in kind” qualitative evidence (that X is necessary for Y) and, most important, an “in degree” quantitative extrapolation states that at least a minimum amount of X is a necessary condition to make achieving a given level Y of output possible.

The IESE Business School of Navarra points out startups based on advanced materials, artificial intelligence, biotechnology, blockchain, drones and robotics and quantum computing as the main sectors of deep-tech ventures (Siota & Prats, 2021). In addition to these sectors, Start-up Genome (2020) also includes AgTech and Big Data business as a category of deep technologies.

To collect data on deep-tech entrepreneurship we used the Crunchbase database. Based on IESE business school (Siota & Prats, 2021) and Start-up Genome (2020) studies on deep technology sectors, we collected data directly from the search engine of Crunchbase website (see www.crunchbase.com/discover/organization.companies) and the sectors (they used the word industry) taxonomy used by them is as follows:

Advanced materials, AgTech, Artificial intelligence (artificial intelligence, intelligent systems, machine learning, natural language processing, and predictive analytics), Big Data, Biotechnology (bioinformatics, biometrics, biopharma, biotechnology, genetics, life science, neuroscience, and quantified self), Blockchain, Drones and robotics (drone management, drones, and robotics), and Quantum computing (Crunchbase, 2021, website).

We have defined the five-year period to ensure that we only select companies that are similar in terms of their growth stage. The period delimitation is also relevant for allowing us to select only innovative start-ups.

To test whether a resource provided by an EE is a necessary condition for deep-tech entrepreneurship, we collected data from the Global Innovation Index (GII). We selected 15 variables (Table 1) that represent innovation inputs provided by EE at country-level (Cornell University, INSEAD, & WIPO, 2020). Using GII indicators is also interesting, since it is a longitudinal and systematic study of the factors that affect innovation in different countries. It is an internationally harmonized and comparable data source that focuses on both innovation inputs and outputs. Our sample corresponds to data from 132 countries participating in the GII 2021 report [3].

To run the NCA we normalized the data from 0 to 1, although it is not needed for NCA purposes. We applied the max-mix method ([Max value observed]/[Max Min]). Outliers’ values were identified and replaced by the interquartile interval method, an approach used in EEs composite indices such as the Global Entrepreneurship Index (Ács et al., 2019) and the European Innovation Scoreboard 2022 (Hollanders et al., 2022). Appendix 1 brings the descriptive statistics for all variables in raw (original) data as well normalized.

Figure 2 allows a visual inspection of the eight scatter plots whose conditions have proven to be necessary for deep-tech entrepreneurship [4]. The scatter plots are the graphic solution of NCA necessary analysis results. Usually, the abscissa axis (X) represents the conditions variable (“condition” and not “independent” according to developers’ syntax), i.e. variables that measure the EEs’ conditions, whereas the ordinate axis (Y) represents the deep-tech startup variable, the outcome. The countries (observations) are represented by the blue dots.

The effect size (d), a criterion to infer if a condition is or isn’t a necessary one, is computed by the ratio between the “empty space” (upper left corner above the ceiling line, the red line in Figure 2) and the “total space.” Thus, the larger the empty space, the more an EEs condition

Table 1
Entrepreneurial ecosystesms variables and definitions

Sources: Based on Global Innovation Index – GII (Cornell University, INSEAD & WIPO 2020) and Crunchbase, Elaborated by the authorsNotes: GII is available at: www.globalinnovationindex.org/Home; Crunchbase is available at: www.crunchbase.com/

Figure 2
Scatter plot between each condition (X) vs outcome (Y)
Source: Elaborated by the authors

constrains the outcome. Additionally, to the visual inspection, the specialized literature recommends using two criteria to consider a condition as necessary: effect size (d) and a p-value [5], although the threshold is up to the researcher’s judgment. We selected the necessary conditions that meet at least medium effect size (d) and a p-value statistically significant at least 5% level, shown in Table 2. The shaded cells are the ultimate necessary conditions.

Thus, the political, regulatory and business environment (PE, RE and BE) are pointed out by the EE literature as moderating (in NCA terminology, an allegation assumption about sufficiency) factors of entrepreneurship (Sendra-Pons, Comeig, & Mas-Tur, 2022). Our results of the multivariate NCA indicate that the two institutional conditions (PE and BE) are necessary conditions for deep-tech entrepreneurship.

Our results agree with the study of Torres and Godinho (2021), which analyzed data from 27 EU member states, with the former member UK, and applied fsQCA and NCA to discover single necessary conditions (NCA was able, whereas fsQCA was not), finding that “Formal

Figure 2.
Deep-tech entrepreneurship
Source: Elaborated by the authors

institutions, regulations and taxation” (Table 3 at p. 38) are necessary conditions for what the authors called digitally enabled unicorns and new business creation. Regarding the regulatory environment (RE), the results showed a nonsignificant value; therefore, it cannot be a necessary condition for deep-tech entrepreneurship.

Regarding the dimension entitled “Human capital and research,” two conditions also attended to our criteria threshold, showing a large (Education – E) and medium (Research and development – RD) effect size and being both statistically significant. Tertiary education (TE), despite showing a small effect size, had a nonsignificant p-value. Therefore, it is not a necessary condition for deep-tech entrepreneurship.

Not surprisingly, the RD variable – which measures the number of full-time researchers, average expenditures of the three largest companies in RD, and quality of universities and research institutes – showed both effect size and significant p-value. Studies on EE (JafariSadeghi et al., 2021; Tavassoli et al., 2021) show that researchers influence entrepreneurship, as well as RD expenditures and the quality of universities. Thus, our results agree with the EE literature.

Table 2
Results of multivariate necessary condition analysis and permutation test (p-value

Source: Elaborated by authorsNotes: ns = not significant, Effect size (d): small = (0 < d < 0.1); medium = (0.1# d < 0.3); large = (0.3# d < 0.5); very large = (d 0.5), p-value: ** = significant at 5% (p < 0.05); *** = significant at 1% (p < 0.01); ns = nonsignificant. Shaded cells are the ultimate necessary conditions to be used in the next step

Table 2
Bottleneck analysis

Source: Elaborated by the authorsNote: NN = not necessary

The next dimension – infrastructure – showed all conditions with effect sizes greater than zero. Although ecological sustainability (ES) and ICT are pointed out in the literature as enabling factors and a spaces of opportunity for creating new ventures (Tavassoli et al., 2021; Tiba et al., 2021), our results showed a nonsignificance statistical p-value which discarded them. Consequently, only general infrastructure (GI) is a necessary condition for deep-tech entrepreneurship. This condition also finds support in the literature on EE (JafariSadeghi et al., 2021; Audretsch, Heger, & Veith, 2015) as a condition that facilitates access to markets.

Regarding the market sophistication dimension, the investment (I) showed a nonsignificant p-value, even though the literature indicates that it (Gigler, 2018; Spigel & Vinodrai, 2020), mainly in the form of venture capital, is important for entrepreneurship. On the other hand, credit (C) can be considered necessary for deep-tech entrepreneurship. Access to credit, particularly in the form of finance and loans – as measured by GII – for early stages deep-tech startups sounds plausible instead of investments series A, B and C rounds – as measured by GII – for a more incumbent startup (Gompers et al., 2020).

Our results indicate the remaining market sophistication condition, named trade, diversification and market size (TDMS) is a necessary condition for deep-tech startups. Indeed, market size is a crucial factor with our results agreeing with previous studies (Ali, Kelley, & Levie, 2020; Tavassoli et al., 2021).

Finally, the last dimension – business sophistication – showed that only knowledge absorption (KA) had both significant effect size and p-value over the threshold. Although knowledge workers (KW) and innovation linkages (IL) have effect sizes greater than zero, their p-values are slightly higher than 5% (0.68 and 0.54, respectively). Knowledge absorption (KA) finds support both in seminal (Kim, 1997) and more recent literature, such as the study of Khan and Tao (2022) that used the same data source as we did – the GII – and encountered evidence of positive correlation between knowledge absorption and innovation performance of manufacture firms.

After the NCA’s necessity analysis, we performed the bottleneck analysis. Here lies the most important power advantage of this technique, which, to the best of our knowledge, no other technique provides: it “[...] precisely identifies what level of X is necessary for what level of Y” (Vis & Dul, 2018, p. 882) or “[...] shows which level of the condition is a bottleneck for a given desired level of the outcome” (Dul, 2020, p. 1518). In Table 3, the first column shows the possible and/or wished levels of outcome (Y = DTS = deep-tech startup) on a scale from 0 to 100% and the remaining columns (from 2 to 9) show the levels of necessity for each condition to obtain a certain desired level of output.

Our results show that even to low DTS levels (Y = 10%) all eight conditions are necessary. However, the level of necessity varies for each condition. To focus our analysis, Figure 4 shows the GII 2021 results of Brazil. On the left side is a radar graph with a benchmarking of Brazil (green line) against: all upper-middle (34) income countries (orange line); all Latin America & Caribbean (18) countries (blue line); and top 10 best performance countries (yellow line).

From a key performance indicators (KPI) evaluation Brazil ranks 11th among the 34 upper middle income and 4th among 18 Latinamerican & Caribbean economies. These results are not surprising since our national system of innovation and entrepreneurship ecosystem is one of the most mature in the region (Alves et al., 2021; Dionisio et al., 2021; Fischer et al., 2022). However, it can improve as shows the radar graph in the left side of Figure 4. Brazil has the strongest positions only in four conditions compared with its counterpart region (E, RD, TDMS and KA) and income group economies (RE, E, RD, TDMS and KA). Obviously, compared with the top 10 performing economies in the GII, Brazil has a long way to go.

From an NCA that aims to support policy, prioritizing the conditions where Brazil has its weakest performance on GII (shaded in gray at the right side of Figure 3), i.e. general infrastructure (GI: 20.5), credit (C: 30.5), and research and development (RD: 31.9), is advisable. This focus is plausible and desirable since the government, specially the Federal Government, struggles due to the expenditure cap (from the Portuguese, “teto de gastos”) on the public budget.

Figure3.
Brazilian conditions and outcome
Sources: Based on Global Innovation index - GII (Cornell University, INSEAD & WIPO 2020) and Crunchabase; Elaborated by the authors

Table 3 shows that these conditions needed to be at least 27.7%, 34.4% and 2.2%, respectively, for Brazil could be able to reach the outcome of high deep-tech startups (Y 80%), measured here in numbers of firms. Thus, as Table 3 is expressed in percentage, we get the minimum necessary scores of 18.6, 30.3 and 2.0 for GI, C and RD, respectively. This is obtained by multiplying each requirement level from Table 3 by the maximum value in the sample (it is the value normalized as 1, for instance: GI = 27.7% 67.3 = 18.6; C = 34.4% 88.0 = 30.3; RD = 2.2% 89.8 = 2.0.

Brazil is very close to this threshold in GI (20.5 against the minimum of 18.6) and C (30.5 against the minimum of 30.3). Thus, attention is needed to improve to overcome the fragilities of these conditions. Therefore, for the Brazilian EE to have the possibility of originating deeptech ventures, allocating efforts – in terms of strategic national programs and goals – to overcome the fragilities of these two necessary conditions (GI and C) is necessary.

Note again the concept of necessary conditions, i.e. “If these conditions are not in place (at the right level), the outcome will not occur.” and “Other conditions cannot compensate for their absence” (Dul, 2016a, 2016b, p. 1522).

Therefore, it does not matter how many and whatever the mix of configurations that the policymakers want to implement, the necessary conditions must be present in these minimum requirements to assure the possibility of realization of the outcome. However, although they are necessary and cannot be replaced by other necessary conditions, the presence of these conditions does not guarantee that the output will occur. In short, if these conditions are not present, the EE capacity to generate deep-tech entrepreneurship is guaranteed to fail.