The first bridge: turning an idea into a company
For an academic start-up, the first pre-foundation financing question is not how to maximise the size of an investment. It is whether the project can cross the threshold from a promising research-based idea to a formally established company at all. The German EXIST–Business Startup Grant (EGS) provides unusually useful evidence for that question: it compares projects near a funding boundary, allowing a more credible estimate of what targeted support changes before private finance is readily available.
What the threshold comparison can tell us
Many promising projects never become businesses. A research team can have a valuable discovery and still stop before incorporation because the founders need income, the prototype needs expensive materials, or the commercial proposition is not yet clear enough to justify personal financial risk. If we simply compare companies that received a grant with companies that did not, however, we cannot tell whether the grant caused the difference. Public programmes usually select projects that appear more promising, so the funded group may have formed companies even without the intervention.
The missing quantity is the counterfactual: what would have happened to the funded project if it had not received EGS support? We cannot observe both versions of the same project. A credible evaluation therefore needs a comparison group that resembles the funded projects in all the ways that matter for formation, apart from receiving the grant.
The EGS selection process creates a useful approximation. Applications receive expert scores based on qualities such as the team, innovation potential, and marketability. Funding is recommended at a defined score threshold. Projects just above and just below that boundary are not identical, but they are plausibly close in their expected quality and motivation. A project barely above the threshold receives the intervention; a project barely below it does not. If the two groups differ sharply in later company formation right at the boundary, the discontinuity is evidence that the intervention—not merely the underlying promise of the projects—contributed to the result.
A simple hypothetical makes the logic clearer. Suppose two project applications receive almost the same assessment of their team, innovation, and market opportunity. One is just above the funding line and one is just below it. A comparison between those two projects is not perfect: the scores contain judgment, and the projects may differ in ways the evaluators did not record. But it is more informative than comparing the funded project with an average rejected application, because the latter may be less technically mature, less motivated, or pursuing a different kind of venture. The closer the cases are to the threshold, the more plausible it is that the funding decision—not a large pre-existing difference—accounts for a jump in formation outcomes.
This reasoning depends on a continuity assumption. In the absence of funding, the chance of company formation should change smoothly as the expert score rises. If projects just below and just above the line would otherwise have similar formation probabilities, a sudden jump at the line can be attributed to the intervention with more confidence. The design would be less persuasive if another policy changed at exactly the same score, if applicants could precisely manipulate their scores, or if the projects on either side were systematically different. The study tests the design with alternative bandwidths and controls, but the estimate remains local rather than universal.
This is the logic of a regression-discontinuity design. The score is the running variable, the funding threshold determines treatment, and company formation is the outcome. The design does not claim that every funded project is comparable with every rejected project. It estimates a local effect for projects near the decision boundary. That limitation is also its strength: by narrowing the comparison to close cases, it reduces the selection problem that makes a broad funded-versus-rejected comparison misleading.
The expert score is not a measure of future revenue. It combines judgments about the team, innovation potential, and marketability, so it reflects the programme’s wider eligibility and support purpose rather than one precisely measured prediction of incorporation. That is why controlling for the score is central to the comparison, but also why the score cannot be treated as an objective measure of venture quality. The authors note that it contains subjective elements and that its relationship with formation is not especially strong. A selection score can be useful for creating a local comparison without being a perfect forecast of which project will become a company.
The Springer study examined 452 funded projects and 332 projects rejected for EGS funding. The projects began, or planned to begin, their funding between January 2021 and June 2023, while company formation was checked in the Commercial Register through September 2024. The researchers also included the expert scores and other available characteristics in their models, then tested different bandwidths—the ranges of cases included above and below the threshold.
The result is large and stable across those specifications. The estimated odds ratios for company formation range from 10.79 in the narrow bandwidth to 7.23 and 5.37 in wider local bandwidths, with a global estimate of 4.83 using the full sample. An odds ratio is not the same as a percentage-point increase, so it should not be read as “a 10.79 percentage-point effect.” In absolute terms, the estimated differences in adjusted success probabilities range from 39.71 to 53.29 percentage points. In every specification, the estimated success probability for funded projects is at least twice that of the comparison group.
The decline in the odds ratio as the bandwidth widens is itself informative. The narrow estimate compares projects closest to the decision boundary, while the global estimate includes cases farther away, where unmeasured differences may be larger. The fact that the effect remains statistically significant across these choices supports the existence of a substantial programme relationship. The different magnitudes also counsel against selecting the most dramatic number and presenting it as the ordinary effect for every applicant. The most defensible statement is that the EGS effect is robust near the threshold and sensitive in size to how broadly the comparison is drawn.
The cautious conclusion is nevertheless powerful: for projects near this German programme’s funding cutoff, receiving EGS support substantially increases the likelihood of formal company formation. This is stronger evidence than a correlation between grants and company creation because the evaluation uses a design intended to approximate a local experiment. It is evidence about a particular intervention, population, period, threshold, and outcome—not a universal multiplier for all grants.
The sequence makes two boundaries visible. First, the comparison is between projects close to a selection threshold, not a random sample of all academic ideas. Second, the measured endpoint is a qualifying company entry, not a complete account of what happened inside the project. The study tells us that the intervention changes the probability of reaching that endpoint. It does not, by itself, tell us whether every element of the support bundle mattered in the same way.
That distinction also clarifies what “causal” means here. It does not mean that the grant guarantees incorporation for any team that receives it. It means that, under the design’s assumptions, moving a project from just below to just above the EGS funding threshold changes its expected probability of forming a qualifying company. The estimate belongs to that margin. A project far below the threshold, a project applying in another institutional setting, or a project receiving a differently designed grant may respond differently.
How the support bundle changes the pre-foundation state
The EGS is designed as more than a one-time cash transfer. It provides up to one year of support for students, graduates, and postdoctoral researchers working on technology-driven or knowledge-based projects. The programme can cover living expenses for up to three recipients at up to €3,000 per person per month, a separate €30,000 budget for project materials, and €5,000 for coaching. Universities and research institutions host the teams, help prepare applications, administer the funding, and may provide facilities, infrastructure, mentoring, and consulting.
Each element addresses a different obstacle between an idea and a company. Living support gives founders time that might otherwise be divided between the project and paid employment. That time is economically important because the founders carry the cost of uncertainty: they may work for months before the venture can generate revenue, and they may not know whether the technical route or market will hold. By absorbing some of the opportunity cost, the grant can make continued experimentation less personally hazardous.
The materials budget addresses a second constraint. A scientific idea cannot become a commercial proposition through argument alone; it needs a prototype, a tested process, or evidence that a product can work outside the original research setting. Materials do not guarantee that proof, but they can remove a practical reason for delay. For a laboratory or engineering project, a small amount of unavailable equipment or fabrication may block the next experiment entirely. Funding that allows the experiment to proceed can change what the team is able to demonstrate to customers, partners, or later investors.
There is also a less visible value in having a funded period with a defined purpose. Before support, the founders may face a choice between preserving personal income and spending concentrated time on a project whose commercial value is uncertain. The grant gives the team a temporary operating window in which technical and commercial assumptions can be tested together. That window does not eliminate risk; it moves the decision from “Can we afford to explore this?” toward “What evidence will tell us whether this deserves a company?” A well-used grant can therefore support disciplined abandonment as well as incorporation. If testing shows that the idea is not viable, avoiding a poorly founded company is a different outcome from proving that the grant failed.
Coaching and institutional support address a different interface. Researchers may know how to investigate a technical problem without knowing how to structure a business plan, define a customer, negotiate a pilot, or present financial needs to an investor. A university host can supply access to facilities and people who understand the research environment. Coaching can help the team convert technical progress into decisions about market entry, roles, milestones, and resource needs. Networking can expose the project to potential users, established companies, and investors whose questions reveal weaknesses that laboratory feedback may not expose.
These are plausible links in a theory of change:
- Financial and temporal runway reduces the risk that founders abandon the project before testing it.
- Project resources allow technical development and prototype work to advance.
- Coaching and education improve planning and commercial interpretation.
- Mentors, customers, investors, and university networks add knowledge and relationships the founding team may lack.
- Better technical and commercial preparation raises the feasibility of formal company formation.
The chain is useful for understanding why a grant could have a large effect before the company exists. It should not be mistaken for a set of independently estimated effects. The EGS study explicitly evaluates whether funded projects are more likely to form companies; it does not empirically verify that living support, materials, coaching, or networking is the mediator responsible for the observed difference. Earlier studies cited by the paper report improvements during EGS funding in product development, business planning, networking, skills, and external capital, but those findings do not convert the present regression-discontinuity estimate into a component-level causal analysis.
Nor should institutional anchoring be confused with institutional control over the venture’s future. The university or research institution can host a team and make facilities or advice available, but it cannot guarantee a customer, a regulatory outcome, a private investor, or a profitable exit. The founders still have to make choices about what to build and which assumptions to test. The grant administrator can allocate support under programme rules, but cannot make the market accept the product. These distinct responsibilities explain why a bundle can improve formation while leaving major post-formation risks intact.
Those risks appear as the project changes state. If the grant produces company formation by giving founders time, the next constraint may be investor access. If materials produce a prototype, the next constraint may be regulatory approval or a paying pilot customer. If coaching improves a business plan, the plan still has to survive contact with the market. A support bundle can move a project from one state to another while exposing a new bottleneck that was previously hidden. This is not evidence that the first intervention was ineffective; it is a reason to measure the next transition separately.
The outcome definition also imposes a boundary. The study counts limited-liability companies recorded in the German Commercial Register, mainly GmbH and UG forms. Partnerships such as a GbR were not included because the relevant registration data were unavailable. The authors report a formation rate slightly above 74% among funded projects in this sample, while an ongoing monitoring report gives a rate near 85%; the difference is explained partly by the monitoring report’s inclusion of partnerships and partly by its allowing funded projects more time after support ends.
Nor does formal formation tell us how quickly the company was founded, whether it survived, whether its product was commercially valuable, or whether it later raised private equity. The EGS study itself calls for longer-term analysis of speed, quality, sustainability, innovation output, economic development, and survival. The recent observation window is especially important: some projects had only recently completed funding when the register search was conducted, so a later formation could not yet be observed. A binary company-entry measure is appropriate for the programme’s first objective, but incomplete for judging the whole financing path.
The study also notes methodological limits around the score and the threshold. Projects just below the funding line could sometimes revise their applications, creating fewer observations immediately below the cutoff. Reviewer judgments may contain bias or inconsistent interpretations, and additional covariates related to success may be missing. Sensitivity analyses, alternative bandwidths, weighting, and balance checks support the stability of the reported effect, but the authors still identify a randomized trial as the cleanest possible design. Longer follow-up and data on mechanisms would be needed to establish whether the same formation effect leads to durable companies in other regions and institutions.
The most defensible reading is therefore two-part. First, a targeted, university-integrated grant can causally increase the chance that a near-threshold academic project becomes a formal company. Second, the reason it works—and whether the new company can finance its next stage—requires more than the formation result. The programme creates runway, technical progress, and potentially commercial capability, but the later path depends on what the team and its support network build before that runway ends.
This is the first bridge in the wider financing story. It gets the venture across the boundary into existence. The next question is what kind of capital should come after that bridge, because grants, assistance, equity, convertibles, and debt do not leave the same options behind them.