(Stalled) Acceleration Goals

The hub and accelerator landscape has continued to expand in nominal terms while contracting in quality and sustainability.AfriLabs' 2024 Impact Report records a network of 514 innovation hubs in 53 African countries, directly reaching over 280,000 entrepreneurs and startups. The total number of hubs - including those outside AfriLabs - remains above 1,000, with approximately 109 accelerators and incubators in East Africa alone as of mid-2025.

Sendemo and Digital Africa's 2024 study of the African SSO landscape quantifies the structural problem beneath this empirical finding. The work points to roughly 1,200 organisations operating across the continent: 95 percent are independent businesses nominally expected to be financially self-sustaining; 30 percent are actually profitable; 70 percent are maintained alive by donor funding. The result is a structural double injunction - SSOs are simultaneously required to serve non-solvable beneficiaries and to generate autonomous revenues - which no business model yet designed can consistently resolve.

The correction period produced significant casualties. Many hubs and accelerator programmes closed between 2023 and 2024 as corporate sponsorship and donor funding tightened. Business models remained fragile: most hubs receive under $100,000 in external funding, which is largely consumed by staff and facilities. Revenue streams are still dominated by membership fees, project-based donor funding, and consulting - the same fragile triangle identified in the original analysis. The most dramatic case was 54 Collective - formerly Founders Factory Africa - which collapsed in 2025 following the Mastercard Foundation dispute. The collapse demonstrated the systemic risk of large-scale donor-backed institutions whose operating viability depends on a single funding relationship.

The most important new signal comes from funding data, not from accelerator reporting. Disrupt Africa's African Tech Startups Funding Report 2024 shows that the share of funded startups with any accelerator experience fell from 64.5 percent in 2023 to 51.3 percent in 2024.

This does not prove that accelerators cause worse outcomes, but it does show that a growing share of ventures able to raise capital are doing so without them. The WDI East Africa study adds field texture: entrepreneurs report attending multiple ESO programmes in succession - cycling through cohorts to access small grants rather than to build capability - a pattern WDI terms "training fatigue." The same study finds investors increasingly distrustful of ESO pipeline quality, with due diligence consistently revealing companies that programmes deemed investment-ready to be underprepared. Combined with founder narratives around time cost and limited value, the direction of travel is clear: founders and investors are beginning to self-correct on a model whose causal link to scaling has always been weak.

"Why is a startup going into an accelerator? They're not going to be accelerated. They're going because they want help, which leads them to funding or customers. These are the only reasons why they should be attending. And if it doesn't do that, then they've damaged their business by wasting time and energy." - interviewee

The academic literature on accelerator effectiveness has matured substantially since the 2022 analysis. Three separate evidence streams now converge on a consistent picture.

The foundational empirical literature. Yael Hochberg's 2016 Innovation Policy and the Economy paper provides the most cited treatment of the accelerator phenomenon in the academic literature, arguing that accelerators differ structurally from earlier incubator models in their fixed-duration, cohort-based, mentorship-intensive design and that the empirical record on whether the design works is more mixed than the field's own reporting suggests. Cohen, Bingham and Hallen's Administrative Science Quarterly paper (2019) extends the analysis through detailed case research on accelerator design choices, finding that the design features that produce learning - concentrated, time-bounded, peer-comparison-rich - are precisely the features that distinguish well-designed from generically-designed accelerators, and that most programmes do not implement them at the level of intensity the academic theory describes.

The treatment-vs-selection question. Gonzalez-Uribe and Leatherbee's Review of Financial Studies paper on Start-Up Chile (2018) is the methodological breakthrough that the live ecosystem debate has not absorbed. Using a regression discontinuity at the Start-Up Chile acceptance threshold to isolate the treatment effect from the selection effect, they find that the cash-grant component of acceleration produces no measurable performance improvement - but the schooling and mentorship component produces meaningful gains in performance, suggesting that the capability-building dimension of accelerators is what works, while the cash transfer dimension is what does not. The implication for African ecosystem policy is direct: the dominant programme architecture (cash grants delivered through generic cohort programming) is the opposite of the architecture the rigorous evidence supports (intensive capability-building delivered with minimal cash dependence). Yu's Strategic Entrepreneurship Journal paper (2020) extends the empirical record with a different finding: accelerator participation accelerates both founder learning and the timing of failure for ventures whose underlying model is non-viable. Accelerators help winners win faster and help losers fail faster - both of which are beneficial outcomes if the ecosystem's purpose is allocative efficiency, but neither of which is what donor-architecture programmes are typically calibrated to demonstrate.

The GALI ten-year synthesis. GALI's ten-year synthesis provides the most comprehensive cross-country evidence base. Accelerated ventures do outperform non-accelerated peers on revenue and investment on average, even after accounting for selection effects, and those differences increase over time. The majority of accelerators are cost-effective mechanisms for driving funds into promising ventures. This matters: the 2022 analysis was more sceptical than the evidence justifies, and the field should not overcorrect on the basis of a funding contraction that exposed design failures rather than disproving the model.

But GALI's nuance is equally important. Most of the upside accrues to the top quartile of ventures, and high-performing firms tend to enter accelerators with stronger teams and more resources - directly consistent with the Gonzalez-Uribe and Leatherbee selection-effect finding. Investment benefits concentrate in high-income country programmes; emerging market ventures see more benefit in revenue growth than capital raised - a finding that maps directly onto African operating conditions where investor-readiness pipelines remain thin. Top-performing enterprises that attribute no success to their accelerator cite a consistent cluster of failures: poor match between programme offerings and their needs, over-emphasis on investment over business fundamentals, lack of meaningful investor connections, and absence of cohort cohesion. These are design failures, not model failures.

If acceleration predominantly helps ventures already on a strong trajectory, its role as a market-failure tool is limited. It becomes a selection device for strong firms rather than a capability-building mechanism for the broader pipeline.

What capability-building actually looks like when it works is worth naming directly. The development-economics literature provides two foundational reference points. Bloom, Eifert, Mahajan, McKenzie and Roberts' QJE paper on Indian textile firms (2013) demonstrated that intensive, firm-specific management consulting raised productivity by 17 percent in the first year, with effects persisting and spreading across plants - but the intervention was sustained, expensive per firm, and tightly tailored. McKenzie, Iacovone and Maloney's April 2026 World Bank paper reports 8–10 year follow-up results from a randomised Colombian group-consulting experiment: firms working collaboratively in peer groups at $10,000 per firm were 11–13 percentage points more likely to survive through 2024, with annual sales up 55 percent, profits up 48 percent, and employment up 17 workers per firm. Group consulting dominated individual consulting on cost-effectiveness. The design features that produced sustained impact - peer benchmarking, collective problem diagnosis, practices embedded in internal systems - are the opposite of generic content delivered through short-cycle acceleration programmes, and the long-term follow-up was funded by the Argidius Foundation through IPA.

A parallel signal comes from Canada. Growth Catalyst, a scale-up programme launched at Mount Royal University's Institute for Innovation and Entrepreneurship, was designed explicitly from an evidence base - the BIG Ten Capabilities research from Kent Business School's 447-firm Promoting Sustainable Performance longitudinal programme, adapted for the Alberta SME context. It targets established firms with $2–100 million revenue rather than startups, runs a cohort-based six-month integrative model, and is being evaluated using the Goldman Sachs 10,000 Small Businesses methodology developed by Professor Mark Hart at Aston University. The design features - selective cohort, revenue-stage targeting, evidence-derived curriculum, quasi-experimental evaluation - are the same cluster that distinguishes Growth Catalyst from the acceleration model critiqued earlier.

Multiple comparable emerging and developed market cases now point in the same direction: structured, evidence-based, cohort-delivered scaleup programming with rigorous evaluation works. What is absent across African markets is not the design template. It is the institutional architecture and evaluative discipline to build, fund, and test equivalent programmes.

The 2022 critique focused on programme design. Subsequent work has added a structural layer that is more consequential. The substantive treatment of role collapse in donor-funded ecosystems sits in Political Economy Ecosystem; the implication for acceleration specifically is that organisations that fund, deliver, and evaluate acceleration programmes simultaneously create a built-in conflict of interest. When the same actors design programmes, select ventures, and write the evaluations that justify continued funding, the result is systematically positive reporting, weak learning, and very little willingness to shut down ineffective models.

Hwang and Horowitt's The Rainforest names what donor-architecture acceleration systematically fails to produce. Their core argument: mature entrepreneurial ecosystems generate value through emergent culture - trust networks, tacit knowledge, peer norms, weak ties between operators across firms - that programmes cannot directly manufacture. Programmes can support cultural conditions; they cannot substitute for them. The acceleration architecture that delivers content through structured curricula misunderstands what ecosystems actually do. Hwang and Horowitt's clinical observation: the most consequential ecosystem outputs - operational candour between founders, frank investor-founder relationships, the diffusion of hard-won lessons from one venture's failure into another's strategy - emerge from culture, not from cohorts. The donor architecture funds cohorts because cohorts are programmable. It does not, and cannot, fund the cultural conditions in which the actual learning happens.

The structural critique of BDS funding architecture treated in Ecosystem Characteristics operates here at the acceleration-specific level. BDS funding is "typically prescriptive, programmatic, and short-term," compelling ESOs to redirect efforts from aiding enterprises towards appeasing donors. The result is generic cohort designs that are familiar to donors and enable large participant numbers using less experienced staff. The four-node vicious cycle the WDI East Africa study identifies - company unwillingness to pay forces grant dependency; grant dependency turns ESOs into donor implementers; competitive RFP structures fragment resources; talent leaves for investment funds in around two years - is the operational form of the architecture problem. Each node reinforces the others, and the equilibrium is structurally produced by the incentives of the actors who benefit from programme continuity.

The evaluation issue is a governance issue.

The design failures identified in 2022 remain largely uncorrected. The WDI East Africa study provides the most granular recent documentation: generic cohort-based programmes overemphasise business models and pitch deck preparation at the expense of substantive support for product-market fit, operational efficiency, leadership, regulatory compliance, and team building. One-size-fits-all cohorts mix sectors and stages. Mentorship is limited to brief, often mismatched interactions across a three-to-six month window. Investment readiness training rarely addresses the deeper technical and communication skills needed to position business models and financials to investors - it teaches founders to build visually appealing pitch decks. Interviewees note that these problems were flagged a decade ago and persist unchanged.

One donor interviewed for the WDI study asks the core question directly: "Are we actually contributing to the problem, or are we actually helping solve the problem? Because the systemic issues that drove the creation of incubators and accelerators still exist, and in some ways have become more entrenched."

The Cohen, Bingham and Hallen design-features research names the structural reason why redesign rarely happens. Programme design is the dimension of acceleration that most directly determines outcome quality, but it is also the dimension that requires the most accumulated expertise and the longest learning cycles to get right. Donor-funded programmes operating on three-year cycles with two-year staff tenure and quarterly evaluation reporting have the wrong institutional architecture to develop, test, and iterate the design features that the academic literature identifies as load-bearing. The institutional discipline the design challenge requires is not what donor-funded ESOs are funded to build.

The World Bank's XL Africa reflection is explicit on the gap that matters most at the scaling stage: mentorship and post-programme support are structural blind spots, not minor design details. The GALI synthesis confirms it: the quality of the network an accelerator is embedded in is a critically important part of the offer, and programmes that facilitate genuine peer networking and strategic introductions outperform those that deliver classroom instruction. This is precisely the cultural-capital dimension Hwang and Horowitt name - and precisely the dimension donor-funded programme architecture cannot deliver through curriculum.

"The accelerators become islands of activity followed by lots of frustration." - Roger Norton, formerly Founders Factory Africa

AI tools have begun materially changing what acceleration can deliver per dollar of support. The substantive treatment of AI's effects on capability building and ecosystem support sits in What AI changes about African scaling; the implication for acceleration specifically is that the classic model - fixed curriculum, three-month cohort, demo day - was designed for a pre-AI world. Off-the-shelf AI tools now make it possible to personalise curriculum paths, automate basic diagnostics, and provide founders with targeted content and expert matching at a fraction of historical cost.

AI-powered personalisation could shift the economics of quality support dramatically. Accelerators able to build venture-specific learning journeys - using cohort data to identify which interventions correlate with improved outcomes, and matching founders to mentors based on actual need rather than convenience - will remain relevant. Those continuing to offer fixed curricula and demo days will increasingly find themselves competing with self-serve AI learning and founder communities that deliver more value at lower cost.

The deeper implication is that AI tooling makes the academic-literature-derived design features (intensive, personalised, peer-comparison-rich, capability-building) cheaper to deliver than they have ever been. The donor architecture that funds generic cohort programming is now sustaining a model that costs more per founder than the better alternative AI tooling enables. The cost-efficiency case for redesign is converging with the evidence-base case for redesign. The institutional question is whether the apparatus that funds the model can let go of the model fast enough for the redesign to scale.