Every budget cycle, ministries and agencies across India and the Gulf commission founder programs. The launch photograph is the easy part. Two years later, the question that actually matters — how many durable, investable ventures did this produce? — is often unanswerable, because nobody specified it at the start. This article sets out what a government is really purchasing when it buys an entrepreneurship program, what four decades of evidence say about which components work, and how to write a mandate that buys activation rather than announcements.
The idea in brief. A government that commissions an entrepreneurship program is buying four things: a selection filter that finds ready founders, a capability transfer that changes what those founders do, network access to capital and customers, and a signal that lowers the cost of trusting them. The evidence is clear that the first two dominate: cash and coworking alone show no measurable effect in the best-designed studies, while structured schooling, personal-initiative training and well-targeted grants produce large, lasting differences. Programs fail when they are specified by inputs (cohort size, events held) rather than outcomes with time horizons. The tender, not the curriculum, is where most programs are lost.
The announcement economy
Innovation policy has a peculiar incentive structure. The people who commission a program are judged on a political calendar; the outcomes that justify the program arrive on an entrepreneurial calendar. A minister needs something to say within twelve months. A venture needs twenty-four to thirty-six months to demonstrate that it exists in any meaningful sense — revenue, customers, a second round of capital, a payroll. The gap between the two calendars is where the announcement economy lives: programs designed to be launched, counted and reported, rather than to change what happens to a thousand founders over five years.
None of this is a criticism of ambition. The scale of public commitment to entrepreneurship across the India–Gulf corridor is historic. India’s Startup India initiative, launched in 2016, has recognised well over 150,000 startups through its national registry, according to the Department for Promotion of Industry and Internal Trade. Bahrain, Saudi Arabia, the UAE, Oman and Qatar have each embedded entrepreneurship in their national visions and built dedicated agencies, funds and labour programs to deliver it. The money and the mandate are there. What separates jurisdictions now is activation: whether the machinery converts intention into ventures that last.
Daniel Isenberg’s much-cited 2010 argument in Harvard Business Review still frames the problem well. Governments, he observed, tend to copy the visible artefacts of successful ecosystems — the incubator building, the venture fund, the university science park — and are then puzzled when the ecosystem does not follow. The artefacts are outputs of a working system, not its cause. The same mistake recurs at program level: the cohort, the demo day and the mentor list are visible; the selection quality, the training dosage and the follow-through are not, and they are what determine whether anything happens.
Four things a government is actually buying
Strip away the branding and every founder program is a bundle of four services. Naming them separately matters, because they have different costs, different evidence bases and different failure modes.
| Component | What it does | What the evidence says | Typical failure |
|---|---|---|---|
| Selection | Finds founders who are ready to use the program | Judges’ scores predict outcomes weakly; structured criteria and observed behaviour predict better | Choosing polished pitchers over capable operators |
| Capability transfer | Changes what founders know and do | Generic training has small effects; intensive, behavioural and practice-based training has large ones | Two-day workshops counted as “training” |
| Network access | Connects founders to capital, customers and talent | Effects depend on network quality; weak networks add little | A mentor list nobody activates |
| Signal | Certifies founders to investors and buyers | Real when the program is selective and known; otherwise absent | A certificate no investor recognises |
The most rigorous single study of a government accelerator remains the evaluation of Start-Up Chile by Juanita González-Uribe and Michael Leatherbee, published in the Review of Financial Studies in 2018. Their finding is uncomfortable for anyone who has ever funded a coworking space: the program’s basic bundle — equity-free cash and workspace — produced no measurable effect on venture performance. The component that did produce measurable gains in fundraising and survival was the entrepreneurship schooling: structured training and mentoring with accountability. The money got founders in the door; the method changed what happened to them afterwards.
A second strand of evidence explains the mechanism. Benjamin Hallen, Susan Cohen and Christopher Bingham’s 2020 study in Organization Science, asking bluntly whether accelerators work, found that the effective ones compress learning: ventures reach milestones faster because intensive, broad and structured consultation forces founders to confront hard questions early. Sandy Yu’s 2020 analysis in Management Science adds a finding that should reshape how ministries read their own dashboards: accelerated ventures were more likely to shut down early and raised less capital, which the author interprets as faster resolution of uncertainty. Founders learned sooner that an idea would not work, and stopped. For a program funder, early closures are therefore not automatically a failure statistic. They may be the program working.
The dosage problem
If capability transfer is the component that carries the evidence, then how much of it, and of what kind, becomes the central design question — and here the development-economics literature is unusually rich, because business training has been the subject of dozens of randomised trials.
David McKenzie and Christopher Woodruff’s review of that literature in the World Bank Research Observer reached a sobering conclusion: standard business training tends to change business practices modestly and profits barely at all, with effects often too small to detect in samples of the size most programs can afford. The GATE experiment in the United States, analysed by Robert Fairlie, Dean Karlan and Jonathan Zinman (NBER Working Paper 17804, later published in the American Economic Journal: Economic Policy), found that subsidised entrepreneurship training produced short-run effects on business ownership that faded within a few years. Generic content, delivered briefly, does not stick.
Against that backdrop, the Togo experiment reported by Francisco Campos and colleagues in Science in 2017 stands out. Comparing a traditional business-skills curriculum with a psychology-based “personal initiative” training — teaching proactive, self-starting, persistent behaviour — the researchers found the behavioural program raised firm profits by roughly 30 per cent, while the traditional curriculum’s effect (around 11 per cent) was not statistically distinguishable from zero. The lesson is not that accounting does not matter. It is that what founders do — how they search for customers, how they respond to setbacks, how they experiment — responds to training in a way that what they know does not, and that programs built around behaviour and practice outperform programs built around content.
For a commissioning agency, the design implications are concrete:
- Intensity beats coverage. A twelve-week program with weekly accountability for 100 founders will produce more measurable change than a one-day event for 5,000. The announcement economy prefers the second; the evidence prefers the first.
- Behaviour beats content. Curricula should be organised around actions — interviewing customers, shipping a prototype, running a pricing test — with content delivered in service of those actions, not the reverse.
- Accountability is a component, not a courtesy. Structured check-ins, milestone reviews and cohort peer pressure are the delivery mechanism through which training becomes practice.
- Founder psychology is a legitimate curriculum subject. Personal initiative, persistence and error-management training are evidence-backed and cheap relative to capital.
Money versus method: what the grant evidence says
Governments like grants because they are visible, countable and fast. The evidence on grants is more encouraging than the evidence on generic training — provided the grants are large enough to matter and the selection is honest about its limits.
The clearest case is Nigeria’s YouWiN! business-plan competition, evaluated by McKenzie in the American Economic Review in 2017. Winners received grants averaging around US$50,000 — large relative to the firms — and were followed for three years. Among new firms, winners were roughly 37 percentage points more likely to be operating a business three years later and roughly 23 percentage points more likely to employ ten or more people than comparable non-winners. Existing firms also grew. Because part of the winner pool was selected by lottery among semi-finalists, the study could estimate causal effects cleanly, and it could also test the judges: their scores were only weakly related to subsequent outcomes. Random selection among founders who had cleared a quality bar did about as well as expert ranking.
Two lessons follow. First, capital that is large relative to the venture’s needs, and unconditional on equity, can be transformative at the early stage — a finding consistent with Sabrina Howell’s work on R&D grants in the American Economic Review, which showed that small early grants roughly doubled the probability that a young firm subsequently raised venture capital. Second, selection is harder than it looks. Panels of experts scoring pitches add less than they believe. The practical response is not to abandon selection but to redesign it: clear a quality threshold with structured criteria, then allocate among qualifiers with less confidence in fine rankings — and, where a program can bear it, use the resulting randomness to evaluate itself.
An activation funnel, honestly drawn
The following funnel is an illustrative model, not a report of any single program; it is drawn from the shape that well-run national founder programs tend to take when their numbers are reported honestly. Its purpose is to make one point: the drops are not equal, and only some of them are failures.
- From applicants to selected. A national call attracts many people who want the grant more than the work. A 10–15 per cent selection rate is healthy; a 60 per cent rate means the program has no filter, and the downstream numbers will show it.
- From selected to completed. Some founders leave because the program is demanding. That is the filter working late. A completion rate above 75 per cent with real accountability is good; a 98 per cent rate usually means nothing was demanded.
- From completed to operating at 24 months. This is where Yu’s finding matters. A share of closures reflects founders who learned quickly that the idea would not work. The metric to watch is not survival alone but survival plus the quality of what survived — revenue, customers, team.
- From operating to externally financed. A minority of any cohort will raise institutional capital, and this is fine: many durable businesses never should. What a government should count here is any independent, outside validation — customers, partners, lenders or investors.
A program that reports only the top of this funnel (“1,000 founders trained”) is reporting an input. A program that reports the bottom, with dates, is reporting an outcome. The difference is the whole difference between an announcement and an activation.
What “Startup Ready” looks like when it is built for activation
HexGn’s founder program, Startup Ready, was designed around these findings and has been delivered as agency-commissioned cohorts on both sides of the corridor. Its architecture is worth describing not as a sales pitch but as a worked example of the design principles above:
- A twelve-week arc with four gates — validation, business model, build, pitch — where each gate requires evidence produced outside the classroom: customer interviews conducted, a prototype used by real people, a pricing test run, a pitch delivered to investors who are free to say no.
- Cohorts sized for accountability. Twenty to forty founders per cohort, so that a program director knows every venture and every founder knows that absence is noticed.
- Mentors with obligations, not titles. Each mentor commits to specific hours, specific ventures and specific deliverables. A mentor list that is a list is a marketing asset; a mentor bench with obligations is a program component.
- No equity. Government-commissioned programs should not create a conflict between the program’s interest and the founder’s. Signal and network are the program’s return.
- Outcome tracking by design. Every cohort reports operating status, revenue bands, team size and external validation at 6, 12 and 24 months. The reporting is built into the founder agreement at intake, which is the only time it can be.
The design deliberately concedes the announcement metrics. A twelve-week cohort of thirty founders is a modest headline. Three cohorts a year across five sectors, each with honest 24-month outcomes, is a national pipeline.
A composite case: the national program that had to be redesigned
The following is a composite drawn from patterns observed across several agency engagements; details are altered and no single client is described.
An economic-development agency in the Gulf launched a national founder program with a large headline: a thousand entrepreneurs in year one, delivered through a chain of two-day bootcamps across the country, with a grant competition at the end. Year one delivered exactly what was promised — the bootcamps happened, the thousand were counted, the winners were photographed. Year two brought the questions. Of the grant recipients, a majority had used the money to survive rather than to grow; of the thousand trained, the agency could not say how many were still operating, because nothing had been recorded that would let it know.
The redesign kept the political commitment and changed the machinery. Selection moved to a structured, criteria-based process with a quality threshold. Training moved from two-day events to twelve-week sector cohorts of thirty, with weekly milestones and a mentor bench drawn from corporates that had agreed to be early customers. The grant became a milestone-based instrument released against evidence. Outcome tracking was written into the founder agreement. The year-two headline was smaller — a few hundred founders rather than a thousand — and the year-four report was the first the agency had ever been able to publish with survival, revenue and financing figures in it. That report, not the year-one photograph, is what the agency now uses to justify its budget.
What ministers should put in the tender
Most program failures are contracted at procurement. A tender that specifies inputs buys inputs. The following checklist converts the evidence above into procurement language:
| Tender clause | Specify | Avoid |
|---|---|---|
| Outcomes | Operating status, revenue bands, employment and external validation at 12 and 24 months, per cohort | “Number of entrepreneurs trained” |
| Selection | Structured criteria, a quality threshold, published selection rate, evaluation of predictive power after the fact | Unstructured jury scoring as the sole method |
| Dosage | Minimum contact hours, cohort size cap, milestone gates, accountability mechanism | Events counted as training |
| Mentors | Named bench, committed hours, sector match, conflict disclosure | A list of names |
| Capital | Grant size relative to venture need; milestone release; no equity for public programs | Small grants spread thinly to maximise headcount |
| Data | Agency owns the outcome data; founder consent at intake; standard definitions | Provider-held data the agency cannot audit |
| Evaluation | Where feasible, oversubscription used to create a comparison group | Provider self-reporting as the only evidence |
| Sustainability | Alumni obligations, corporate co-funding path, handover plan | Programs that end when the contract does |
The evaluation clause deserves emphasis. Most national programs are oversubscribed. That oversubscription is an evaluation asset: if selection clears a quality bar and then allocates places transparently among qualifiers, the program creates its own comparison group at no cost. This is precisely how the YouWiN! and Start-Up Chile evidence became credible, and it is available to any agency willing to specify it.
What could go wrong
- Headcount inflation. The target is set in founders trained; the provider optimises for it; the cohort becomes a lecture hall. Antidote: cap cohort size in the contract.
- Grant capture. A subset of applicants are professional grant-seekers. Antidote: milestone-based release and observed behaviour in selection.
- Mentor decoration. Prominent names agree to be listed and never appear. Antidote: committed hours and per-venture assignment.
- Metric drift. Reporting migrates to whatever is easiest to count. Antidote: definitions fixed at tender, reported on a schedule, audited.
- Provider dependence. The program lives only while the contract lives. Antidote: alumni structures, corporate co-funding and a handover clause from day one.
- Misreading early closures. A rise in closures after an intensive program is read as failure and the program is cut. Antidote: track closures alongside the quality of survivors, and brief decision-makers on the evidence before the numbers arrive.
Questions ministers actually ask
“How many startups will this produce?” Fewer than the headline and more than the sceptics expect — if the funnel is designed for it. A realistic national program produces a cohort in which roughly a third to a half of ventures are operating with revenue at 24 months and a minority attract outside capital. Any provider promising materially better should be asked for the 24-month data behind the claim.
“Should we give grants or training?” Both, sequenced. Training with accountability first, because it changes what founders do; capital second, sized to matter and released against milestones, because capital handed to unprepared founders funds survival rather than growth.
“Can we do this at national scale?” Yes, by running many cohorts rather than one large one. Scale in this field comes from replication, not from enlarging the room.
“How do we know the provider is any good?” Ask for outcome data from previous cohorts with definitions and dates, and ask which of their programs failed and what they changed. A provider with no failures has no outcome data.
“What is the political risk of honest metrics?” Lower than the risk of dishonest ones. Early closures and modest survival rates are the normal shape of the evidence; a program that reports them credibly is defensible in a way that a program with a thousand photographs and no outcomes is not.
Methodology & data notes
The effect sizes cited from the academic literature are reported approximately and readers should consult the original papers, linked below, for exact estimates, confidence intervals and context; Start-Up Chile, YouWiN! and the Togo experiment are cited because they used randomised or lottery-based designs that permit causal interpretation. The activation funnel is an illustrative model drawn from the shape of well-reported programs, not a dataset. The composite case is a pattern assembled from multiple engagements with details altered. Startup India registry figures are as published by DPIIT and change continuously. Companion articles in this series examine ecosystem measurement (article 32), incubator and accelerator design (article 37) and the founder-readiness evidence in detail (article 40).
References & further reading
- González-Uribe, J. & Leatherbee, M. (2018) — The Effects of Business Accelerators on Venture Performance: Evidence from Start-Up Chile, Review of Financial Studies
- Hallen, B., Cohen, S. & Bingham, C. (2020) — Do Accelerators Work? If So, How?, Organization Science
- Yu, S. (2020) — How Do Accelerators Impact the Performance of High-Technology Ventures?, Management Science
- McKenzie, D. (2017) — Identifying and Spurring High-Growth Entrepreneurship: Experimental Evidence from a Business Plan Competition, American Economic Review
- Campos, F. et al. (2017) — Teaching personal initiative beats traditional training in boosting small business in West Africa, Science
- McKenzie, D. & Woodruff, C. (2014) — What Are We Learning from Business Training and Entrepreneurship Evaluations around the Developing World?, World Bank Research Observer
- Fairlie, R., Karlan, D. & Zinman, J. (2012) — Behind the GATE Experiment: Evidence on Effects of and Rationales for Subsidized Entrepreneurship Training, NBER Working Paper 17804 (published 2015 in American Economic Journal: Economic Policy)
- Howell, S. (2017) — Financing Innovation: Evidence from R&D Grants, American Economic Review
- Isenberg, D. (2010) — How to Start an Entrepreneurial Revolution, Harvard Business Review
- Startup India — DPIIT startup recognition and program statistics
- Start-Up Chile — program information
HexGn designs and runs founder, campus and research-commercialisation programs for governments, agencies and universities across India and the Gulf — specified by outcomes, delivered end to end, and reported honestly at 24 months.