How to select companies in an open call: requirements, selection criteria, and jury

Key Takeaways

• Selection decides much of what the program will end up showing. In data from the Global Accelerator Learning Initiative (GALI), companies accepted by accelerators already had nearly twice the revenue of rejected ones when they applied.
• Juries are better at screening out low-potential companies than at ranking the best ones. In a business plan competition in Nigeria, judges' scores were not correlated with survival, employment, sales, or profits three years later (McKenzie and Sansone, 2019).
• Which judge a company gets matters. At the Cali Chamber of Commerce accelerator, 34% of judges scored systematically more generously or more harshly than the rest (González-Uribe and Reyes, 2021).
• A program's effect depends on the profile of who gets in. In the same Cali case, the program helped high-potential companies and did not improve low-potential ones that got in thanks to a generous judge.
• When there are more eligible companies than places, a lottery among those with similar scores is a fair rule and makes it possible to measure impact. That is how more than 700 winners of Nigeria's YouWiN! program were chosen (McKenzie, 2017).

The open call looks like a formality before the program, and it is one of the design decisions with the biggest consequences. It determines who the resources reach, how much can change in those companies, and whether it will be possible to say what the program did. A program that selects companies that would have grown anyway will look good in the report even if it changed little.

If you put together open calls for a chamber of commerce, an economic development department, a foundation, or a multilateral, here are seven steps to define requirements and criteria, set up the jury, decide the cutoff, and leave the selection ready for evaluation, with the evidence on what works.

What does selecting well mean in a business support program?

Selecting well means choosing, among the companies that meet the conditions to benefit from the program, those for which it can change something they would not have achieved on their own. The IDB book Rethinking Productive Development (Crespi, Fernández-Arias, and Stein, eds., 2014), when reviewing innovation policies, uses the concept of additionality for this: whether support generates something the company would not have done on its own. It recommends targeting the companies most exposed to the failure the program aims to correct.

This does not mean choosing the weakest. The Cali case, below, shows that the program mainly helped high-potential companies and did not transform weak ideas. The question is which of the companies with potential need the support to realize it, and which already have the means to do it alone.

In practice, selection has three layers worth separating:

Why is it so hard to predict which companies will grow?

It is hard because at the top of an open call the differences between companies are small, and growth depends on many things a jury cannot see. The study by McKenzie and Sansone (2019), with more than 2,000 participants in a business plan competition in Nigeria, found that the scores of 20 expert judges, using a ten-criterion template and anonymized plans, did not predict three-year outcomes. A few entrepreneur characteristics predicted somewhat better, though with little precision, and machine learning models did not improve much.

An experiment in Ghana compared expert panels with surveys (Fafchamps and Woodruff, 2017). An ability measure taken by survey was strongly correlated with growth, and the panel score added some information. With the caveat that the sample was small, the authors conclude that panels are better at separating out low-potential companies than at distinguishing among the best, and that surveys are much cheaper.

A well-designed jury is therefore useful for filtering the bottom of the list. Within the top, scores are less precise than they seem, and it pays to decide the cutoff with a transparent rule.

How to select companies in an open call step by step

1. Define who the program is for

Selection starts with the theory of change: what change the program seeks and which companies can achieve it with this support and not without it. Write the profile around the problem the companies share, on top of their size or sector.

Check who gets left out if the profile is written too broadly. A program open to "SMEs with potential" will receive more applications from companies that already have staff to apply, and selection ends up rewarding the ability to write a good application.

2. Separate eligibility requirements from selection criteria

Eligibility requirements are yes-or-no conditions, verifiable with a document: territory, size, years in operation, formal registration. Selection criteria rank eligible applicants and carry a score.

A minimum screen before the jury saves work and improves selection. YouWiN! excluded applicants below a minimum score before the lottery, and at the Cali accelerator an initial screen left 135 of 255 applications before they went to the judges. Also check that the requirements do not push out the beneficiary the program says it wants: procedures only the largest or long-formalized companies can complete leave out precisely those who could gain the most.

3. Size the open call against places and dropout

Estimate how many applications you need to fill the places with companies that really fit the profile, and how many will be lost along the way. In GALI's data, acceleration programs select on average fewer than 13% of applicants per cohort.

Not every program has that demand. In Ghana the target was 400 applicants, 92 arrived by the first deadline, and the territory had to be expanded to reach 335. Then 30% did not attend any training and only 42% made it to the panel. Early dropouts worked more hours and were more often in retail, and on average had less schooling and lower financial literacy: dropout mixed opportunity cost with lower capability. If demand does not exceed the places, the work shifts to the call itself: outreach with local partners, a shorter application and, if you want to evaluate, phased entry.

Keep the application short. In Nigeria, plans averaged 13,000 words and scoring each one took 30 to 45 minutes; a long form screens for available time more than for potential.

4. Set up the jury with a rubric and reduce noise between judges

Use several judges per application, assign them at random, and give them a rubric with defined criteria and a description of each level. This is an illustrative rubric; each program defines its criteria and weights according to its theory of change and publishes them in the call:

Then adjust each score for the judge's generosity. A simple way, which can be done in a spreadsheet, is to subtract from each score the average of the scores that judge gave and add the overall average across all judges. This adjustment is only valid if judges were assigned at random and each scored several applications; otherwise a judge looks harsh just because they got weaker applications. Keep both the raw and adjusted scores, and publish the rule in the call.

Pay attention to the committee's composition too. In GALI's data, programs with more than 45% women on the selection committee, which is the median, received more applications from all-women teams. It is a correlation, not a proven effect.

5. Combine the jury score with objective data

Add a short survey to the application with data that can be verified: sales, employees, practices the company already uses, team experience. In Ghana, surveys explained more of growth than panels did, and in Nigeria a few entrepreneur characteristics predicted somewhat better than the judges. That does not mean scoring by age or gender; it means a survey with verifiable data adds information a jury cannot see.

Data reported to win a place tends to be inflated. For the baseline, apply the same instrument to all applicants before announcing results, without it counting toward the score, and check it against records where possible.

6. Decide the extremes by score and the middle with a transparent rule

Use the score to rule out those who clearly do not fit the profile and to secure those who clearly do. For the middle group, where scores are similar, decide with an explicit rule: a lottery among companies with similar scores, or quotas by territory or gender defined before opening. If you combine quotas and a lottery, draw within each quota.

Lotteries meet resistance in many institutions in the region. The board or funders may feel the institution is giving up its decision, and oversight bodies ask about objective criteria. A lottery is easier to defend when presented as what it is, a tie-breaking rule among companies the jury cannot tell apart, announced in the call, carried out in public with minutes, and leaving the jury in charge of the ends of the list. YouWiN! drew lots among semifinalists for two reasons: the uncertainty of scores in that range, together with the risk of capture through political or ethnic ties, and the possibility of measuring impact.

7. Leave the selection ready for evaluation and review the call before publishing

Keep the scores of every application, with the judge who gave them, including those dropped at the initial screen, along with contact details and survey data for non-selected companies. If there was a lottery, write down the range drawn, the procedure, and a lottery-ordered waiting list for replacements. If there are enough applications near the cutoff, the evaluation can use a regression discontinuity, as long as judges score without knowing the cutoff and every exception is recorded.

In GALI's data, rejected applicants answered follow-up surveys at a 47% rate, compared with 77% for participants, so ask for their authorization when they apply and explain how their data will be used. How to choose the evaluation method based on the selection rule is covered in how to measure the impact of a business program.

Before publishing, check that the call states the requirements and the document that proves each one, the criteria and their weights, who scores and how scores are adjusted, the rule for the middle group, the authorization to use non-selected companies' data, and the date results will be announced.

Real case: what the Cali Chamber of Commerce accelerator taught about selection juries

The ValleE accelerator, run by the Cali Chamber of Commerce in Colombia, offered training, advice, and visibility, without cash. The Chamber made an unusual design decision: assign judges at random and keep every score. That allowed Juanita González-Uribe and Santiago Reyes to study one of its calls (González-Uribe and Reyes, 2021).

Of 255 complete applications, an initial screen left 135. Each was scored by three randomly assigned judges, from a pool of 50, on five criteria: clarity of the model, innovation, scalability, potential profitability, and team. The 35 with the best average score got in, a number set in advance by budget and space.

Because judges were assigned at random, the authors could measure how much the judge mattered. 34% deviated systematically from the rest, compared with 6.8% expected by chance, and three-judge panels were not enough to correct it. Scores adjusted for each judge's generosity did rank companies well by potential; the unadjusted average was not correlated with performance.

The same random assignment made it possible to estimate the program's effect on companies whose entry depended on the judge they got. Over the three years after applying, their annual revenue was on average about 66 million pesos higher, 166% above the rejected applicants' average. The effect came from high-potential companies that a harsh judge would have left out. Low-potential companies that got in thanks to a generous judge did not improve, and participants valued advice and visibility most (74%) over standard training (8%).

Frequently asked questions about selecting companies in open calls

What selection criteria are used in a business program?

The most common assess the business model, team capacity, growth potential, innovation, and fit with the program's objective. The Cali accelerator used five: clarity of the model, innovation, scalability, potential profitability, and team. More important than the list is that each criterion has a rubric with described levels and measures what the program can change.

Should companies pitch to the jury?

A pitch can add information, and it also opens the door to presentation biases. In an experiment where the same pitch was narrated by different voices, 68% of participants chose to fund the one narrated by a male voice (Brooks et al., 2014), although those participants were not investors. If you use pitches, use a strict rubric and a diverse panel, and combine them with the written application and the survey.

At Suricata Labs, since early 2026, we have used an AI tool we built to evaluate and train pitches. It has let us score with far more objective criteria than human judges, especially when dozens of pitches have to be reviewed in a few days.

How do you avoid gender bias in selection?

With a diverse selection committee, explicit rubrics, and a review of results by gender before closing the list. GALI found that women-led teams do not have lower acceptance rates at accelerators but do have more difficulty raising capital afterwards, and that the presence of women on the committee and among mentors is associated with more applications from all-women teams.

Is it fair to allocate a program's places by lottery?

It is fair when the lottery is held among companies that meet the requirements and have similar scores, because it gives all of them the same chance and reduces the risk of favoritism. The IDB and World Bank handbook Impact Evaluation in Practice considers it a transparent rule when there are more eligible applicants than places. It should be announced in the call and carried out in public. With few companies in the lottery, comparing winners and non-winners works for outcomes that vary little or when several editions are pooled.

What happens to companies that are not selected?

They should be told clearly why they did not get in and, if possible, offered an alternative path or feedback. They are also valuable to the program: with their authorization, their data serve the evaluation and help adjust the profile and criteria of the next call.

What to have in writing before opening the call

A well-designed open call puts in writing, before opening, the profile of company the program serves, the requirements that protect it, a rubric with described levels, the way jury scores will be adjusted, and the rule for deciding the middle of the list. With that, the program reaches the companies it can help most and keeps what it needs to show it later.

If you share the draft of your next open call with us, we will review the requirements, the rubric, and the selection rule with you before you publish it.

Request a design session | How we work with institutions

Read also: How to design a business support program | How to write the terms of reference for an operator