Accelerator Selection Day: How to Run Interviews and Pitches for Your Cohort

For its 2025 cohort, MedTech Innovator received nearly 1,500 applications, heard 213 pitches at Road Tour events in Los Angeles, College Park and Dublin, collected feedback from 430 subject matter experts and accepted 65 companies, the top 4 per cent (Medical Economics, June 2025). Read those numbers as an organiser and you see a timetable before you see a jury. Two hundred pitches and four hundred experts only produce a fair result if someone decides who hears whom, when, and against which scale.

Most accelerators run a smaller version of the same problem. After the online screening, a shortlist of a few dozen teams comes in for one intense day, and a handful of jurors have to turn that day into a cohort. The person who defends the result is usually you, the programme lead. You defend it in front of the partner or investment committee that signs off the cohort. Three months later, that partner asks why one startup made it and another did not.

This guide argues that a selection day is a scheduling problem before it is a judgement problem. Which jurors see which startup, in which order and after what calibration shapes your cohort as much as the criteria do. Below you will find the interview matrix, the question set, the calibration routine, the deliberation and the rejection plan.

What is an accelerator selection day?

The format borrows from several others, so a definition helps. An accelerator selection day is the final stage of a cohort application process. Shortlisted teams meet the selection panel, usually for a short pitch followed by an interview. Then the programme decides who joins the next cohort.

A qualitative study of ten accelerators found the same staged pattern almost everywhere: online application, expert screening, then a jury or committee (Beyhan, Akçomak and Cetindamar, Entrepreneurship Research Journal, 2021). These juries had 3 to 40 members, drawn from angel investors, venture capital firms, corporates, sponsors, mentors and academics. The same study found two main criteria: the opportunity, where a working prototype and a clearly identifiable market signal a short time to market, and the team, judged on commitment, coachability, market knowledge, preparedness and how well the founders work together. The team weighed more than anything else, while financial projections were rarely used. Your interview format should therefore test the team, not the spreadsheet.

Formats vary in length. Y Combinator runs 10-minute video calls with two or three partners and spends them only on questions and on what the founders have built (YC Interview Guide, 2026). The timbuktoo ManuTech programme run with UNDP (the United Nations Development Programme) gave each of 67 eligible startups a 10-minute pitch. It scored them out of 100 on seven criteria and admitted the top 20 above a 60 per cent threshold (UNDP, March 2026). Both examples show that a short slot works, as long as every minute of it is planned.

A selection day is not a pitch competition. A competition ranks entrants in public for a prize, and its scoring rubric and jury design are covered in our guide on how to run a pitch competition. A selection day decides membership behind closed doors, often with several panels interviewing in parallel. The result is a group that has to work together for months. It also differs from the shortlist of an open innovation challenge, where a business unit picks the startup that fits one specific problem.

Why the schedule decides as much as the rubric

Parallel panels are where the trouble starts. In a replication of the review process of the US National Institutes of Health (NIH), 43 reviewers rated the same 25 grant applications. Their ratings showed no agreement on quality, even though all of them had received the same instructions (Pier et al., PNAS, 2018). The authors concluded that the outcome depended more on the reviewer an application was assigned to than on the application itself. If each startup meets only one panel, your cohort partly reflects the panel draw.

Order adds a second distortion. An analysis of the Eurovision Song Contest and of figure skating championships found that ratings rose with a candidate’s position in the running order. That held whether judges scored at the end or straight after each performance (Bruine de Bruin, Acta Psychologica, 2005). Interviewers also grade relative to their day. In more than 9,000 MBA admission interviews, a candidate’s predicted score fell when the applicants interviewed earlier that day had scored higher (Simonsohn and Gino, Psychological Science, 2013). The team in the eleventh slot, after three excellent interviews, starts with a handicap it never sees.

None of this means your jurors are careless. It means the rubric cannot do the work alone. The timetable has to.

Build the interview matrix

Because the timetable carries part of the judgement, design it before you invite a single juror. A workable slot is 30 minutes. It holds 5 minutes of pitch or demo, about 12 minutes of questions and 3 minutes for the founders’ own questions. The last 10 minutes belong to the jurors, who score alone while the next team walks in. A day from 08:00 to 18:30 with a briefing, a calibration lunch and a deliberation leaves two blocks of seven slots per panel. Keep the last slot of each block empty as a reserve, and you have 12 interviews per panel.

Then decide on coverage. With three panels of two jurors, 12 interviews each give you 36 interviews in total. If every startup meets one panel, you can see 36 teams. If every startup meets two different panels, you see 18, and each result rests on four jurors instead of two. Given the NIH finding, the second option is usually the better trade for a final round. Fewer teams, but no decision hinges on the draw.

Five rules turn that grid into a fair one:

• Two panels per finalist: each team meets two panels with no juror in common. No single juror sees a team twice, and no single pairing decides its fate.
• New pairs after lunch: reshuffle the juror pairs between the morning and the afternoon block. A particularly strict pair then cannot pull down a whole block of teams.
• Conflict exclusions: a juror who has invested in a competitor, advises the team or works for one of its customers never gets that slot. Collect these declarations a week before and build them into the matrix, not into the morning of the day.
• A drawn running order: draw the order of teams and tell them it was drawn. Avoid placing the same team late in both of its interviews. Randomisation does not remove the order effect, but it gives every team the same chance of a good slot (Bruine de Bruin, 2005).
• One reserve slot per block: it absorbs a delayed train or a broken video link without moving everybody else.

Picture the alternative. At 14:10 a founder’s flight is late, the coordinator swaps two teams by hand, and the swapped team lands in front of the juror who mentors its closest competitor. Nobody notices until the deliberation, when the low score is already on the table. A matrix with built-in exclusions refuses that swap.

Ask every startup the same questions

A fair matrix is wasted if each panel asks something different. In personnel selection, structured interviews (the same questions in the same order, rated on a defined scale) reached a mean validity of 0.42 in a 2022 re-analysis. That was the highest value among the predictors the authors discussed (Sackett et al., Journal of Applied Psychology, 2022). Founders are not job candidates, but the mechanism carries over. When every team answers the same questions, the differences you score are differences between teams, not between interviewers.

The wording matters as much as the consistency. A study of question and answer sessions at TechCrunch Disrupt found that investors tended to ask male founders about potential gains and female founders about potential losses. In a follow-up experiment, founders asked gain-oriented questions received twice as much funding (Kanze, Huang, Conley and Higgins, Academy of Management Journal, 2018). The fix is simple: write a question set in which every team gets the same mix of upside and risk questions.

A set for a 12-minute interview could look like this:

• What is your company working on, in two sentences? Y Combinator says this is typically its first question (YC Interview Guide, 2026).
• What has changed since you applied?
• Who is your customer, and how do you know they will pay?
• How big can this become in five years?
• What would make this fail in the next twelve months?
• How do the founders split decisions, and when did you last disagree?

For the score sheet, reuse the five-criterion logic from the pitch competition guide above, but weight it towards the team, which carried the decision in the accelerator study. Anchor every score with a description. A team score of 5 might read: “answered the risk question with evidence, and the founders corrected each other constructively”. A juror who has to match a description scores against the description. A juror with a bare scale scores against the previous team.

Calibrate before the first interview and again at lunch

Questions and descriptions only align your jurors if they have used them together once. The NIH study found that reviewers did not even agree on how many weaknesses justified which score. Calibration targets exactly that translation step.

Run it in three moves:

• Anchor case at 08:00: all jurors watch the same recorded pitch, for example from a past applicant who has agreed to it. Everyone scores alone, then the scores go up on a screen. Discuss every gap of more than one point until the group shares a reading of the descriptions. This takes 30 minutes and saves hours of argument later.
• Lunch check: compare the average morning score of each panel. If one panel sits a full point above the others, it is probably not seeing better startups, because the draw spread teams evenly. More likely, it is using a different scale. Talk about it before the afternoon block.
• Score before you speak: in every slot, each juror locks their scores before the two jurors exchange a word. Discussion first anchors the panel to whoever talks first.

Remind jurors of the MBA finding at lunch as well. Teams are scored against the descriptions, never against the team before.

Deliberate once and plan the rejections

With scores from two panels per team, the deliberation can be short and focused. Sort the finalists by their combined score and discuss only the band around the cut line. For a cohort of ten, that might be places 7 to 13. Teams far above or below the line do not need another hour of debate.

Keep cohort fit as a separate, explicit step after the individual scores. Sector overlap, mentor capacity or a second team with the same customer are legitimate reasons. But they belong in a visible second round, not hidden in the individual scores. For every team on the cut line, write one sentence that states why it is in or out. When the partner asks in March why team 14 is not in the cohort, that sentence is your answer.

Then plan the rejections, because most of the teams you interviewed will hear no. Y Combinator gives interviewed teams that are not selected feedback by email and notes that many reapply the following batch and get accepted (YC Interview Guide, 2026). The timbuktoo panel gave every startup feedback on pitch day and kept those outside the top 20 in its alumni community (UNDP, March 2026). Tell every team at the end of its slot when it will hear, keep that date, and give rejected finalists two or three sentences from their score notes. Today’s rejected team is next year’s applicant. It talks to other founders.

The accepted teams move straight into programme planning. The cohort you choose on this day becomes the cohort brief you write at T minus twelve weeks in our demo day playbook. A careful selection pays off again when your founders pitch to investors.

Solution: Converve schedules selection interviews the way it schedules B2B meetings. You define the slots, which jurors may meet which startups and the exclusions for conflicts of interest. The meeting matrix then builds a conflict-free timetable that jurors and founders see in their personal schedule. Changes on the day are logged, so you can later show which jurors saw which team. Our solution for startup and investor events shows how the same matrix runs founder and investor programmes.

Frequently Asked Questions

What is an accelerator selection day?

It is the final stage of an accelerator application process. Shortlisted teams pitch briefly and are interviewed by a panel of investors, mentors and programme staff, and the programme decides who joins the next cohort. Research on ten accelerators found juries of 3 to 40 members (Entrepreneurship Research Journal, 2021).

How long should an accelerator interview be?

Short formats work if they are structured. Y Combinator interviews last 10 minutes with two or three partners (YC Interview Guide, 2026). For an in-person selection day, a 30-minute slot with a 5-minute pitch, about 12 minutes of questions and time for silent scoring is a workable standard.

How many jurors should interview each startup?

At least two panels of two jurors, so every decision rests on four independent judgements. In a replication of NIH grant review, 43 reviewers showed no agreement on the same 25 applications. The outcome depended more on the assigned reviewer than on the application (Pier et al., PNAS, 2018).

What criteria do accelerators use to select startups?

Most accelerators assess two things: the opportunity, often judged by a working prototype and a clearly identifiable market, and the team, judged on commitment, coachability, market knowledge, preparedness and how well the founders work together. The team usually carries the decision, while financial projections are rarely used at this stage (Beyhan, Akçomak and Cetindamar, 2021).

Should rejected startups get feedback?

Yes, at least the finalists you interviewed. Y Combinator sends feedback by email to interviewed teams it does not select, and many of them get in on a later application (YC Interview Guide, 2026). Two or three sentences from the score notes are enough.

Conclusion: fix the timetable, then trust the scores

A selection day is a scheduling problem before it is a judgement problem. Two panels per finalist, conflict exclusions, a drawn running order, one question set and a calibrated scale do more for a fair cohort than another criterion. Get the timetable right, and the scores start to mean what you need them to mean. You can then defend the cohort to your partner one sentence at a time.

If you want your next selection day to run on a matrix you can show to anyone who asks, talk to the Converve team.