Accelerator selection looks like venture screening, but the shape is different. A fund sees deals all year and backs a few. An accelerator opens a window, receives most of its applications in the final days, and has to choose a whole cohort at once, often with a pool of volunteer reviewers who each read a slice.
Understanding how accelerators select startups helps on both sides. Founders learn what reviewers look for. Programme managers get a process that holds up when applications triple in the last week. The sections below use what well-known programmes say publicly about their own process, and then set out a practical approach any programme can run.
What large programmes say about their selection
Y Combinator is the most visible example, and it publishes a fair amount about its process. Its FAQ describes an application process open to any startup anywhere in the world, and recommends applying as soon as you have a founding team and an idea you are excited about. It says it is important for the founding team to have the skills to build the product themselves. It also notes that in a typical batch about half the companies applied more than once before being accepted, and that on average 40% of funded companies are just an idea. Its FAQ states that it invests $500,000 in each company.
On interviews, Y Combinator's application page says promising applicants are invited to interview, that most interviews are held by video conference, and that it typically makes decisions the same day as the interview, with feedback for everyone who interviews. Techstars describes its accelerators as three months of intensive, mentorship-driven support for early-stage startups.
Two lessons carry over to smaller programmes. Publish what you look for, and commit to when founders will hear back.
How accelerators select startups, stage by stage
Programmes vary, but most selection processes pass through the same stages. Each one assesses something different and has its own common failure.
| Stage | What is assessed | Who usually does it | Common failure |
|---|---|---|---|
| Eligibility | Stage, sector, location, legal entity, time commitment | Programme team, often from form answers | Long forms before eligibility is checked |
| Written review | Problem, customer, progress, team's ability to build | Reviewers, each reading a share of applications | Reviewers apply different standards |
| Shortlist discussion | Borderline cases and disagreements | Programme team and lead reviewers | Loudest reviewer decides |
| Interview | How founders think, answer follow-ups and divide work | Partners, programme leads or alumni | Judging confidence and polish instead of substance |
| Cohort decision | Individual quality plus the programme's mix and capacity | Selection committee | No record of why companies were chosen |
| Notification | Outcome and, where possible, feedback | Programme team | Late or missing replies to declined founders |
Write criteria reviewers can apply the same way
Volunteer reviewers bring useful experience and very different instincts. Written criteria are what make their scores comparable.
- Separate eligibility from quality. Eligibility rules, such as stage or location, are yes or no and can be checked before anyone reads the essay answers.
- Define four to six quality criteria, such as the team's ability to build, understanding of the customer, progress since starting, and clarity about what the programme would change.
- Describe what a strong, middling and weak answer looks like for each criterion, with examples.
- Score each criterion separately before giving an overall view, so one impressive detail does not carry the whole application.
- Treat missing information as a question, not a zero. An idea-stage team will not have revenue, and that should not count against them if your programme takes idea-stage teams.
Our guide on how to screen startup applications covers triage outcomes and replies in more detail, and the VC application form template can be adapted for a programme.
Run the review and interviews at deadline volume
- 1Recruit and brief reviewers before the window opens, with the criteria and a short calibration exercise.
- 2Calibrate: have every reviewer score the same five applications, compare results and discuss the largest differences.
- 3Assign applications so each is read by at least two reviewers who do not see each other's scores first.
- 4Flag large disagreements for discussion instead of averaging them away.
- 5Use a consistent interview: the same core questions for every team, with follow-ups drawn from their application.
- 6Assess interview answers on substance and evidence. Accent, fluency, presentation polish, appearance and background are not evidence of a strong company.
- 7Keep a short written reason for each final decision, including declines.
Programmes with large volumes sometimes add a short interview that applicants take on their own schedule before the live one, so more teams get a chance to explain themselves without a scheduling burden. Our guide on asynchronous founder interviews covers when that helps and how to design one, and our guide on bias in startup screening covers safeguards for reviewers.
Choose the cohort, then tell every founder
The final step is different from venture screening. A cohort is a group that will spend months together, so programmes weigh the mix as well as each company. Be careful with that. Balancing sectors or stages so that mentors and content fit is reasonable. Using "fit" as an unexamined reason to pass on founders who are unlike past cohorts is not.
Then reply to everyone. Publish the date founders will hear by, and meet it. Declined founders are future applicants, and Y Combinator's own figure, that about half of a typical batch applied more than once, is a reminder that a good decline helps people come back stronger. Our guide on how to say no to founders has wording you can adapt.
Common mistakes at this stage: deciding the cohort in one long meeting without written reasons, letting late-arriving strong applications skip the process, and sending declines weeks after the accepted founders have announced their place.
Where Delsivo fits
Delsivo is designed for venture funds first, and it also suits accelerators. In the pilot release, a programme gets a branded application portal on its own subdomain, with save and resume and a receipt for every founder. Each PDF deck is read with slide citations, applications are checked against the programme's written criteria with results of Met, Mismatch or Unknown, and applications wait in a review queue with owners. A mismatch is surfaced for review, never an automatic rejection.
Live AI interviews are planned for the first release: an applicant shares their screen and pitches for about 5 minutes, then an AI interviewer asks the programme's questions out loud, within a cap the programme sets of 15 to 20 minutes. A text interview is always available. The AI assesses only the transcript, on substance and evidence; accent, fluency, presentation polish, appearance, background and the video and screen images are excluded. Reviewers can override any assessment with a reason, and the programme team decides every place in the cohort. See Delsivo for accelerators and accelerator cohort selection.
Sources
- Y Combinator: Frequently asked questions (checked 6 November 2026)
- Y Combinator: Apply to Y Combinator (checked 6 November 2026)
- Techstars: Accelerators (checked 6 November 2026)
Delsivo publishes this guide and makes software for venture funds, so we have a stake in the topic. Facts about other companies, research and regulations come from the sources listed, on the dates shown, and can change. Corrections are welcome through our contact page.