Key takeaways
- Founder interviews are a first release feature, not part of the pilot.
- Nobody from your fund has to be on the call, and the founder chooses when to take it.
- Questions are your fixed set plus follow-ups drawn from the pitch, the deck and the deal's open claims.
- The AI never sees video or the screen. Accent, fluency, polish, appearance and background are excluded.
- The partner call doesn't disappear. It moves to after the interview, for founders you already want to meet.
Planned feature: not in Delsivo yet, and details may change before it ships.
What a first screening call is really for
A first call usually does three jobs. It fills gaps the deck left open, such as round size or who is paying today. It lets a partner hear how a founder explains the problem. And it decides whether the deal deserves more partner time. All three matter, but the call is an expensive way to do the first one, and it varies with who takes it, on which day, in which time zone.
A recorded interview with the same questions for every selected founder can cover the first two jobs on the same terms, so the live conversation with a partner is spent on deals where the third job is already half done.
When to replace the first call with an AI interview, and when not to
| Situation | AI interview | Partner call |
|---|---|---|
| Many applicants at the same stage | Same questions and time plan for all, taken when it suits them | Hard to keep consistent across reviewers |
| Founder in a distant time zone | No scheduling across time zones | Someone takes a late or early call |
| Deck leaves basic questions open | Follow-ups are drawn from the unknowns and open claims | Fine, but uses partner time on basics |
| A founder you already know well | Usually unnecessary | Better |
| A competitive round moving fast | May slow things down | Better |
| The founder asks for a conversation | Offer both, plus the text interview | Your team can always run the call instead |
How the interview works for the founder
- 1The founder gets an invitation in your fund's brand, with an expiry date and reminders before it lapses.
- 2Before the call, one screen states the purpose, the length and the two parts, what is recorded and who sees it, and asks for separate consent for camera, microphone and screen.
- 3A device check runs on a desktop browser. A founder on a phone, or one who declines recording, can take the text interview with the same questions and rubric.
- 4The founder shares their screen with the camera on and pitches for about 5 minutes while the AI listens.
- 5The AI interviewer asks your fixed questions out loud, then follow-ups based on the pitch, the deck and open claims, and closes before the cap you set.
- 6If the connection drops, the founder rejoins within 10 minutes and continues. Afterwards they can flag a transcription error, kept next to the original line.
What your team receives after the interview
- The recording, with bookmarks for the pitch and each question and a view switch between screen and camera.
- A transcript with timestamps, so any assessment can be checked against the exact moment it cites.
- An assessment against your criteria, shown as an AI suggestion with cited excerpts, for an analyst to accept, edit or dismiss.
- New or changed claims, such as a founder clarifying that 12 of 20 customers are unpaid pilots, added to the claim tracker with both sources.
- A manual path, so an analyst can run the interview personally when a founder can't or won't use either online format.
Fairness rules written into the rubric
No interview format removes every source of unfairness, and Delsivo doesn't claim to. What the design does is narrow what is assessed. The AI reads only the transcript and judges the substance of answers and the evidence behind them. It never sees the video or the screen, and accent, fluency, presentation polish, appearance and background are excluded. Speech metrics such as pace and filler words are not computed for investors.
Keeping a text interview under the same rubric matters too. A founder who is more precise in writing, or who has no quiet room for a camera, is not marked down for choosing text.