Key takeaways
- Planned for the first release, not part of the pilot. Details may change after design partner pilots.
- One live call in two parts: a screen-share pitch of about 5 minutes, then spoken questions from an AI interviewer.
- Camera, microphone and screen recording each need their own consent before the call starts.
- The AI assesses only the transcript. People at your fund watch the recording.
- A text interview with the same questions and the same rubric is always available.
Planned feature: not in Delsivo yet, and details may change before it ships.
What is an AI interview for startup founders?
It replaces the first screening call for founders you have already shortlisted. Nobody from your fund needs to be on the call. The founder opens a link on your branded portal whenever suits them inside the invitation window, pitches with their screen shared, and then answers questions from an AI interviewer that speaks in a neutral voice with live captions. Your team watches the recording later, with a transcript and citations beside each answer.
The media runs over WebRTC, the technology browsers use for video calls, and camera, microphone and screen are recorded on the server, so a crashed browser loses nothing already sent. The interview is a step inside the deal record, not a separate tool. It starts from the deal, uses your rubric and the deck facts already extracted, and its findings land next to the claims and the thesis fit report.
How the live interview works for a founder
- 1An analyst sends the interview from the deal. If your fund requires partner approval for outbound messages, a partner approves it first.
- 2The founder gets a branded email with a link to your portal. The link expires after a period you set (7 days by default), with reminders before it does.
- 3Before the call, one screen explains the purpose, the total length and the two parts, how answers are assessed and who sees the recording for how long. It asks for separate consent for camera, microphone and screen recording, warns that a shared screen can show notifications, and links to the text interview.
- 4A device check shows the camera preview, a microphone level, a test screen share and a connection test. Screen sharing needs a desktop browser, so a founder on a phone is asked to switch device or take the text interview.
- 5Part one is the pitch. The founder shares their screen, usually the deck, with the camera on and pitches for about 5 minutes (your fund can set 3 to 10). A timer shows the time left, and the AI stays silent and listens.
- 6Part two is the questions. The AI interviewer asks your core questions in your order, then follow-ups drawn from the pitch, the deck and open claims, at most one per question. The founder answers out loud, can keep sharing the screen to show the product, and can ask for a question to be repeated.
- 7The interviewer closes politely one minute before the cap you set, and the call stops at the cap. The founder gets a receipt, and later the timestamped transcript, where they can flag transcription errors. Founders never see scores.
Live AI interview or text interview: what changes
The format changes how the founder answers, not how the answer is judged. Your rubric, core questions and follow-up limits are the same in both.
| Live AI interview | Text interview | |
|---|---|---|
| Format | One call: a screen-share pitch, then spoken questions | A written pitch summary, then the questions in a chat |
| Consent | Separate ticks for camera, microphone and screen recording | No recording beyond the written answers |
| Time | A cap your fund sets, usually 15 to 20 minutes, with a visible timer | The same time guidance, with a word limit for the pitch |
| Device | A desktop browser with a camera and microphone | Any browser, including a phone |
| What your team reviews | The recording with a bookmark per question and a synced transcript | The written answers |
| What the AI assesses | The transcript, for substance and evidence | The written answers, for substance and evidence |
A founder who declines recording, has no quiet room or simply prefers to write keeps the text interview. Nobody is pushed onto camera to stay in the process.
What is never assessed
The AI assesses only the transcript. It never sees the camera or the screen recording, and accent, fluency, presentation polish, appearance and background are excluded from the assessment. Speech measures such as words per minute or filler words are not computed for investors at all. The founder is told this before the call.
The recordings are for people at your fund, who can watch the pitch and the answers. Every founder on the same interview template gets the same core questions, the same interviewer persona and the same time plan. How answers are scored is covered on the interview scoring rubric page.
When things go wrong
- Dropped connection: the founder rejoins the same session within 10 minutes and continues where it stopped, in one interview record.
- Technical failure: one retry by default, allowed only after a failure the system recorded, never to improve answers.
- A strict network blocks the live call: the browser records locally and uploads in short chunks, and the questions run turn by turn.
- Expired link: your analyst can extend it or send a fresh one from the deal.
- A founder can't take an online interview: an analyst can switch to a manual path and record notes from a live call, with a consented transcript if there is one.
Your fund sets how long camera, screen and the transcript are kept, each on its own schedule. Withdrawing consent deletes the recordings.