How-to guides

Run a scoping interview

Describe the project; the AI figures out which cards you need. Or paste a call transcript and skip the questions.


The Interview tab is where you scope a project through a conversation with the AI. You do not need to know which Install Cards to use — describe the project and the AI builds the estimate for you.

Audience level

Before starting, select who is being interviewed using the three-chip selector above the description field:

  • Sales / PM (default) — Questions are phrased entirely in business language. No technical jargon, no equipment counts. The AI asks about coverage areas, outcomes, and physical reality — never "how many access points?" or "what's the PoE budget?"
  • Technical PM — System-level terms are permitted (access points, PoE switch, NVR, VLAN) but low-level configuration details are avoided.
  • Field Engineer — Full technical vocabulary. The AI asks as one engineer would ask another: AP density, PoE budget, VLAN topology, uplink redundancy, and so on.

Your selection is saved to the project and restored if you return to the interview later.

Discovery Agenda — prep before you call the client

Before you even open the interview, click Discovery Agenda above the audience selector to build a call sheet for the discovery conversation itself. Pick a deal type (Video Surveillance, AV Buildout, Network Refresh, Access Control, and more — defaulted from your org's vertical) and adjust the domain chips if the deal spans more than one. Click Generate Agenda and DarkBird assembles two things instantly, with no AI call involved:

  • The questions that matter — pulled from your org's own scoping-question bank, grouped by domain, so you walk into the call knowing what to ask instead of discovering the gaps afterward in the interview.
  • "Watch For" callouts — cards from your calibration history that consistently run over or under estimate, with the reason why (e.g. "Camera installs run 22% over — ceiling type is rarely confirmed up front"). This is your own project history telling you what to double-check before you quote it again.

Use the Copy button to paste the agenda anywhere, or Copy as Granola template to paste it into a note-taking tool before the call — the questions become the literal headings of your meeting notes, so the call and the agenda stay in sync. The generated agenda is saved to the project, so it is still there if you step away and come back.

Running the interview

  1. Set audience level and describe the project

    Choose who is being interviewed, then type a plain-language description of what the client needs. "They need a new network setup with WiFi throughout the office and a firewall" is enough to start.

  2. Answer clarifying questions

    The AI asks follow-up questions to narrow down the right cards. Use the tappable quick-reply chips for common answers, or type your own.

  3. Watch the Proposed Estimate build

    The sidebar on the right shows matched cards updating in real time as the interview progresses. You can see hours and price building as you answer.

  4. Review and confirm

    When the interview is complete, the proposed estimate is ready to review in the Estimate tab.

Or paste a call transcript

If you already have a discovery call transcript or meeting notes — a recorded conversation already contains far more detail than a short description — click Or paste a call transcript below the Start Scoping button instead of typing a description. Paste the full text and click Draft Estimate from Transcript. The AI reads the whole thing, extracts every requirement it can, and goes straight to a full set of recommendations rather than asking questions one at a time — the same recommendation review, Add-card buttons, and Assumptions & Exclusions panel appear either way.

Tip

It only asks a clarifying question if something is a genuine blocker the transcript does not answer. The more detail in the transcript (existing infrastructure state, device counts, anything explicitly excluded), the more confident the resulting estimate.

Quick-reply chips

After each AI question, tappable chips appear with common answer options (e.g. Yes / No, 1-5 / 6-20 / 20+, Drop ceiling / Concrete / Open). Tap a chip to submit that answer instantly. You can always type a different answer if none of the chips fit.

Library match: confidence indicator

A status bar at the top of the interview shows how well your current project matches the card library:

LevelWhat it means
StrongAll requirements have matching templates. Your estimate will be accurate.
PartialMost requirements match. Some items may need manual adjustment after the interview.
Discovery Mode ActiveA service was described that has no matching card. The AI is learning from this conversation.
Tip

When Discovery Mode is active, consider looping in a technical team member before finalizing the estimate, or proceed and flag the unmatched items for review.

Voice mode

Toggle Voice Mode using the microphone button in the interview. When active, AI responses are read aloud ("The Bird Speaks"), you can speak your answers hands-free, and quick-reply chips remain visible as a backup.

Revisiting a previous answer

If you answered a question incorrectly or the client changes their mind, use the Revisit a topic link below the chat input. Select the question you want to go back to — the AI re-asks it naturally and the estimate updates from that point forward. You do not have to start the interview over.

Flagging irrelevant questions (Owner / Admin)

If the AI asks a question that is not relevant to this type of project, Owners and Admins can flag it using the thumbs-down icon that appears to the right of the question text.

  • Immediate effect — The AI receives a note in its session context and will not re-ask the same question for the remainder of this interview.
  • Long-term effect — The dismissal is stored with the surrounding conversation context so you can later teach your Expert Profile not to make the same reasoning error again.
Tip

Example: If a firewall project prompts "How many floors need wireless coverage?" — that question makes sense on a Wi-Fi deployment but is irrelevant when Wi-Fi was only mentioned as a VLAN segment. Flagging it teaches the AI to distinguish between the two contexts.

Flagged questions accumulate in Settings → Expert Center → Interview Feedback. When you have a batch worth reviewing, click Refine from Feedback to update your Expert Profile's reasoning based on the flags.

Quote assumptions & exclusions

At the end of the interview, after the recommendations appear, a two-column Quote Assumptions & Exclusions panel is displayed below the proposed estimate. The AI derives these directly from what the client said during the interview — for example, if the client confirmed no backup internet connection, that becomes an exclusion; if they confirmed new cabling throughout, that becomes an assumption.

  • Assumptions — Conditions the estimate depends on (e.g. "All cabling will be new install — no existing infrastructure to reuse").
  • Out of Scope — Items explicitly not included (e.g. "Redundant internet connection excluded — client confirmed no backup ISP").

Each item is editable: click the × to remove an item you disagree with, or type a new item in the add field at the bottom of either column. Changes save automatically. When you generate a Scope of Work, these items are merged verbatim into the SOW's Assumptions and Out of Scope sections alongside any assumptions/exclusions on the Install Cards themselves.

Tip

Review this panel before generating the SOW — it is your last chance to remove auto-generated items that do not apply or add anything the AI missed.