Completing the Interview

Once the AI has covered all topics — or you feel everything important has been captured — you finish the interview explicitly to trigger deliverable generation.

Three ways to finish

• Finish & generate (shortcut): click the "Finish & generate" button at the top of the interview at any time. It skips the remaining topics and generates deliverables from whatever has been captured so far — you don't need to answer every topic for a useful result.
• Answer everything: when all topics are covered, the progress bar fills, the AI tells you all topics are covered, and a "Complete interview & generate" button appears under the recap. A long recap scrolls inside the interview card, the same way a long note above any question does, so that button stays on screen — you never have to scroll the whole page to reach it.
• Ask in chat: type "end the interview" and the AI wraps up and presents the completion button.

If every topic reads as covered but the button hasn't appeared, a notice above the answer box explains what is still holding it: something the organizer asked for that hasn't been answered yet, or a difference between an uploaded document and your answers that the AI left open. Tell the AI which is right (or who would know) and the button appears.

When the interview is about a process already in your library, a difference between that documentation and what you described has to be classified before the complete button appears. The interviewer asks, one difference at a time, which it is: the documentation is out of date, this is how your region, team, site, or category of work does it, or it is your own workaround. If you said up front that your scope runs the process differently, it does not ask. Finish & generate still skips the remaining questions; the difference is kept either way.

Important: Deliverables only generate when you click the button — simply chatting or closing the browser will not trigger generation. Generation usually takes a few minutes; you can close the browser and come back, and you'll get an email when your deliverables are ready. If generation shows an error, click Try again; if it keeps failing, add more detail in the conversation first, then try again.

The completion screen

Once documents are generated, the completion screen is organized into tabs — Summary, Deliverables, Transcript, Insights, and (when available) Recording and Documents, plus People and Processes.

The Deliverables tab leads with the Library process this interview feeds: its name, whether it is Published · vN, a Draft or has No deliverable yet, and SOP / BPMN / flowchart cards that open the process's documents on the process page (the same ones the Library and Chat use). The cards show the version this interview stands behind — the draft it added when that is the process's newest version, otherwise the published version — so what you open is labelled with its version number, and you edit it there (see The process page in the Processes section: only the latest version can be edited, and a process change becomes the next version to review). When the version an interview produced is no longer the newest, the block is marked Stale · vN and explains which version replaced it, when, and why (the newer version's change note and whether it is published or a draft); the cards then open the latest version, and a Read vN as it was link opens the older one read-only. An interview that feeds two processes gets a block for each; a process that has been archived (no longer surfaced by Architect and never published — the Library hides it) gets none, so a duplicate retired in favour of another process does not show up here as a draft to publish. Below, As recorded in this interview folds away the documents the AI wrote from this recording alone — the record the process was built from. They are read-only and downloadable; it is open by default only until the process has a document of its own.

The People tab lists the people this interview surfaced — your own role and responsibilities plus any colleagues you mentioned — each with a short quote showing where the detail came from. It's a read-only summary that org admins review and save to the People directory.

Completing also teaches the Initiative's Learned from interviews document in the Library: the abbreviations, system names, roles, sites, rules and durable facts your interview established are added to it so the next interview on this Initiative doesn't ask what they mean, and Architect reads them the way you used them. See Written documents under Processes for how to read or correct it.

If a reviewer asks you for more, an "Asked of you" card appears at the top of the Insights tab. Click "Answer in the interview" to drop their request into the chat and answer it in your own words — nothing you already recorded is lost; if your answer changes the process, the interviewer reopens the interview so you can finish it again. Click "Mark handled" when it's done.

Chatting after completion

You can keep chatting with the interviewer after completion to ask questions about the process or make a correction. The interviewer can't edit the documents itself — for rewording or reformatting it will point you to the process page's Edit → AI Edit tools. But if you tell it something about the process is actually different (a step is wrong, missing, or out of order), it treats that as a correction and may reopen the interview so you can finish it again with the correction in it.

Reviewing a finished interview (organizers and admins)

If you sent the invitation — or you're an org or Initiative admin — open the finished session's Insights tab. At the top you get a Reviewer view block the interviewee doesn't see:

• The interview's quality score out of 100, with a one-line explanation for each of the five things it's judged on (depth, clarity, evidence, consistency, completeness) and a score per topic.
• Add context — a note recorded against this interview with your name on it, for something the interviewee couldn't know ("we already license that connector"). Adding one marks the deliverables Needs update; the next time this interview's documents are generated (an interview that is reopened and finished again) it is included as a named reviewer correction. It never edits the transcript and is never written as something the interviewee said. Since Architect reads the process's SOP, put what the note changes into the process's latest version from its page if you want it to reach the next analysis run. A note keeps applying to every future generation until you click "Withdraw" on it.
• Ask for more — send the interview back to the person who gave it:
  1. Next to any "Still to clarify" gap, click "Ask for this" (the gap is pre-written as a question), or write your own request.
  2. Click "Send back". The deliverables are marked Needs update.
  3. They get an email and an "Asked of you" card on their Insights tab, click "Answer in the interview", and answer in their own words.
  4. Either of you can click "Mark handled"; only a reviewer can reopen it. Once the answer is in, take what it changes into the process's latest version from the process page.

A reviewer never types on the interviewee's behalf — a send-back is a request, and the answer always comes from the person who does the work.