PlacementFlow's AI does the operational work — sourcing, screening, chasing, scheduling, analysing — while you approve the high-stakes actions. This page is the complete, honest map: what runs automatically, what always waits for you, and the lines the AI never crosses.
| Surface | The AI does | Waits for your approval |
|---|---|---|
| Sourcing | Generates search filters from your brief, sources candidates, drafts outreach copy | Campaigns start as drafts — you rehearse them in the Playground and launch them yourself |
| Screening | Runs the conversational screening chat, asks your non-negotiables first, probes vague answers, scores responses | You review every screening result before a candidate advances; disqualifications sit in a review queue |
| Reply handling | Classifies every reply in seconds, stops follow-ups, handles opt-outs and out-of-office | Counter-offers, spam threats, and anything uncertain route to you — always |
| Scheduling | Proposes slots, creates calendar events and video links, sends reminders, handles reschedule requests | Nothing — scheduling coordination is low-risk and runs automatically once an interview exists |
| Interview analysis | Records (with consent), transcribes, and analyses interviews; surfaces a verdict and evidence | The verdict is a recommendation — you decide what happens next |
| Submissions | Drafts the submission email and follow-up sequence | A submission never sends to a client without your explicit approval — this is capped in the autonomy system and cannot be raised to fully automatic |
| Offers | Pre-fills offer terms from the interview analysis, drafts counter-offer responses for you | Creating and sending every offer; responding to every counter-offer |
How much of the routine work runs without a per-action click depends on your autonomy level — but the caps above hold at every level.
These are not settings. They hold at every autonomy level, for every agency, and cannot be configured away:
We'd rather you know exactly where the edges are:
[SCORING_UNAVAILABLE] warning and review the responses yourself — never a silently fabricated score.Every AI action — proposed, approved, rejected, or executed — is written to a per-decision audit log with a timestamp, the confidence score, and who (human or AI) executed it. You can export the decision history as a CSV from Settings → Compliance, and candidates can download their own AI decision report from the candidate portal. "Trust us" isn't the model; "check us" is — see Auditable AI: Bias Monitoring and the Audit Export.