AI assistance
AI replies and tuning
RouteFlex AI can read your inbound candidate emails and texts, write a draft reply, and (when you trust it enough) send the reply automatically. This page explains how to shape that behavior so it matches your voice and your tolerance for risk.
What RouteFlex AI actually does
When a candidate replies to one of your emails or texts, an automation can route that reply to RouteFlex AI. The AI reads the message, looks up the relevant Q&A pairs in your knowledge base, and writes a response in your voice. Depending on how confident it is, it either sends the reply automatically or queues it for a human to review.
The AI never replies about things outside your knowledge base. If a candidate asks something it doesn't have an answer for, it escalates to the contact you set in AI settings.
AI settings, one by one
Open Settings › AI. The page has five things you can tune.
Screenshot
The AI settings page with fields for instructions, escalation contact, confidence threshold, and disclosure.
AI instructions is the prompt that shapes tone and behavior. Tell the AI how to sound (warm, professional, casual), what it can and can't talk about, and how strict to be about staying on topic. If you applied a template, this is filled in; tweak it to match your voice.
Escalation email and phone is the contact info the AI gives candidates when it can't answer their question itself. Use a real recruiter's address, not a no-reply.
Confidence threshold is a number between 0 and 1. When the AI is more confident than this threshold, it sends the reply automatically. Below it, the reply is queued for a human to review. Higher numbers mean more human review and fewer auto-sends.
Candidate disclosure controls whether the AI tells candidates it's an AI. We recommend leaving this on. The disclosure text is editable.
Reviewing drafts
When the AI is below the confidence threshold (or you've set things up to always require review), drafts land in your notifications. Open the draft, edit if needed, and click Send. Edits are silently used to keep improving the AI over time.
When the AI gets something wrong
Usually it's because the knowledge base is missing the answer. Add a new Q&A pair in the Knowledge base section of the product, and the AI will pick it up immediately. If you're seeing tone problems (too formal, too chatty), edit AI instructions: a single short sentence usually does the trick.