The moment a customer submits a feature request is the best moment to improve it: they're still there, and they know what they meant. Kai PM has three automations that act right then, before the request reaches your queue. All three are under Feedback portal → Kai PM → Other settings, and all three only act on requests customers submit themselves.
Thin requests score badly for the wrong reason. "Export please" could be a spreadsheet or an API; Kai can't judge the effort or the fit, and neither could you.
With Ask for more context on, Kai holds a request that's too thin to judge and asks the customer in the request's own conversation:
It would be great if I could export my data.
Happy to pass this on! Which data would you export, and how often would you need it?
The customer sees the question in the widget and answers there. Kai asks at most two questions, thanks them, and then scores the request. If nobody answers within 24 hours, the request is released and scored on what's there.
While held, a request isn't on the public portal and isn't scored. Your team sees it on the overview marked "Kai asked for details", and can accept or decline it at any point.
Every open roadmap collects the same idea five times in different words, and the votes scatter. Kai compares each new request against the existing board. When it finds a clear match, the new request is merged into the existing one, the customer is told which request now carries their vote, and the vote moves with it.
Kai merges only when it's confident. Near-misses stay separate for you to merge in chat later. This runs as part of Let Kai decide; see that article.
Some requests are for features you shipped last year. Tell customers when it already exists lets Kai check a new request against your connected code repository and, when the feature exists, reply with where to find it:
Could you add a way to get my reports by email?
Good news: this already exists! You can turn on email reports in your notification settings.
The request is closed as released. If the customer replies that this isn't what they meant, it reopens and goes through normal scoring. No reply within a week and it stays closed.
This needs a code repository connected under AI & Automation → Knowledge; without one the setting does nothing. It's off by default because a wrong "it already exists" is more annoying than a wrong score. Turn it on once you've seen a few effort estimates you agree with.
Requests your team creates from a conversation, imports, or requests created by agents skip all three. They're already yours; Kai scores them and stops.
