Can it check real order or account information?
Yes, where we integrate with the relevant system — that's usually what separates a useful chatbot from a generic FAQ widget.
What happens with questions it can't answer?
It escalates to a human with full conversation context, so the customer doesn't have to repeat themselves.
Is pricing per conversation or per seat?
No — since it's your own codebase, there's no per-seat or per-conversation platform fee scaling against you.
Can it work across our website and other channels?
We scope this based on where your customers actually reach out — website, app, or other channels — rather than assuming one channel fits every business the same way.
How do you keep it from giving wrong answers?
We ground responses in your real product data and policies rather than letting the model answer from general knowledge alone.
Can we add more capabilities to it later?
Yes — because you own the codebase, adding a new capability is a normal development task, not a request to a vendor's roadmap or a wait for a feature release.
Does it support multiple languages?
We scope this based on your customer base during discovery — modern language models handle multiple languages well, so it's usually a configuration decision rather than a major engineering lift.
How quickly does it respond?
Response time depends on the complexity of what it's checking, but for most product and policy questions it's near-instant — the main latency comes from live lookups against external systems, which we optimize during the build.
Can it hand off smoothly if a customer prefers a human from the start?
Yes — we always build a visible, easy path to a human agent rather than forcing every customer through the bot first, since some customers simply prefer that and fighting it hurts the experience.
Can it start a conversation proactively, not just respond?
We can build proactive messaging — for example, following up on an abandoned cart or a stalled signup — as part of the scope, though we're careful to design it so it feels helpful rather than intrusive.