We start by checking whether a digital human is actually the right interface. If it is, we design what it should say and do, connect it to the right information, and test it with real users. We also plan sensible handover to humans, so nobody gets stuck.