The voice stack’s real bottleneck is turn-taking
For builders evaluating real-time voice AI, latency is a chain problem: recognition, endpointing, orchestration and speech output must work together.
Use XBert to collect tenant maintenance details across voice and text, then keep urgent decisions and work-order routing tied to your existing process.
After-hours maintenance calls often start with an incomplete report: a leak, a broken lock, no heat. Staff then have to call back for the unit number, assess urgency, and enter the request somewhere else. A voice-and-SMS intake flow can collect those basics before a person takes over. It can reduce avoidable back-and-forth, but it should not make safety decisions on its own.
XBert, developed by Nextiva, handles business conversations across voice and text, among other channels. Its AI Receptionist and AI Employee capabilities can support an intake workflow, while its Agent Assist capabilities help human teams with coaching, transcript summaries, and drafted responses. The steps below describe how to plan that workflow. Specific routing depends on which of your business tools XBert can connect to and what those integrations support.
Write down your existing maintenance priorities before automating intake. Separate immediate danger from urgent repairs and routine requests. Use your organization’s own definitions and escalation process; do not ask the AI to invent a priority or decide whether a situation is safe.
For example, your intake plan might direct a tenant reporting active flooding, a suspected gas leak, or a person at immediate risk to the contact or emergency procedure your organization has designated. A failed appliance with no immediate hazard may follow a different path. Those are examples, not a substitute for your local procedures, lease obligations, or emergency guidance.
Set a clear boundary: XBert can gather the report and follow the agreed handoff path, but a person or established emergency service handles situations requiring judgment. Tell tenants what to do if they believe someone is in immediate danger, rather than making them wait for an automated conversation to finish.
Keep the questions consistent across calls and texts. Ask for the tenant’s name, property and unit, a callback number, the affected area, what happened, when it started, and whether the problem is getting worse. For urgent reports, ask focused follow-ups that clarify observable facts: Is water actively spreading? Is anyone unable to leave safely? Avoid leading tenants toward a diagnosis.
Ask whether the tenant can safely take a simple protective step only if that advice is already approved by your organization. Do not have an automated agent improvise repair instructions. Close by confirming what was reported, what happens next, and whether a staff member will follow up.
This checklist is the content your team should define and validate. Do not assume an existing product setting will ask each question in precisely this order. Review a demo or trial against your real scenarios before relying on the flow.
For a tenant maintenance call, the AI Receptionist can be part of the voice intake. For a tenant who prefers to text, XBert supports text conversations as well. Use the same core checklist in both channels so that a caller and a texter are not asked for completely different information.
Keep the conversation brief. Capture the issue and the location early, then ask only the follow-ups needed to understand the request and apply your escalation plan. If a tenant cannot provide a detail, record that it is unknown rather than filling in a guess. If the conversation needs a person, make the handoff path explicit and tell the tenant what to expect.
XBert’s NEXT.IQ architecture includes a Knowledge Layer, Customer Journal Layer, and Guardian Layer. The product describes the Knowledge Layer as grounding answers in sources, the Customer Journal Layer as keeping interactions in a continuous customer story, and the Guardian Layer as providing guardrails and permissions. For maintenance intake, have your team review what information and approved guidance the workflow should use, and check how the connected systems handle records and permissions.
List the destination for each type of request before launch: for example, the system your team uses to manage maintenance, an on-call contact, or a human review queue. XBert integrates with CRMs, calendars, and other business tools, but that does not establish that a particular property management system is supported or that a specific work-order action is available. Confirm the connection and the exact data flow with XBert before promising automatic work-order creation.
If a supported integration can pass the request into your system, agree on the fields your team needs: tenant and unit, issue description, reported time, urgency indicators, and contact details. Test whether missing information stays visibly missing, whether a human can correct the record, and whether the tenant’s follow-up instructions are preserved. If the system cannot receive the request directly, keep a staff review step rather than treating an unverified handoff as a completed work order.
Run sample calls and texts for routine repairs, incomplete reports, repeat issues, and urgent scenarios. Have staff compare the resulting conversation record with what they would need to act. Check that the timeline is understandable and that a person can see the reported facts without treating an AI summary as a substitute for the original details.
Set named owners for reviewing escalations and checking requests that did not reach the intended destination. Start with a limited rollout and review misses, incorrect summaries, and unnecessary escalations before expanding it. For a closer look at human handoffs, see how to plan escalation rules that carry conversation context to a live agent.
The useful outcome is not simply fewer calls reaching staff. It is a cleaner first report, a traceable conversation history, and a clear route to a person when the situation calls for one. Measure those outcomes against your current process, and keep emergency judgment with the people and procedures responsible for it.
For builders evaluating real-time voice AI, latency is a chain problem: recognition, endpointing, orchestration and speech output must work together.
Use after-hours call routing, clear intake rules and a staffed escalation path to keep callers moving when your team is away.
A paid search lead is only valuable if someone answers, captures the need and moves the prospect to a clear next step.