What routing decides
Inbound routing decides three things for every call that reaches one of your numbers: whether a Finn answers, which Finn answers, and what happens when the Finn cannot finish the job alone.
You wire the first two in the dashboard under Settings → Phone numbers → [Number] → Default Inbound Finn. There is no API for this today. A number with no Default Inbound Finn set will not be answered by a Finn, which is the single most common reason inbound calls fall through to voicemail. See phone-numbers for provisioning and telephony-overview for how numbers reach the platform.
The three inputs you can route on
| Input | What it is | Typical use |
|---|---|---|
| Time of day | The caller's arrival time against a business-hours window | Different Finn (or human) during staffed hours vs after hours |
| Dialed number | Which of your numbers the caller dialed | Sales line and support line answered by different Finns |
| Caller ID | The number the call came from | Known customers routed differently from unknown callers |
Routing depth is the same as a keypad IVR. The difference is that a Finn routes on stated intent rather than on menu numbers, so callers do not have to guess which branch holds their problem.
Business hours
A business-hours window is a local-time range. Calls inside the window and calls outside it can be handled by different configurations.
The two patterns that work:
- Humans in hours, Finn after hours. Your existing line stays as-is during staffed hours. Outside the window, the number routes to a Finn. This is the after-hours overflow case, and in practice 5-15% of those calls still require a human call-back.
- Finn always, humans as escalation. The Finn answers every call and transfers when the conversation needs a person. For a 24/7 receptionist Finn, expect 15-30% of calls to transfer to a human. Tier-1 support runs higher, 25-40%.
Those percentages are what the same use cases produce across accounts, not a guarantee for yours. Measure your own rate in call-outcomes before you staff around a number.
Business hours are a local-time concept. If you serve callers across time zones from one number, a single window will be wrong for someone. Decide whose local time the window is expressed in and write it down, because nothing in the call log will tell you later.
Overflow
Overflow is what happens when calls arrive faster than you can answer them. Two limits bite, and they bite in different places.
The first is your account's concurrency limit. Calls beyond it are not queued indefinitely by the platform. See concurrency for the limit on your plan and how to read current usage.
The second is your transfer destination. A Finn can hold many calls at once. The human cell phone or office line you transfer to cannot. If the Finn escalates 30% of a spike, the destination absorbs the spike, not the Finn.
Fallback
Fallback is what happens when the intended handler does not pick up. This is the part most setups skip, and it is the part callers notice.
| Failure | What the caller hears without fallback | What to configure |
|---|---|---|
| Transfer destination does not answer | Ringing, then dead air or a dropped call | Fallback to voicemail, or back to the Finn |
| Transfer destination is busy | Busy tone | Same, plus a second destination if you have one |
| No Default Inbound Finn on the number | Straight to voicemail, silently | Set the Default Inbound Finn |
| Finn cannot answer the question | A loop, or a guess | Knowledge base coverage plus a human path |
Configure a fallback for every transfer destination. See tools-transfer for how the transfer itself is set up and what the Finn says while it is happening.
What will go wrong
Calls go to voicemail when you expected a Finn. Almost always a routing misconfiguration rather than a telephony fault. Check the Default Inbound Finn on the dialed number first, then the business-hours window, then whether the call reached the platform at all. troubleshooting covers the number-level checks.
Knowledge base gaps produce loops. A Finn that cannot find an answer will circle or invent one. Before you route real volume, audit coverage for the top 20 questions callers actually ask. See agents-knowledge-base.
Some callers want a human immediately. A share of callers hang up as soon as they realize they are talking to AI. A softer welcome message and an explicit "I'd like a human" path recover most of them. Do not bury the escalation behind three turns of qualifying questions.
Transfers fail quietly. A failed transfer looks like a completed call in aggregate metrics. Check transfer outcomes specifically in call-logs, and subscribe to call events via webhooks if you need to alert on them.
Next
- quickstart-inbound for a number answering calls end to end.
- phone-numbers for buying numbers or pointing an existing one at the platform.
- byoc-sip if the number stays with your carrier.