A call goes out normally, connects, rings, and shows up on the handset under a warning your recipient reads before deciding whether to answer. Nothing in your own systems looks wrong, because from your side nothing was.
Labelling happens at the far end after your provider has finished its part, and a company you have no relationship with applies it.
Who applies the label?
The carrier that delivers the call to the handset, usually through an analytics partner.
The Federal Communications Commission (FCC) makes providers responsible for analytics work they give to another company. In its order in volume 86 of the Federal Register (FR), page 17726, published 6 April 2021, the Commission says it "expects voice service providers to satisfy those requirements, whether or not they use a third party" and "declines to allow voice service providers to avoid liability solely because they make use of a third-party" analytics service. The division is a commercial one, and the responsibility is not divided with it.
This is why the instinct to raise a label with your own provider mostly fails. Your provider originated the call. The label was written by the terminating provider's analytics against a number, and there is no field on your side to unset. What your provider controls is what it can assert about you on the way out, which is a different lever and the subject of what is STIR/SHAKEN.
To understand what Bird can control, use the caller IDs guide. It lists what happens to a number after it leaves the platform and names "Spam and fraud filtering applied by terminating operators" and "Do-not-originate lists and regulator-mandated blocking" among the things outside a carrier's control, alongside anti-spoofing frameworks and origin-based restrictions.
Was my call labelled, or was it blocked?
Different outcomes, and your own records tell them apart.
A blocked call does not connect, and it says so. The FCC requires a terminating provider that blocks a call to "immediately notify the caller that the call has been blocked by sending either a Session Initiation Protocol (SIP) or ISDN User Part (ISUP) response code, as appropriate", and it requires every provider in the path to carry that code back to the origin. So a block arrives at your end as a failed call with a response code attached. What do SIP response codes mean covers reading one, and the call log is where the code and the outcome sit together.
A labelled call connects. It rings, the record shows a normal call, and if nobody answers it reads as an unanswered call rather than a rejected one. There is no code, no rejection reason and no event. The only reliable evidence is a phone in the destination country, which is why labelling is usually discovered by a customer rather than by monitoring.
Separate the two outcomes because they have different causes and different amounts of recourse.
What recourse do I have?
For a block, a defined process. For a label, considerably less, and that was a deliberate choice.
The same FCC order sets out what a blocking provider owes a caller: immediate notification with a response code, a list of blocked calls disclosed to subscribers on request, and, "when a calling party disputes whether blocking its calls is appropriate", a requirement that the provider "provide a status update to the party that filed the dispute within 24 hours".
The Commission created no corresponding redress process for labelling and instead encouraged providers and analytics partners to work with callers. Under the heading "No Redress Requirements for Labeling", it wrote:
The Commission declines to extend redress mechanisms to erroneous call labeling at this time. Rather, the Commission encourages voice service providers and their analytics partners to work in good faith with callers to avoid erroneous labeling so consumers can better decide whether to answer a call.
So the asymmetry you are running into is real and intentional: a wrongly blocked call has a clock attached to it, and a wrongly labelled one has an encouragement.
Each terminating provider's blocking-dispute contact must also handle callers trying to verify their calls after caller ID authentication information affected them. The Commission "requires that the point of contact which terminating voice service providers have established to handle blocking disputes also handle contacts from callers that are adversely affected by information provided by caller ID authentication seeking to verify the authenticity of their calls". That contact exists at each terminating provider, it is obliged to take a call about authentication information, and it is the closest thing to a defined path when a label is following a number rather than a campaign.
What actually changes the outcome?
The two things that are yours: what your provider can vouch for, and what your traffic looks like from the outside.
What can be attested. Attestation is one of the inputs the standards intend for this exact purpose, described in the FCC's earlier order as information to "enhance the spam identification solutions that terminating voice service providers enable for their customers". The level a provider can assert turns on whether it can confirm both the caller and the caller's right to the number, so a number the carrying provider issued sits differently from one you own elsewhere and present through it. Bird's own guidance is to prefer numbers procured through the platform for exactly this reason, and to verify the ones you do not.
What the traffic looks like. Analytics act on a number's behaviour, and the behaviour is yours. A number that dials a great many recipients who do not answer looks different from one that has conversations, and a number nobody can call back looks different from one that rings a real desk. Neither of those is a rule any provider publishes, which cuts both ways: there is no threshold to stay under, and no configuration to change once a number has a reputation.
A labelled number is difficult to rehabilitate and easy to replace, so the defence is spread rather than repair. Keep the numbers you present tied to a real return path, and treat a rising unanswered rate as the early symptom rather than the eventual complaint. The call log is where that shows up first, and what is a call detail record covers reading answer rates without double-counting the legs of a forwarded call.
In short
The label belongs to the receiving carrier, not to yours.
It is applied by the provider terminating the call and its analytics partner. The provider that carried the call has no field to unset.
Blocked and labelled are different outcomes with different evidence.
A blocked call comes back with a SIP or ISDN User Part (ISUP) response code. A labelled call connects and rings, so your own logs show it as a normal call.
The FCC required redress for blocking and declined it for labelling.
Blocking carries notification, a list on request and a status update within 24 hours. For an erroneous label the Commission asked for good faith instead.
There is one contact point the rules do reach.
The blocking dispute contact also has to handle callers adversely affected by caller ID authentication information, which is the nearest thing to a labelling route.