Why Your Calls Show Spam Likely on Caller ID

An agent buys twenty fresh numbers on Monday. By Thursday, four of them are showing “Spam Likely” on Android and going straight to Silenced Unknown Callers on iPhone. He buys twenty more. Same result, faster.

This is the single most expensive misunderstanding in phone-based insurance sales, and it comes from a wrong mental model. The label is not a punishment attached to a number. It is a score attached to a calling pattern, and the number is only one of its inputs. Rotate the number without changing the pattern and you have bought yourself four days.

Who actually applies the label

Not your dialer vendor. Not the carrier you buy numbers from. Three analytics engines do the overwhelming majority of the labelling in North America, and the mobile carriers license from them:

• First Orion — powers T-Mobile and Metro’s Scam Shield labelling
• Hiya — powers AT&T Call Protect and Samsung’s built-in caller ID
• TNS (Transaction Network Services) — powers Verizon Call Filter and publishes a twice-yearly Robocall Investigation Report on how the ecosystem is behaving

Each runs its own model. This is why a number can be clean on Verizon and labelled on T-Mobile the same afternoon — three scores, three thresholds, no coordination. There is no central registry to appeal to, which is the practical reason “just get it removed” is not a strategy.

STIR/SHAKEN sits underneath all of it. Attestation levels (A, B, C) travel with your call and tell the terminating carrier how confident the originating carrier is that you are entitled to the number you are displaying. The FCC’s call authentication framework is the governing scheme. Full A-level attestation does not exempt you from labelling — it removes one reason to be labelled. Analytics still score behaviour on top of it.

What the models actually measure

Vendors do not publish their weights, but the observable inputs are consistent across all three engines and consistent with what the FCC’s consumer guidance on call blocking describes:

Velocity. Calls per number per hour. A number that places 400 calls in an afternoon does not look like a person. Human-plausible volume per number is the strongest single lever you have.

Answer rate and duration. A number whose calls are answered and produce 4-minute conversations scores well. A number whose calls are declined at 92% and average 6 seconds scores badly. Note the feedback loop: once labelled, your answer rate collapses, which drives the score down further. Labelling is self-reinforcing, and this is why a burnt number rarely recovers on its own.

Complaint signal. Explicit “Report spam” taps, and implicit signals — immediate hangups, blocks added by the recipient.

Number reputation history. Reassigned numbers carry their predecessor’s baggage. A cheap DID from a pool that a debt collector burned last quarter arrives pre-damaged.

Pattern signature. Sequential dialing through a number block, identical call durations, calling at machine-regular intervals.

Why buying more numbers is a treadmill

Because velocity is per-number, splitting 4,000 calls across twenty numbers does lower per-number velocity — that part works. What does not work is doing it without fixing answer rate, because the other inputs travel with the calling pattern, not the DID. Twenty numbers dialing a burnt list at a 4% answer rate produce twenty numbers with a 4% answer rate.

There is also a hard ceiling on this tactic. Aggressive number rotation is itself a pattern the engines detect, and a large block of numbers with correlated behaviour and no call history is a recognisable signature. You are not hiding; you are announcing.

Rotation is a hygiene practice, not a fix. It buys headroom for a calling pattern that is otherwise healthy. It cannot rescue one that is not.

The three levers you control

1. Volume per number. Cap it. A defensible working range for insurance sales is roughly 60–100 calls per number per day, spread across hours rather than fired in a block. That is a business decision about how many numbers you provision, and it is the lever with the most immediate effect.

2. Answer rate — which means list quality. This is the lever agents skip because it is the expensive one. Dialing aged, worked-out, wrong-number-heavy data guarantees a bad score no matter how the numbers are arranged. Cleaning the list, checking it against the vendor audit framework, and dropping sources whose contact rate is below threshold does more for caller-ID reputation than any dialer setting. It also, not coincidentally, does more for your commissions.

3. Registration and attestation. Register your numbers in the free reputation portals the analytics vendors operate — First Orion, Hiya and TNS each accept business registration that associates a number with a verified legal entity and a display name. Confirm with your provider that your traffic gets A-level attestation, which requires that the provider has verified you are entitled to the number. Branded calling (your business name rendered on the handset instead of a bare number) is a paid layer on top and measurably lifts answer rate, which then feeds lever two.

None of the three is a submission form that clears a label overnight. All three change the score that produces the label.

The remediation sequence that actually works

1. Check every number across all three engines before assuming a problem is universal. Free lookup tools from each vendor report their own status; a number clean on two of three needs a different response than one burnt on all three.
2. Rest the burnt numbers. Not rotate — rest. Take them out of rotation entirely for 30+ days. Score decay is real but slow.
3. Fix the pattern before reintroducing capacity. New numbers into an unchanged pattern reproduce the problem on a two-week lag.
4. Register and attest everything currently in rotation.
5. Monitor continuously. Reputation is not a state you achieve; it is a number that moves daily. Shops that check monthly find out about a burnt number three weeks after their connect rate fell.

What this costs you if you ignore it

A labelled number does not fail loudly. Your dials still go out. Your dashboard still shows activity. The number that moves is answer rate, and it moves gradually enough that most agents attribute it to the list, the season, or the market. Run the delta through the dialer ROI arithmetic and a drop from 12% to 7% answer rate is not a 5-point problem — it is a 42% cut to every downstream number, including issued policies.

Related reading: local presence dialing, which interacts with all of this, and connect-rate benchmarks for what “normal” actually looks like before you diagnose a problem.