Written for the operations lead sizing a caller ID pool: a sizing method with the reasoning shown, the two failure patterns carrier analytics punish, and the aging, monitoring and retirement policies that keep a pool clean.
How many phone numbers do you need for outbound dialing?
Work it out from peak daily dial volume, not a magic number. Keep daily dials per caller ID modest (in our experience, low hundreds per number per day is already pushing it), then add headroom so numbers can rest. Too few numbers concentrates volume and trips velocity flags; too many with constant rotation looks like snowshoeing, which analytics engines also punish.
Carrier analytics score the shape of your traffic, and both extremes of pool sizing produce a bad shape.
The question usually arrives as a procurement line item, a count of DIDs to put on the order. The more useful framing is what shape of traffic each of those numbers will carry, because the analytics engines that decide whether your calls display as a normal number or as Spam Likely are scoring shape, not intent.
Concentrate too much volume on too few numbers and you trip velocity scoring. One caller ID pushing thousands of dials a day into a single carrier network, most of them short or unanswered, is the easiest pattern in the world to flag, and the engines were catching it long before STIR/SHAKEN existed. That failure mode is well understood, which is why the standard overcorrection exists.
The overcorrection is worse. Buy hundreds of numbers, rotate constantly, keep per-number volume tiny, and you have rebuilt a pattern the industry calls snowshoeing. TNS, the analytics engine behind Verizon's call protection, describes bad actors doing exactly this in its robocall reporting: spreading traffic across many numbers, keeping volume per number low, and disposing of numbers quickly. When a legitimate lender does the same thing for innocent reasons, the model sees the same thing. Nobody at the engine reviews your business case.
So both directions are punished, and the safe region in the middle is not published. None of the three US engines (First Orion, Hiya, TNS) discloses its thresholds, and the models are retrained continually, so any vendor quoting you an exact safe dials-per-day figure is inventing it. What follows is how we reason about sizing on the floors we run, with the logic shown so you can adjust it when the environment moves.
Peak daily dials, divided by a modest per-number ceiling, plus headroom for resting.
Start from peak daily dial volume, not average. Campaigns breathe: fresh lead batches, end-of-month pushes, seasonal spikes. Size for the heaviest realistic day, because the day you exceed your per-number ceiling is exactly the day you can least afford new labels.
Then pick a per-number daily ceiling. This is the contested variable, and the honest answer is that no defensible universal figure exists. Our operating position, built from running outbound floors and watching per-DID answer rates react: keep it modest. Low hundreds of dials per number per day is the upper end of what we are comfortable with on warm, consented traffic, and on cold lists we keep it well below that. The flag risk is never volume alone; it is volume combined with what happens to those calls, meaning answer rate, call duration, and how often recipients block or report the number. A cold list drags all of those the wrong way, so it earns a lower ceiling.
Divide peak daily dials by the ceiling to get your active pool, then add headroom, because some fraction of the pool should always be resting rather than dialing. Worked through as arithmetic (an illustration of the method, not an industry standard):
| Peak daily dials | Per-number ceiling (your call) | Active numbers needed | With resting headroom |
|---|---|---|---|
| 2,000 | 100 | 20 | 25 to 30 |
| 10,000 | 150 | 67 | 85 to 100 |
| 50,000 | 150 | 334 | 420 to 500 |
Two things to notice. The ceiling is a management dial, not a constant: if per-DID answer rates hold steady for a month, you can edge it up, and if they sag, bring it down before the engines bring it down for you. And the headroom is what makes resting and retirement possible without emergency purchases, which is where most operators get pushed into the churn pattern they were trying to avoid.
The free 17-question Five9 Health Check covers caller ID setup, campaign structure and pacing, and shows you where your configuration sits against the outbound floors we run every day.
A fresh block of numbers at full volume on day one is what a snowshoe block looks like.
A set of brand-new numbers that all light up on the same morning at full volume is indistinguishable, from the engine's side, from a fresh snowshoe block. New numbers have no reputation history, and no history reads as risk.
The fix is boring and cheap: buy ahead of need. Provision numbers weeks before the campaign that will use them, leave them live and able to receive calls, and register them before the first outbound dial. In our own practice, a number that has existed for around a month, registered and reachable, behaves noticeably better in its first weeks of dialing than one bought on Friday and worked on Monday. We cannot see inside the models to prove the mechanism, and we will not pretend to, but the pattern has been consistent across the floors we manage.
Registration in the US runs through the Free Caller Registry, a shared portal operated by the three analytics engines themselves: First Orion (which serves T-Mobile), Hiya (AT&T) and TNS (Verizon). It costs nothing, takes up to 20 numbers entered individually or a file upload for larger batches, and asks for business identity, call purpose and expected monthly volume. Two caveats before you rely on it. Registration is explicitly not a guarantee against labeling, a point Hiya makes plainly in its own material. And the registry restricts itself to businesses calling on their own behalf, which excludes outsourcers and service providers calling for clients, an awkward gap if you run a BPO and one reason per-client number strategy matters in outsourced setups.
Ramp volume too. A registered, aged number still should not go from zero to its full ceiling overnight. Start low, watch the per-DID answer rate, and step up over the first couple of weeks. It costs a little pacing capacity and buys a pool that is not on fire in month two. Every number in the pool also needs to route somewhere sensible when called back; missed-call callbacks are inevitable at volume, and a number that rings dead when returned is both a reputation signal and, in the UK, a regulatory problem.
The engines and the recipients treat number types differently, and that changes pool math.
Pool sizing interacts with the kind of numbers you buy, because a local number strategy and a toll-free strategy produce very different pool shapes.
Numbers that look local to the recipient tend to get answered more readily, which is the pitch behind local presence dialing. The cost is pool multiplication: matching area codes across a national campaign means many more numbers, each carrying less volume, which pushes the pool toward the thin-spread pattern the engines dislike. If you run local presence, run it with aged, registered, answerable numbers per region, not a rented pile of one-week DIDs.
Toll-free concentrates volume onto fewer numbers with a distinct reputation ecosystem and no pretense of being local. For service, collections and callback-heavy traffic that is often fine, and the smaller pool is easier to monitor and defend. For cold outreach, expect lower answer rates than local presentation in most of the setups we audit.
Branded calling sits on top of either. Carriers and their engine partners sell verified display of a business name, and in some programs a logo and call reason, on the recipient's screen, and the FCC opened a formal rulemaking look at call branding in late 2025. Branding does not replace pool discipline: it is bought presentation on top of behavior, and the behavior still gets scored. The same goes for STIR/SHAKEN attestation. Full A-level attestation, which your provider applies when it knows you and knows your right to the number, has been the baseline for US outbound since the FCC required large voice providers to implement the framework from mid-2021 under the TRACED Act. Unsigned or C-attested traffic starts every call at a disadvantage, whatever your pool looks like.
Reputation damage spreads through whatever pool a bad list touches.
Run every campaign out of one shared pool and a single aggressive win-back file can poison the same numbers your payment reminders and service callbacks ride on. That coupling is the strongest argument for structuring pools per campaign, or at minimum per traffic type: sales prospecting on one set of numbers, service and lifecycle traffic on another.
Isolation buys you two things. Blast radius control first: when a campaign misbehaves, the damage stays inside its own pool and the retirement decision is cheap. Second, interpretability: per-DID answer rate movements map to a single campaign and list, so diagnosis takes minutes rather than an archaeology project. The price is more numbers overall and more administration, which is real but modest next to the cost of a labeled service line.
In Five9 this maps naturally, since outbound campaigns carry their own caller ID configuration, so a pool-per-campaign layout is an assignment exercise rather than an engineering one. For BPOs the same logic goes a step further: separate pools per client, without exception. Sharing numbers across clients means one client's list quality decides another client's answer rate, and explaining that in a quarterly review is not a meeting anyone enjoys. The Free Caller Registry's restriction to businesses calling on their own behalf points the same way, since registration and branding end up needing the end client's identity attached, which only works cleanly when that client's numbers are theirs alone.
Nobody tells you a number has been labeled. The number tells you, if you are watching it.
You will almost never be notified that a number has been labeled. The recipient sees Spam Likely; you see silence. So the signal you can actually observe is behavioral: a number whose answer rate falls away from its pool.
The mechanics are simple enough to run in any reporting stack. Track answers divided by dials for every caller ID on a rolling window (we use seven days) and compare each number against the average of its own pool. List quality, time of day and seasonality move every number together; a label moves one number alone. A DID drifting well below its pool baseline for consecutive days, while its poolmates hold, is flagged somewhere until proven otherwise. Pull it from rotation the same day and investigate, rather than waiting for the campaign-level connect rate to sag, by which point you are weeks behind the event. If the whole pool has slumped at once, the problem is upstream of numbering, and we have written separately about diagnosing a sudden answer rate drop.
Back the metric with direct observation. Each engine labels independently for its carrier partner, so a number can be clean on AT&T and labeled on T-Mobile. Keep a cheap handset per major carrier, or use one of the reputation monitoring services that query the engines, and call your own DIDs on a schedule. Ten minutes a week of someone phoning the pool from carrier handsets has caught labels for us that dashboards took days to reflect. When a number does show a label, remediation starts with the engines' own dispute processes, which we cover in fixing Spam Likely caller ID.
Consistency beats churn, and lifecycle policy is what prevents the panic swap.
A flagged number is not always a dead number, but the panic swap (retire everything Friday, buy fifty fresh DIDs, blast Monday) is exactly the churn signature this article keeps warning about. Lifecycle policy is what stands between you and that reflex.
Resting means taking a number out of dialing rotation while keeping it live and answerable. Rest it, then fix the behavior that burned it, because a number returned to the same calling pattern will be re-flagged on schedule. Check its registration is current, dispute the label with the engine concerned, and when it comes back, bring it back at aged-number ramp volumes rather than straight to ceiling. How long to rest is another figure nobody publishes; we work in weeks rather than days, and we permanently retire any number that has been labeled more than once despite remediation, on the logic that its history now costs more than the number is worth.
None of this works without records. The floors we audit almost never have them, and the absence turns every incident into guesswork. Keep a per-DID register holding:
It is one small table, and it converts number management from folklore into something you can hand to the next campaign manager without a two-hour oral history.
Ofcom regulates what may appear in the caller ID field, which constrains pool tactics directly.
UK operators get less freedom to improvise here, because Ofcom regulates the caller ID field itself. The CLI guidance requires CLI data to be a valid, dialable number that uniquely identifies the caller, and presentation numbers specifically have to come from ranges Ofcom has declared available for allocation, have to be capable of receiving a return call, and have to be numbers the caller has authority to use. Originating providers are expected to bind their customers to those conditions contractually.
Read against pool strategy, those requirements cut down the US-style playbook. A large rented pool of unrelated geographic numbers used for local presence sits uncomfortably against the requirement that a presentation number uniquely identify the caller and be one the caller has the right to use. So UK presentation pools tend to be smaller, owned rather than rented, and consistently answered, and operators typically put their effort into answer handling and CLI accuracy rather than scale rotation.
The rules are also a moving target. Ofcom published updated CLI guidance on 29 July 2024, applying from 29 January 2025, as part of its measures against scam calls, and a further update published in July 2026 takes effect in 2027. Guidance moves quickly here and enforcement practice moves with it, so check the current text on Ofcom's site before building a numbering plan around any summary, including this one, and take advice from qualified counsel where the commercial stakes justify it. Our companion piece on Ofcom's dialler rules covers the wider UK obligations for outbound floors.
There is no published safe threshold, and anyone quoting one is guessing, because the analytics engines that assign spam labels keep their models private and retrain them continually. Our operating practice is to keep daily dials per number modest: low hundreds at the very top end on warm traffic, and well below that on cold lists, adjusted against each number's answer rate. Treat any ceiling as a dial you manage weekly, not a constant you set once.
Usually the opposite. Constant rotation across a large pool reproduces the pattern analytics engines call snowshoeing, spreading low volume across many short-lived numbers, which is a documented tactic of actual spammers. Engines score traffic shape, not intent, so churn reads as evasion. Stable, aged, registered numbers carrying moderate consistent volume build reputation over time, while rotation resets it. Rotate for coverage reasons if you need to, never as a reputation repair strategy.
Snowshoeing is spreading call or message volume thinly across a large set of phone numbers so that no single number carries enough traffic to trip volume-based detection. TNS, one of the three major US analytics engines, describes bad actors using this technique and disposing of numbers quickly. Because it is a known attack signature, legitimate businesses that heavily rotate large number pools can be scored the same way, which is why an oversized pool is a risk rather than a safety margin.
No carrier or analytics engine publishes an aging requirement. In our own operations we buy numbers weeks ahead of need, keep them live and able to receive calls, register them with the Free Caller Registry, and only then start dialing, ramping volume gradually over the first weeks. Around a month of age has worked noticeably better for us than a few days, though we cannot see inside the engines' models and will not pretend to know the exact mechanism.
It depends on the traffic. Local presentation tends to earn higher answer rates for cold outreach but multiplies pool size, which pushes toward the thin-spread pattern engines penalize, so it needs aged, registered numbers per region. Toll-free concentrates volume on fewer, easier-to-monitor numbers and suits service, collections and callback-heavy traffic. Many operators run both, split by campaign type, with separate monitoring for each pool. In the UK, Ofcom's CLI rules constrain local presence tactics further.
You will rarely be notified directly. The practical early-warning signal is per-number answer rate: track it on a rolling window for every caller ID and compare each number to its own pool average. A single number falling well below its poolmates for several consecutive days, while the rest hold steady, is very likely labeled somewhere. Confirm by calling your own numbers from handsets on each major carrier, since each analytics engine labels independently for its carrier partner.
It is a free portal run jointly by First Orion, Hiya and TNS, the analytics engines behind T-Mobile, AT&T and Verizon call protection, where businesses register outbound numbers, identity and call purpose. Register every number before first use. Two limits matter: registration does not guarantee you avoid labels, and the registry only accepts businesses calling on their own behalf, which excludes BPOs and outsourcers calling for clients, so outsourced setups need the end client involved.
The US analytics-engine ecosystem is specific to US carriers, but the UK has its own constraints. Ofcom's CLI guidance requires caller ID data to be valid, dialable and to uniquely identify the caller, and presentation numbers have to be ones the caller has the right to use and that can accept return calls. Updated guidance applied from January 2025, with a further revision published in 2026 to follow. Check the current text on Ofcom's site, since this area keeps moving.
We size, age and monitor DID pools as part of running outbound floors daily, with a dedicated pod on your account rather than a ticket queue. Across our optimization portfolio the aggregate result is a 40 percent lift in connect rate. Tell us your dial volumes and we will tell you what we would do.
Talk to us