An AI calling cadence is the schedule that decides how soon, how often, and how many times an AI voice agent may call a single lead before it stops. It is the one setting that quietly controls your connect rate, your number reputation, and a fair share of your compliance exposure.
The question every call center manager asks in week one is how many times the AI should call a lead. The answer is less obvious than it sounds. No federal law says stop after five calls, no single number fits every vertical, and most cadence advice published online was written for SDR teams running multi-channel email sequences rather than for voice agents placing thousands of dials a day.
This guide covers what the rules actually require in 2026, what carriers do when you ignore them, and a calling cadence framework you can copy into your own campaign settings today.
TL;DR
There is no federal cap on how many times you may call a lead, so the working limit is practical. After roughly five to seven unanswered attempts to the same number, the chance of connecting on the next attempt is close to nothing and the damage to your number reputation keeps compounding.
A defensible calling cadence starts fast, spaces out, and stops early. Place the first attempt within seconds of a form fill, spread the rest across seven to fourteen days at different times of day, then stop and move the record to another channel.
Cadence rules only help if something enforces them. If you run a self-serve platform with no number rotation and no per-number frequency cap, this framework will not save you, and a low-volume team calling a few hundred records a week does not need it at all.
Key takeaways
- No federal statute sets a per-lead call frequency cap, but consent, calling hours, and revocation rules all constrain cadence.
- Five to seven attempts across seven to fourteen days is the responsible ceiling for most regulated outbound programs.
- Attempt one should land within seconds of a form submission, because that call carries most of the value in the sequence.
- Carrier filtering is the hidden limit. Frequent, short, unanswered calls from one number get it labeled and that label follows every future call.
- The FTC caps abandoned calls at 3 percent of answered calls per campaign over 30 days, which AI operations hit through failed transfers.
- Revocation of consent must be honored promptly and across channels, with requirements phasing in through 2026.
- A managed platform enforces the calling cadence automatically. A self-serve platform will let you set anything and will not stop you.
Table of contents
- What an AI calling cadence is
- Why calling cadence matters more with AI than with reps
- What the law actually says about call frequency
- The calling cadence framework
- The cadence schedule at a glance
- Number health: the hidden cadence limit
- The abandoned call problem
- How cadence changes by vertical
- Who enforces the cadence, and how
- Cadence mistakes that cost you numbers
- AI calling cadence FAQ
- The bottom line
What an AI calling cadence is
An AI calling cadence is a documented set of rules covering the timing of the first attempt, the interval between attempts, the time of day each attempt is placed, the maximum number of attempts per record, and the condition that ends the sequence. Written down properly, it is four or five lines of configuration.
The reason it deserves its own policy is scale. A human rep who cold dials all day might work 80 to 120 records. An AI voice agent can place tens of thousands of dials in the same shift. Every mistake in the cadence is therefore repeated at industrial volume before anyone notices it on a dashboard.
Cadence is not the same thing as consent, and it does not replace it. Consent decides whether you may call at all. Calling cadence decides how you behave once you have it. You need both, and the compliance vocabulary around them is collected in the AI calling glossary.
Why calling cadence matters more with AI than with reps
When a human team dials, the natural slowness of manual work acts as a built-in limiter. Nobody calls the same person six times before lunch because there is not enough time in the day. Automation removes that limiter, so you have to put it back deliberately.
An over-aggressive human team is annoying. An over-aggressive AI campaign creates a compliance record, a number reputation problem, and a carrier filtering problem at the same time, and it does all three faster than your reporting cycle.
- Carrier filtering starts. The major carriers monitor outbound patterns. High call frequency to the same destinations, high unanswered rates, and very short call durations are the signals that get a number labeled. Once labeled, it displays as Spam Likely on every call it places, and connect rates fall sharply across the whole campaign.
- Leads complain. Somebody who receives four calls in one day from a number they do not recognize does not read that as persistence. They block it, report it, or file a complaint, and TCPA damages run $500 per violating call and up to $1,500 for willful violations.
- Your data degrades. Past five to seven unanswered attempts, the probability of connecting on the next one is close to zero. Every further dial spends budget and number reputation for almost no expected return.
What the law actually says about call frequency
No single federal law states a maximum number of calls per lead. Several rules do shape when you may call, what you need before you call, and what has to happen when someone tells you to stop.
The TCPA
The Telephone Consumer Protection Act sets no per-lead frequency cap. It requires prior express written consent before placing marketing calls to a cell phone with an autodialer or an artificial or prerecorded voice, and it requires you to honor an opt-out. The statutory damages sit in 47 U.S.C. 227. The FCC confirmed in February 2024 that AI-generated voices count as artificial voices for this purpose, so an AI campaign inherits every one of those obligations.
The FTC Telemarketing Sales Rule
The TSR caps abandoned calls at no more than 3 percent of answered calls per campaign, measured over a 30-day period, and sets the standard 8 AM to 9 PM window in the called party’s local time. It also carries disclosure and record keeping duties. The FTC publishes a readable compliance guide that is worth an hour before you launch anything.
Revocation of consent
This is the rule that touches cadence most directly. A revocation has to be honored promptly, treated as applying across channels rather than only to the channel it arrived on, and the remaining pieces of that requirement are phasing in through 2026. In practice it means a text reply of stop has to end the calls too, and your cadence engine needs to check the suppression list before every attempt, not once a night.
The one-to-one consent rule that never took effect
You will still find articles asserting that separate one-to-one consent for each individual seller became binding law. It did not. That rule was vacated by the Eleventh Circuit in January 2025 and never took effect, and prior express written consent under the existing TCPA standard remains the operative requirement. Adopting one-to-one consent as internal policy is still a reasonable decision, because it produces a cleaner record and narrows your exposure if a lead vendor’s disclosure is challenged later.
State rules
Several states run narrower dialing windows than the federal one, require specific opt-out language, or impose cooling-off periods between calls to the same number. A few set explicit frequency limits. Our summary of the state-specific calling rules covers the ones that come up most. None of this is legal advice, so have counsel review your program before you scale it.
The calling cadence framework
This framework balances speed to lead at the front with restraint at the back. It is written for inbound-generated leads in regulated verticals, which is where the sequence matters most.
- Attempt 1, within 30 seconds of submission. This is the highest-value dial in the whole sequence. Lead response research has consistently found that the odds of reaching and qualifying a lead collapse within the first few minutes, and automation is the only realistic way to hold that window on evenings and weekends inside legal dialing hours.
- Attempt 2, two to four hours later. Same day, intent still fresh, no answer on the first try. A second same-day touch is defensible because the person raised their hand hours ago.
- Attempt 3, next business day at a different time of day. If the first two attempts were morning calls, make this one an afternoon call. Rotating the time of day is the cheapest connect-rate improvement available in any calling cadence.
- Attempts 4 and 5, one every two to three days. Keep varying the hour. By this point the person either answers or is not reachable at that number, and you are testing the second possibility.
- Attempts 6 and 7, optional, one per week. These are the lowest-probability calls in the sequence. Plenty of good operations stop at five, and stopping at five is a defensible choice rather than a missed opportunity.
- After attempt 7, stop. Move the record to email or SMS if you hold consent for those channels, or mark it unreachable. Continuing past this point spends number reputation for a return that rounds to zero.
The cadence schedule at a glance
| Attempt | When | Time of day | Purpose |
|---|---|---|---|
| 1 | Within 30 seconds of the form fill | Whenever the lead arrives, inside legal hours | Catch live intent |
| 2 | Two to four hours later, same day | Opposite half of the working day | Second chance while intent is fresh |
| 3 | Next business day | Different from attempts 1 and 2 | Test a new availability window |
| 4 | Day 4 or 5 | Rotate again | Reach shift workers and travelers |
| 5 | Day 7 or 8 | Rotate again | Last high-value attempt |
| 6 and 7 | Day 10 to 14, weekly pace | Rotate again | Optional tail, low expected return |
| Stop | Day 14 onward | No further dials | Move to another channel or mark unreachable |
Cadence audit
Find out what your cadence is doing to your numbers
We will review your current attempt schedule, dialing windows, and number rotation against the rules above. It takes about twenty minutes and you keep the findings either way.
Number health: the hidden cadence limit
Most managers think about calling cadence in terms of the lead experience. The more expensive dimension is number health, because it affects every campaign at once rather than a single record.
Every outbound number carries a reputation with each major carrier, shaped by call frequency, answer rate, average duration, and complaint volume. When that reputation degrades, the number gets labeled or blocked, and the damage is not confined to the leads that caused it.
The chain runs in a predictable order. High frequency to the same destinations attracts monitoring. Low answer rates on those attempts pull the reputation score down. The carrier applies a spam label or blocks the number. Every subsequent call from that number, to anyone, carries the label. Connect rates drop across the board. You rotate to fresh numbers, and unless the cadence itself changed, the same sequence repeats on the new ones.
Rotation without a cadence fix is the most common wasted spend in outbound. New numbers cost money, take time to warm up, and burn at the same rate as the last set if the underlying attempt frequency did not change.
The abandoned call problem
An abandoned call is one where a person answers and no agent, human or AI, is there to speak with them. The TSR limit is 3 percent of answered calls per campaign over 30 days, and AI operations reach it in ways human floors usually do not.
At scale that ceiling is tight. On 10,000 answered calls it is 300 abandoned calls before you are outside the limit, and a single afternoon of failed warm transfers can spend the whole allowance.
Two controls matter. First, response latency has to be low enough that whoever answers hears a voice immediately rather than silence, since silence is what people hang up on and what regulators count. Second, warm transfers have to actually complete. A transfer that drops the caller into dead air is an abandoned call regardless of how well the AI portion of the conversation went.
How cadence changes by vertical
The framework holds across regulated outbound, but the tail length changes with the value and shelf life of the lead.
- Insurance. Shopping windows are tied to renewal dates, so a lead that goes quiet is often genuinely busy rather than uninterested. The full seven attempts are usually worth placing.
- Mortgage and lending. Intent moves with rates. Run the standard sequence, then re-enter the record only when there is a real reason to call again, such as a rate move, and treat that as a new sequence with its own consent check.
- Solar. Speed dominates everything else. The first attempt inside 30 seconds does most of the work, and the tail past attempt five rarely earns its cost.
- Debt relief. Aged databases are the main use case, and consent age is the constraint rather than attempt count. Confirm the consent record is current and defensible before you start a sequence on a two-year-old record.
Who enforces the cadence, and how
A calling cadence policy that lives in a document and not in the dialer is not a policy. The practical question is which layer refuses to place the eighth call.
| Approach | Attempt caps | Number rotation | Where it fails |
|---|---|---|---|
| Managed AI calling service | Enforced before each dial | Monitored and rotated by the vendor | Less freedom to override the rules for a one-off campaign |
| Self-serve AI calling platform | Configurable, not enforced by default | You build and monitor it | Nothing stops a bad campaign setting from running all week |
| Manual list management | Whatever the supervisor remembers | Rarely done at all | Breaks down above a few thousand records |
One clarification before anyone reads that table as a dialer comparison. Bigly Sales does not sell a predictive dialer or a CRM, so this is not a like-for-like feature contest with those categories. It is a comparison of who carries the enforcement burden. If your team has engineers and a compliance officer with time to spare, the self-serve column is a legitimate choice. Most regulated outbound teams do not, which is the argument for a TCPA-compliant managed platform.
Cadence mistakes that cost you numbers
- Calling at the same hour every time. Seven attempts at 10 AM test one hypothesis seven times. Rotate the hour and you test seven.
- Treating a callback request as a new sequence. A booked callback is a scheduled appointment. Placing your regular attempts around it reads as harassment to the person waiting for the call they agreed to.
- Checking the suppression list once a night. Revocation has to be honored promptly and across channels. A nightly sync means you can legally call someone who opted out this morning, and you will.
- Rotating numbers instead of fixing frequency. New numbers buy a few weeks. They do not change the pattern that got the last set labeled.
- Running one cadence for every list. A 30-second-old form fill and a two-year-old aged record are not the same asset and should not get the same schedule.
- Measuring attempts instead of connects. Attempt counts always go up. Connect rate per attempt is the number that tells you whether the cadence is working.
AI calling cadence FAQ
How many times can you legally call someone?
There is no single federal number. The TCPA sets no per-lead frequency cap. What it does require is prior express written consent for automated marketing calls to cell phones, and an immediate stop when someone opts out. State rules and the FTC Telemarketing Sales Rule add further restrictions, including the 3 percent abandoned call limit and calling window rules. The practical ceiling for most operations is five to seven attempts, after which you are spending money and number reputation for very little.
What is a good calling cadence for AI outbound?
First contact within 30 seconds of the form submission. Second attempt two to four hours later if nobody answered. Third attempt the next business day at a different time of day. Then one attempt every two to three days for the following week, with an optional weekly tail out to attempt seven. Stop after that and move the record to another consented channel. This shape works across insurance, mortgage, solar, and debt relief.
Does calling too often damage my phone numbers?
Yes, and this is the most underrated cost of a bad calling cadence. Carriers score every outbound number on call frequency, answer rate, call duration, and complaints. A number that places many short unanswered calls to similar destinations gets labeled or blocked. The label then applies to every call that number places, including calls to leads who would have answered. Recovering a burned number is slower and more expensive than avoiding the pattern.
How does a platform enforce cadence automatically?
On a managed platform the check happens before the dial rather than after. The system looks at how many attempts this record has already received, when the last one was placed, whether the local time is inside the legal window for that area code, and whether the person appears on any suppression or do-not-call list. Any failed check cancels the call. Self-serve platforms generally let you configure the same rules but will not stop a campaign that violates them.
Can I call leads on weekends?
Federal rules allow calls between 8 AM and 9 PM in the called party’s local time, which includes weekends, but several states run narrower windows and some restrict Sundays specifically. The safe implementation checks the lead’s location against state rules at dial time rather than relying on a campaign-level setting. If you are running this manually, weekend calling is the easiest way to end up outside a state window without noticing.
What happens if I ignore cadence limits entirely?
Three things, in order. Connect rate per attempt falls toward zero past the fifth or sixth try, so the extra dials produce almost nothing. Your outbound numbers pick up spam labels, which suppresses connect rates on every other campaign running from those numbers. And complaint volume rises, which is where TCPA exposure at $500 to $1,500 per violating call becomes a real number rather than a theoretical one.
Is one-to-one consent required before I call?
No. The rule that would have required separate consent for each individual seller was vacated by the Eleventh Circuit in January 2025 and never took effect. Prior express written consent under the existing TCPA standard is what applies. Many operators still collect consent one-to-one voluntarily, because it produces a cleaner audit trail and reduces the argument available to a plaintiff if a lead vendor’s disclosure page is ever put in front of a court.
How fast do I have to stop calling after someone opts out?
Promptly, and across channels rather than only on the channel that received the request. Someone who replies stop to a text message has revoked consent for calls as well, and the requirements around cross-channel revocation are phasing in through 2026. Practically this means your suppression list has to be checked at dial time, not synchronized overnight, and the opt-out has to propagate to every system that can originate contact.
Should aged leads get the same cadence as fresh ones?
No. A fresh form fill justifies an immediate call and a same-day second attempt because the intent is live. An aged record does not, and the first question there is whether the consent on file is still current and documented. Aged campaigns generally work better with a shorter sequence, wider spacing, and a script that acknowledges the age of the inquiry instead of pretending it just arrived.
What is the single most important attempt in the sequence?
The first one, by a wide margin. Lead response research has found for years that reach and qualification rates fall away sharply within the first minutes after a form submission, and no later attempt recovers what a slow first call gives up. If you can only fix one thing about your calling cadence this quarter, make it the delay between the form hitting your CRM and the phone ringing.
The bottom line
The calling cadence you configure decides three things at once. How many deals you close, how long your numbers stay usable, and how much regulatory exposure you carry. Too few attempts and you lose people who would have answered on the third try. Too many and you burn numbers, invite complaints, and pay for dials with no realistic chance of connecting.
Start fast because speed wins the conversation. Space the attempts out and vary the hour because persistence without patience reads as spam. Stop at five to seven because the math stops working and the risk keeps climbing. Then make sure something in your stack actually refuses to place the eighth call, because a cadence policy nobody enforces is just a document.
Enforced, not suggested
Let the platform stop the eighth call for you
Attempt caps, dialing windows, suppression checks, and number rotation are enforced before every dial on a managed deployment. Most teams are placing live calls within days.
