# ElevenAgents Campaign Manager: Calling Windows, Callbacks and Stop Rules

Before moving outbound calls into Campaign Manager, check the policy boundaries that can survive a workspace settings change.

By Paras Madan | 2026-11-11


ElevenLabs launched ElevenAgents Campaign Manager on November 9 to manage outbound retries, callbacks and outcomes. For an operator setting it up, the most consequential details concern when an agent may call again, how a stop request is recorded and which settings apply to a running campaign.

The [launch announcement](https://elevenlabs.io/blog/elevenagents-campaign-manager) says it is available in the Agents sidebar and supports phone calls, WhatsApp voice and WhatsApp messaging. Email and SMS are described as coming soon. This is a documentation-based setup review, not a claim that I ran a campaign.

## Understand the policy boundary before importing contacts

The [campaign documentation](https://elevenlabs.io/docs/eleven-agents/phone-numbers/campaigns) says each campaign takes a policy snapshot when created. Changing workspace defaults affects future campaigns, not running ones. To change a running campaign’s policy, the documented route is to stop it and create another.

That creates a migration risk. If your team changes its calling policy after complaints or an operational mistake, saving the workspace setting is insufficient. Keep an inventory of active campaigns and their policy versions. Treat a policy change as an operational rollout with an owner and a verification step, rather than a settings edit that is assumed to apply everywhere.

Before recreating a campaign, reconcile every recipient’s history. Starting another campaign should not mean treating previously contacted people as untouched leads.

## Calling windows have a callback exception

Use separate campaigns for different timezones. Recipient-requested callbacks can fall outside normal calling windows, according to the documentation.

Consider a hypothetical customer who asks for “tomorrow at four.” Before that request becomes a scheduled call, your process needs to establish the date and timezone and capture the request accurately. If your team cannot determine those details reliably, route the case for human review rather than treating a vaguely understood time as permission to call.

Calling windows also cannot tell you whether a person consented to this particular campaign. Check eligibility before enrollment, and have the appropriate compliance owner review the intended use and jurisdictions. Product controls do not replace those decisions.

## Keep a suppression record outside individual campaigns

Campaign-level do-not-call state is not a cross-campaign suppression system. The documentation tells operators to deduplicate recipients across campaigns. It also says the agent must collect the signals used for callbacks, redirects and stop requests.

**Use a shared suppression record before every new enrollment.** If a customer declines in one campaign, a new CSV upload should not quietly reintroduce them through another. Match identities across the phone numbers your business already knows, and establish who reviews ambiguous matches rather than letting an assistant decide that two similar records are different people.

The launch describes a per-contact decision timeline, CSV exports and a needs-attention queue. Use those to audit why another attempt was scheduled. A dashboard total showing more contacts reached does not tell you whether the right people were called at an appropriate time.

## Test stop behavior before retry performance

For a small internal pilot, use consenting test participants and write the expected outcome before each call. Include an unanswered call, a specific callback request, an ambiguous request, a request to stop and a contact who already appears in the suppression record. These are suggested acceptance cases, not results from tests performed here.

Your test should answer two questions: did the agent capture the correct intent, and did the campaign take the correct next action? A clear transcript with the wrong scheduled action is still a failed test. Inspect the recorded decision as well as what the agent said.

Zero-retention workspaces and agents cannot use campaigns. If that is a firm requirement, resolve the mismatch before migration.

## Expand only when the controls are observable

The announcement emphasizes carrying context between calls and stopping when the desired outcome is reached or a person asks to stop. That is the product’s promise. Your rollout should make those transitions visible enough that someone can diagnose a mistake without reconstructing an entire day of calls.

The useful improvement is less manual coordination around legitimate follow-ups. If the system makes unwanted repeat contact easier, a higher completion rate is the wrong success metric. Begin with a narrow, permissioned workflow and expand only after you can demonstrate that its stopping conditions work.

I’m Paras. I work at the intersection of AI, engineering and distribution.

I help teams explain complex AI products through clear technical writing and practical launch stories.

[Book a 30-minute call](https://cal.com/paras-madan/30min)