How to Handle Voicemail Detection & Drops in AI Outbound

How to Handle Voicemail Detection & Drops in AI Outbound | Thoughtly

Last updated May, 2026

How to Handle Voicemail Detection and Voicemail Drops in AI Outbound

Outbound AI voice agents are only useful if they reach a live person—or, at minimum, leave a message worth returning. In high-volume lead follow-up, a meaningful percentage of calls land on voicemail. If your agent sits through a thirty-second greeting, says nothing, and hangs up, you have burned credits and wasted a touchpoint.

This guide walks through configuring voicemail detection and voicemail drops in Thoughtly so your outbound agents handle answering machines correctly: detect them fast, deliver a personalized message, end the call cleanly, and route the outcome to your CRM or automation stack for the next follow-up step.

What You Will Learn

How Thoughtly's answering machine detection (AMD) identifies voicemail systems versus live callers
How to enable voicemail drops and write messages that earn callbacks
How to use voicemail variables for dynamic personalization
How to route voicemail outcomes in post-call automations
Best practices for voicemail message length, content, and compliance

What You Will Need

A Thoughtly account with at least one deployed outbound voice agent
A phone number connected to the agent with outbound calling enabled
Access to Thoughtly's Agent Builder (v1.8 node-based builder)
A working automation (recommended) using the On Call Completed trigger for post-call routing
Familiarity with Thoughtly variables and metadata (helpful but not required)

Step 1: Understand How Voicemail Detection Works in Thoughtly

Thoughtly uses answering machine detection to distinguish between a live person picking up and a voicemail system playing a recorded greeting. When your outbound agent dials a contact, AMD listens during the first few seconds of audio. If it detects a voicemail greeting pattern—long uninterrupted speech, a beep prompt, or a carrier-level "leave a message" recording—it classifies the call as voicemail.

This classification happens before the agent's start node fires. If voicemail is detected and you have voicemail drops enabled, the agent skips the conversation flow entirely, plays your pre-configured voicemail message, and ends the call. If voicemail drops are not enabled, the agent simply ends the call without engaging—and, importantly, calls that go to voicemail without drops enabled do not incur charges.

One key point from Thoughtly's documentation: do not attempt to simulate voicemail detection inside your start node's outcomes. The docs explicitly warn that this approach is unreliable. Instead, rely on the platform-level AMD, which uses dedicated AI models tuned for this classification.

Detection Result	Voicemail Drops Enabled	What Happens
Live person detected	N/A	Agent enters normal conversation flow from start node
Voicemail detected	Yes	Agent plays voicemail message, then ends the call
Voicemail detected	No	Agent ends the call immediately (no charge)
Screening service detected	N/A	Call Screening Assist responds, then proceeds to conversation or voicemail

Step 2: Enable Voicemail Drops in Agent Settings

To turn on voicemail drops for your outbound agent:

Open Agent Settings. Navigate to your agent in the Thoughtly dashboard and select the Settings tab.
Find the Voicemail section. Scroll to the Voicemail settings area.
Enable voicemail. Toggle the voicemail option on. When detected, the agent will play your voicemail message and end the call automatically.
Write your voicemail message. Enter the text the agent will speak when it reaches voicemail. This is a text-to-speech message, not a pre-recorded audio file—it uses the same voice your agent uses in live conversations.
Save your settings. Confirm the changes. Your agent will now leave voicemail drops on all outbound calls that reach an answering machine.

The voicemail message fires only when AMD classifies the call as voicemail. It does not play during live conversations, during call screening, or when the recipient picks up.

Step 3: Write a Voicemail Message That Earns Callbacks

A voicemail from an AI agent follows the same rules as a voicemail from a human rep—except you get to test and iterate the exact wording across thousands of calls. Thoughtly's documentation recommends keeping voicemail messages under 20–25 seconds, including a call-back request and number or URL, and avoiding sensitive information.

Here is a framework for structuring an effective voicemail drop:

Identify yourself. State the agent's name and the company. Example: "Hi, this is Ava from Acme Home Services."
State the reason. Be specific about why you are calling. Vague reasons like "important matter" get deleted. Example: "I'm calling about the quote request you submitted for HVAC repair."
Include a call to action. Tell them what to do next—call back, visit a URL, or expect another attempt. Example: "You can call us back at this number or visit acmehvac.com/schedule to book a time that works for you."
Keep it short. The entire message should land under 25 seconds. Cut filler words. Every sentence should earn its spot.

Example voicemail message for a mortgage lead follow-up:

Hi, this is Alex from Summit Lending. I'm following up on your recent rate inquiry. We have some options that match what you were looking for. You can reach us back at this number or visit summitlending.com/rates to see current offers. Looking forward to connecting with you.

Example voicemail drop — mortgage lead follow-up

Example voicemail message for an insurance re-engagement campaign:

Hi, this is Jordan from Brightside Insurance. I'm reaching out because your auto policy renewal is coming up and we found a few ways to save on your premium. Give us a call back at this number or visit brightside.com/renew to review your options. Talk soon.

Example voicemail drop — insurance re-engagement

Step 4: Personalize Voicemail Messages with Variables

Static voicemail messages work, but personalized ones convert better. Thoughtly supports voicemail variables, which let you inject dynamic values into your voicemail message text—the same way you use variables in speak nodes and advanced prompts.

You can reference contact attributes, metadata, or system variables inside the voicemail message field. When the agent reaches voicemail, it resolves these variables before speaking the message.

Common variables to include in voicemail drops:

Variable	What It Resolves To	Use Case
{{system.contact.first_name}}	The contact's first name	Personalized greeting
{{attributes.lead_source}}	Where the lead came from	Contextual reason for calling
{{metadata.campaign}}	The campaign label for this call	Campaign-specific messaging
{{attributes.service_type}}	The service the lead inquired about	Specific follow-up context

Example personalized voicemail message:

Hi {{system.contact.first_name}}, this is Ava from Acme Home Services. I'm following up on your {{attributes.service_type}} request. Give us a call back at this number or visit our website to schedule a time. We would love to help.

Personalized voicemail drop with variables

Make sure the variables you reference are populated before the outbound call fires. If you are using automations to trigger calls, set contact attributes via the Create or Update Contact step or pass metadata in the Call Contact step before the agent dials.

Step 5: Route Voicemail Outcomes in Post-Call Automations

Leaving a voicemail is not the end of the follow-up sequence—it is the beginning of the next step. Thoughtly's On Call Completed trigger fires after every call, including calls that land on voicemail. The trigger output includes voicemail detection status, which you can use to branch your post-call automation logic.

Here is how to set up voicemail-aware post-call routing:

Create an automation with the On Call Completed trigger. Scope it to your outbound agent or to All Agents if you want a universal voicemail handler.
Add a Conditions step. Check whether the call ended with a voicemail drop. The call outcome data includes voicemail detection status. You can also check the disposition—Thoughtly's auto-disposition labels calls as "Left voicemail" when a drop was delivered.
Branch: voicemail detected. If the call hit voicemail, trigger a follow-up SMS, schedule a retry call for the next day, or update the CRM contact record with a "voicemail left" tag.
Branch: live conversation. If the call connected to a live person, route to your standard post-call flow—CRM update, lead scoring, booking confirmation, or notification to the sales team.

Example automation pattern:

Trigger	Condition	Action
On Call Completed	Disposition = Left Voicemail	Send follow-up SMS with callback link
On Call Completed	Disposition = Left Voicemail AND retry_count < 3	Schedule retry call via Delay For + Call Contact
On Call Completed	Disposition = Qualified lead	Update CRM, notify sales rep
On Call Completed	Disposition = No answer (no voicemail)	Schedule retry with different time-of-day

This pattern ensures that no voicemail touchpoint is wasted. Each voicemail drop feeds back into the follow-up sequence, and the contact continues moving through your pipeline until they respond or are suppressed.

Step 6: Pair Voicemail Drops with Call Screening Assist

If your outbound agent calls contacts who use phone screening services, spam protection apps, or device-level call filters, you may want to pair voicemail drops with Thoughtly's Call Screening Assist feature. Call Screening Assist lets the agent respond to automated screening prompts—"Who is calling?" "What company are you with?"—before the call connects or goes to voicemail.

To configure Call Screening Assist alongside voicemail drops:

Enable Call Screening Assist. In your agent or phone number settings, turn on Call Screening Assist and fill in the agent name, company name, and reason for calling.
Write a truthful reason. Use a specific, honest reason like "following up on your quote request" rather than vague phrasing like "important business matter." The docs are explicit: avoid spammy language.
Test with a screened number. Place a test outbound call to a number that uses call screening. Review the transcript in Thoughtly's call history to verify the agent responded to screening prompts correctly before either connecting live or leaving a voicemail.

The two features operate in sequence: Call Screening Assist handles the front gate, and voicemail drops handle the answering machine. Together, they maximize the chance that your agent either reaches the person or leaves a useful message.

Step 7: Monitor Voicemail Performance and Optimize

Once voicemail drops are live, track the metrics that matter:

Voicemail rate: What percentage of outbound calls reach voicemail? A voicemail rate above 60–70% may indicate timing issues—try shifting call windows to earlier in the morning or late afternoon when pickup rates tend to be higher.
Callback rate: Of contacts who received a voicemail drop, how many called back? This is the direct measure of your voicemail message quality.
Retry-to-connect rate: Of contacts who did not call back after a voicemail, how many picked up on a subsequent attempt? This tells you whether the voicemail primed them to answer next time.
Credit efficiency: Since calls that go to voicemail without drops enabled do not incur charges, compare cost-per-contact-reached with drops on versus off. In most outbound campaigns, enabling drops is more cost-effective than silent hang-ups because the voicemail itself generates callbacks.

Use Thoughtly's call history and disposition filters to segment calls by outcome. Export voicemail-tagged calls to your CRM or BI tool for deeper analysis. If callback rates are low, test different voicemail messages—change the reason, the call-to-action, or the length—and measure the lift.

Common Mistakes

Simulating voicemail detection in the start node. Thoughtly's documentation explicitly warns against this. The start node's outcomes should not attempt to detect whether the recipient is a voicemail system. Use the platform's built-in AMD instead—it uses dedicated AI models and is significantly more reliable than prompt-based classification.
Writing voicemail messages that are too long. Keep it under 20–25 seconds. A voicemail that rambles past 30 seconds gets skipped or deleted. Front-load the reason for calling and the callback instructions.
Including sensitive information in voicemail messages. Do not include account numbers, medical details, financial specifics, or anything covered by HIPAA, GLBA, or similar regulations. The voicemail may be heard by someone other than the intended contact.
Forgetting to populate variables before the call. If your voicemail message references a variable that has not been set—say, the contact's first name—the agent may speak the raw variable placeholder or skip the token entirely. Always verify that contact attributes and metadata are populated in the automation steps before the Call Contact step.
Not building a post-call voicemail branch. Leaving a voicemail without a follow-up action is a wasted touchpoint. Always route voicemail outcomes into your automation sequence—whether that means sending an SMS, scheduling a retry, or updating the CRM.
Running voicemail drops outside of compliant hours. Voicemail drops are still outbound contacts. Respect quiet hours, TCPA requirements, and any state-specific calling windows. Thoughtly's automation scheduling and quiet hours configuration can enforce these boundaries.

Measuring Success

Voicemail drops should be measured as part of your overall outbound conversion funnel, not in isolation. Here are the specific metrics to track:

Metric	What It Tells You	Target Range
Voicemail detection accuracy	Is AMD correctly classifying calls?	> 95% accuracy (false positives should be rare)
Voicemail drop delivery rate	Of detected voicemails, how many successfully played the message?	> 98% (failures indicate carrier or timing issues)
Callback rate from voicemail	Are recipients returning calls after hearing the drop?	5–15% depending on industry and lead temperature
Retry pickup rate	Do contacts who received a voicemail answer the next call attempt?	20–40% higher than contacts with no voicemail left
Cost per contacted lead	How does enabling voicemail drops affect your cost efficiency?	Lower than no-drop campaigns due to callback volume
Follow-up SMS response rate	Are voicemail + SMS sequences outperforming voicemail-only?	Track incrementally; SMS should add 5–10% lift

Review these metrics weekly during initial rollout, then monthly once your voicemail message and follow-up sequence are stable. Use Thoughtly's disposition filters and On Call Completed automation data to feed these numbers into your CRM or reporting dashboard.

Frequently Asked Questions

Does Thoughtly charge for calls that go to voicemail?

Calls that go to voicemail without voicemail drops enabled do not incur charges. If you have voicemail drops enabled and the agent delivers a message, normal call charges apply based on the duration of the voicemail playback.

Can I use a pre-recorded audio file instead of text-to-speech for voicemail drops?

Thoughtly's voicemail feature uses text-to-speech with the same voice assigned to your agent. This ensures the voicemail sounds consistent with live conversations. The benefit is that you can update the message text instantly without re-recording audio files, and you can include dynamic variables that resolve per-contact.

How long should a voicemail drop be?

Thoughtly's documentation recommends keeping voicemail messages under 20–25 seconds. This is long enough to identify yourself, state the reason for calling, and include a callback prompt. Messages longer than 30 seconds see diminishing returns—recipients tend to stop listening or delete without finishing.

Can I use different voicemail messages for different campaigns?

Yes. You can use metadata variables to customize the voicemail message per campaign. Pass campaign-specific context as metadata in your automation's Call Contact step, then reference those metadata keys in the voicemail message field. Alternatively, if you need completely different messages per campaign, you can create separate agents with different voicemail configurations and route calls through the appropriate agent in your automation logic.

What happens if voicemail detection misclassifies a live person as voicemail?

A false positive—where AMD thinks a live person is voicemail—means the agent plays the voicemail message to a live caller instead of entering the conversation flow. This is uncommon with Thoughtly's enhanced AMD, which uses dedicated AI models for classification. If you notice false positives in your call history, review the transcripts and contact Thoughtly support. The platform has improved AMD accuracy significantly in recent updates.