Signal-triggered branching

Signal-triggered branching changes work already scheduled for a contact or account. The event should change the next decision in the cadence, while the record continues to show what has already happened. Treat a buying signal as a state transition inside the active sequence.¹ Keep the enrollment and delivered-message history, then change only future work.^{2, 3}

Start with a signal you can explain

Define the trigger clearly enough that another person can understand why it fired. A vague interest label creates branches that are difficult to review later.

A signal is a measurable event that indicates a change in an account's buying context.⁴ Useful workflow categories include intent signals, job changes, funding events, website visitors, reply routing and opportunity reactivation.⁵ Score each event against your validated ICP. A single event is weak, while combined signals can show which accounts deserve attention.⁶

Ask:

• What exact event occurred?
• Does it describe the contact, the account, or both?
• What makes this event strong enough to change the current cadence?
• What evidence would make the branch expire?

Keep the trigger narrow at launch. First check whether the event produces a useful next action, then add more signal types if needed.

Put ownership and suppression before cadence

Give the branch an owner and a stop path before writing the message that follows the event. Without a suppression check, a trigger can keep working after the context has changed.

Each outbound workflow needs a trigger, a single owner, a suppression check and a written stopping rule.⁷ Before the cadence changes, the eligibility gate should resolve the event and check contact and account eligibility.⁸

For email, CAN-SPAM requires an opt-out to be honored within 10 business days; the mechanism must work for at least 30 days after the message, charge no fee, and require no more than a reply or a visit to one web page; each violating email can carry a penalty up to $53,088.⁹ FCC telemarketing rules require revocations made in any reasonable manner to be honored within a reasonable time of no more than 10 business days, and they prohibit an exclusive revocation method.¹⁰ An internal do not call list needs a written policy available on demand, trained staff and a request recorded when it is made.¹¹

If the branch depends on visitor identification, account for browser-level global privacy controls that communicate an opt-out of sale or sharing.¹²

Read the current state before writing

After the event passes the eligibility gate, read the sequence as it exists. The branch should make a controlled change to that state, with enough history to explain why the next action was selected.

Store an idempotency key so retries cannot create duplicate touches.¹³ The orchestration layer should evaluate the event against the current record and then write a single transition.¹⁴ It should never create a second active sequence just because the account produced another event.¹⁵

Check enrollment behavior in the system you use. In HubSpot, records enter a workflow only the first time they meet its triggers unless re-enrollment triggers are added; sequences default to unenrolling contacts after a reply or booked meeting, and both behaviors can be switched off.¹⁶

Record the current sequence, the last delivered step, the next unsent step and the signal that caused the review. These fields distinguish a deliberate branch from a duplicate enrollment.

Choose the branch action

Change the next action only when the event changes what you need to learn or say. Keep the action vocabulary small so the branch can be tested and explained.

• Continue the current path when the signal adds context without changing the reason for outreach.
• Change the next unsent step when the event gives you a more relevant reason to contact the account.
• Pause for human review when the event needs interpretation or a tailored response.
• Suppress when eligibility fails, a stop condition applies or another owner has taken control.

Set the pace from engagement. Modern cadences increase outreach frequency for engaged prospects and pause for people who request space.¹⁷ A website visit or research event can justify a faster review, while a request for space should prevent the branch from adding pressure.

Leave delivered messages unchanged. Edit the next unsent action instead.

Test the branch and watch the stop

For the first run, use a single trigger, a single owner and a segment small enough to read by hand.¹⁸ Inspect the run for bad event resolution, missed suppression and duplicate activity before the branch handles a wider segment.

Measure valid-contact rate, time to first action, positive-reply rate and qualified pipeline against opt-outs, complaints, duplicate touches and misrouted replies.¹⁹

Design the stopping behavior with the same care as the send behavior.²⁰ Review every branch that pauses, suppresses or changes a next step. If the event is valid but the action is unclear, leave the record for human review and improve the rule before expanding the trigger.

What not to do

Keep these failure modes beside the branch specification during review.

• Treating intent as a promise of purchase is a mistake. Intent is a prioritization signal with no promise of a purchase.²¹
• Accounts differ in buying stage. Some are actively researching while most are not, so identical effort spends attention on accounts at very different stages.^{22, 23, 24}
• Do not send long, complex emails or skip verification, inbox warm-up and proper ramping. Those choices can damage deliverability before campaigns begin.²⁵
• Check policy before automating social touches. LinkedIn's user agreement prohibits bots and unauthorized methods for adding contacts or sending messages.²⁶