Tasks, Notes and Manual Actions

Tasks are the handle automation has on people. When a workflow reaches something only a human can do, it creates a task: a titled item with an owner, a due date and a link back to the contact or opportunity it concerns. Notes are the other half of the same idea, a written record attached to a contact of what happened, what was said and what was decided.

Most builds under-use this module and suffer for it. Automation that only sends messages produces a client who receives leads and still misses them, because nothing told a specific person to do a specific thing by a specific time. A task converts an automated journey into an accountable one. It is also the safety net for everything AI cannot finish: the ambiguous classification, the escalated conversation, the review that needs a human reply.

The mental model is a work queue with provenance. Every task should say what to do, to whom, by when, and carry enough context that the assignee does not have to reconstruct the history. The failure mode is a queue nobody works: tasks created by workflows, assigned to nobody in particular, with no due date, accumulating until the whole list is ignored. A task system is only as good as the discipline around closing items, and designing for that discipline is part of the build, not the client's problem afterwards.

The task

• Title and description, with context merged in from the contact, opportunity or conversation
• Assigned user, or an assignment rule such as the contact's owner or a round robin
• Due date and time, usually set relative to creation
• Priority or type, where the account distinguishes them
• The linked record, so opening the task opens the history

Creation and automation

• Workflow actions that create a task at any point in a journey
• Triggers on task completion, so finishing a task can advance the flow
• Overdue handling: reminder, escalation to a manager, reassignment
• Whether the workflow waits for the task or continues regardless

Notes

• Free-text notes on the contact, written by a user or by a workflow
• AI-generated call and conversation summaries written as notes
• Visibility, and whether notes are internal only

Manual action steps

• A step in a workflow that pauses for a human to act, such as a call to make or a document to review
• What happens on completion, on skip, and on timeout

Views and accountability

• Per-user task lists, team views and filters by due date, status and owner
• Notifications on assignment and on overdue
• Reporting on open, overdue and completed volume by user

1. Decide with the client which steps in their process genuinely require a person. Everything else should be automated, and everything on this list becomes a task.
2. Agree who owns tasks by type before you build. An unassigned task is a task nobody does.
3. Confirm the users exist in the account with the right permissions, and that they will actually see notifications. A task assigned to a dormant login is invisible work.
4. Write task titles as instructions in the imperative: call this lead about their inquiry, review this negative review, confirm this appointment. A title that names a state rather than an action gets skipped.
5. Put the context in the description using merge fields: what the contact asked for, which campaign they came from, what the AI summary said. The assignee should not need to go hunting.
6. Set due dates relative to creation and make them short enough to mean something. A task due in two weeks is a task nobody treats as urgent.
7. Add the task creation steps into the workflows at the points where automation runs out: AI handover, failed classification, negative review, stale opportunity, no reply after the final chase.
8. Decide per workflow whether the flow waits for the task or continues. Waiting is right when the human step gates the next message; continuing is right when the task is a parallel safety net.
9. Build the overdue path: a reminder, then an escalation to a manager, then reassignment. Without this, the queue silently fills.
10. Configure assignment rules so routing survives absence, and test what happens when the assigned user is away.
11. Decide what gets written as a note automatically: AI call summaries, conversation summaries, workflow outcomes. Keep automated notes short and clearly labeled so they do not bury what a human wrote.
12. Agree the closing convention with the client: a task is closed when the action is done and the outcome is recorded, not when it is read.
13. Test the full loop with a real user account, not an admin one: creation, notification, opening, completing, and the workflow continuing afterwards.
14. Review open and overdue volume with the client after the first two weeks. A growing overdue count means either the queue is too noisy or the ownership is wrong, and both are build problems you can fix.

Tasks are created by Workflows and attached to Contacts and Opportunities. They are the standard destination for every AI handover, and they are what makes Pipelines honest by forcing a decision on stale cards.

AI handover with accountability

The conversational agent escalates. A workflow creates a task for the contact's owner titled with the reason, carrying an AI summary of the conversation, due within the hour. If it is not closed, a reminder fires and then the manager is notified. The escalation is no longer a notification in a feed that nobody opened.

Stale opportunity review

A workflow finds opportunities that have not moved and creates a task for each owner to progress or close with a reason. Completing the task advances or closes the card. This is the mechanism that keeps the pipeline total from drifting into fiction.

High-value lead manual call

A lead matching defined criteria enters a workflow that sends the automated first touch and simultaneously creates a call task for a named rep, due in fifteen minutes, with the inquiry text in the description. The nurture sequence continues in parallel but exits the moment the rep logs a connection.

• Unworked queues are the default outcome. Tasks created with no owner, no due date or no clear action accumulate until the list is abandoned. Every task you generate needs all three.
• Volume kills adoption faster than anything. A workflow creating a task for every lead will produce hundreds a week. Condition task creation tightly and reserve it for steps that genuinely need a person.
• Completion means nothing by itself. Marking a task done does not record what happened. If the outcome matters, require a note, a field update or a pipeline move as part of closing it.
• Workflows that wait on a task can stall forever. If the task is never completed the contact sits in the workflow indefinitely. Always add a timeout path.
• Notifications are easy to miss. Email notifications get filtered and in-app badges get ignored. Confirm with the client how their team will actually see tasks, and test it on their real devices.
• Assignment to an absent user is invisible. Holidays, departures and deactivated logins strand tasks. Build reassignment rather than relying on someone noticing.
• Automated notes drown human ones. A note per workflow run makes the contact timeline unreadable. Log only what someone will need, and label it.
• Notes are not a secure place for sensitive detail. Anyone with contact access can read them, and they travel on exports and merges. In regulated industries, agree with the client what may be written in a note at all.
• Tasks are not a project tracker. They are good for short, single-step, contact-linked actions. Multi-step internal work belongs in the client's own tooling, and forcing it here produces a queue nobody trusts.
• Overdue reporting is the real health metric. Open count tells you little; overdue count and age tell you whether the human half of the build is working. Watch it from go-live.
• Merges and deletions take tasks with them. Merging duplicate contacts can move or lose linked tasks and notes. Check after any bulk hygiene work.