Skip to Content
LeadsPipelinesAI agents and workflows

AI agents and workflows in pipelines

FitStack uses AI at the stage level. When a Lead enters an agentic stage, the assigned published workflow starts and the selected agent drives that workflow toward the stage goal.

The workflow supplies the repeatable cadence and actions. The agent supplies the communication identity and enabled channels. The stage supplies the mission, success destination, and optional failure destination.

Who controls what

Decision areaHuman administrators and staffConfigured AI agent
Pipeline design
Human administrators and staffChoose the type, stages, order, probabilities, stale policy, and terminal outcomes
Configured AI agentNo independent authority
Stage assignment
Human administrators and staffChoose Manual or assign one published workflow and one capable agent
Configured AI agentRuns only the assignment saved for that stage
Goal and routing
Human administrators and staffWrite the goal and choose success and optional failure destinations
Configured AI agentUses the goal and may advance only within that configured context
Communication
Human administrators and staffApprove channels, sender configuration, content boundaries, and when to intervene
Configured AI agentExecutes published steps through channels enabled for that agent
Exceptions and judgment
Human administrators and staffReview replies, consent, sensitive cases, stale Leads, and unexpected outcomes
Configured AI agentCan continue its configured work; it does not replace accountable human judgment
Outcome review
Human administrators and staffConfirm the Lead belongs in the next stage and close Won, Lost, or Disqualified deliberately
Configured AI agentRecords activity and can report a configured goal reached

AI support does not remove human accountability. Staff should review consent, sensitive or unexpected replies, stale Leads, exceptions, and every terminal outcome that requires business judgment.

Four pieces of an agentic stage

  1. Published workflow — the approved sequence of outreach, waits, reply handling, logging, and completion.
  2. AI agent — the specific agent that communicates through the workflow’s required channels.
  3. Stage goal — the plain-language outcome injected into the agent’s work.
  4. Routing — the required success stage and optional failure stage.
Demo Prospect pipeline editor with AI Engagement stage expanded to show workflow, agent, goal, and routing
  1. 1Choose the published workflow that owns this stage on entry.
  2. 2Select one agent with every channel the workflow requires.
  3. 3Write one observable goal in plain language.
  4. 4Route success and optional failure deliberately.
An agentic stage binds one workflow and one agent to a human-defined goal and destination.

Choose or build a workflow

Open Leads → Workflows. You can clone a platform template or create a blank workflow. Templates are starting points, not active assignments.

Edit the draft, review its channels and paths, then publish it. A draft never executes. Only an active workflow with a published version can be assigned to a pipeline stage.

Lead workflow library in the Demo Admin with custom workflows and templates
  1. 1Start blank only when no template fits the operating goal.
  2. 2Custom workflows are editable definitions owned by your team.
  3. 3Templates are proven starting structures; clone before editing.
  4. 4View the sequence and required channels before choosing it.
Clone a proven template or build a custom workflow, then publish it before assigning it to a stage.

What happens after stage entry

  1. FitStack checks for the stage’s assigned published workflow and agent.
  2. If the Lead is eligible and is not already active in that workflow, the workflow enrolls the Lead.
  3. The agent follows the published steps and uses the stage goal as its mission.
  4. Calls, messages, replies, and supported outcomes are recorded as activity. A workflow wait advances enrollment execution but does not appear as a Lead Activity timeline record.
  5. When the goal is reached, the agent can advance the Lead only in the full context of this stage and pipeline.
  6. If the person engages without reaching the goal, the Lead stays for ongoing conversation or human follow-up.
  7. If the workflow completes without response, the Lead follows the optional failure destination or stays for manual or stale handling.

Lead-pipeline SMS steps wait for the Lead club’s quiet-hours window. Email and Call steps do not. A person who replies during quiet hours can be answered immediately. See Quiet hours and messaging.

Demo Sales Engagement workflow showing AI opener, wait for reply, no-response logging, and completion
  1. 1The published sequence starts with its configured opener.
  2. 2A wait gives inbound engagement time to arrive.
  3. 3The no-response branch records the operating result.
  4. 4Completion ends this workflow path; stage routing remains separate.
View the workflow graph from the stage assignment to confirm its sequence and no-response path before saving.

A Lead cannot be enrolled twice in the same workflow while that workflow is pending or in progress. An in-flight run keeps the published workflow version it started with.

Inbound replies when the stage is manual

The Inbound Fallback Agent is not an outbound stage worker. It answers replies when the Lead’s current stage has no AI agent. It has no stage goal and does not advance the Lead. If no pipeline fallback is selected, the brand default agent answers under the same goal-less boundary.

The reply order is:

  1. current stage workflow agent, when active and capable;
  2. pipeline Inbound Fallback Agent;
  3. brand default agent.

Human intervention and changes

Staff can intervene whenever judgment or a handoff is needed. A manual stop does not trigger an automatic downstream action.

Changing a stage assignment affects future stage entrants:

  • switching workflows does not rewrite a workflow already in progress;
  • switching to Manual does not stop a workflow already running;
  • publishing a new workflow version does not change an in-flight run.

Review active Leads before making a change, then decide whether any running work needs a separate human stop or follow-up.

Last updated on
Design stages and ownershipCreate a pipeline