The CSA's New Job: Supervising 3 AI Agents Instead of Doing 30 Tasks

The CSA role at growing RIAs is not shrinking. It is shifting.

From doing 30 tasks to supervising 3 agents that handle them. The CSA's judgment, relationship context, and exception-handling remain essential. The repetitive execution — copy-paste between systems, triggered sequences, templated emails that follow a consistent pattern — that's what agents are absorbing.

A small set of RIA firms are running this now. Here's what the transition looks like, what the Three-Agent CSA Stack consists of, and how to build it on tools most firms already own.

The Task-Tier Breakdown

The typical CSA at a 100-150 household RIA manages somewhere between 25 and 40 recurring task types per week. Mapped by decision requirement, they fall into two tiers:

Execution tasks (agent-appropriate): Tasks that follow a consistent trigger → action → output pattern with no situational judgment required. Drafting post-meeting follow-up emails, sending DocuSign envelopes on status change, logging confirmed transfers to the CRM, queuing annual review scheduling tasks, generating client notification drafts for routine account activity.

Judgment tasks (human-required): Tasks where the right response depends on relationship context the system doesn't hold. Handling a client call with unexpected emotional content, escalating a compliance concern, recognizing that a "routine" account transfer is part of a broader estate situation, determining which meeting action item needs immediate follow-up versus can wait a week.

Based on workflow mapping across RIA operations, the split is roughly 60-65% execution tasks and 35-40% judgment tasks by volume. Agents handle the execution tier. The CSA supervises agent output and owns the judgment tier.

This isn't the CSA doing half the work. It's the CSA doing different, higher-quality work — and covering more client relationships because execution is no longer the bottleneck.

The Three-Agent CSA Stack

The configuration that maps to a realistically deployable first version at most RIA firms involves three agents, each handling a distinct category of execution work.

Agent 1: The Follow-up Agent

What it does: After every client meeting, the follow-up agent reads the transcript, extracts client commitments and action items, drafts the recap email, and queues tasks in the CRM assigned to the appropriate team member by name.

What the CSA does: Reviews the draft email, confirms action items are correctly attributed, approves or edits, sends. Time: 2 minutes per meeting instead of 15-20.

Tool configuration:

• Transcription: Fireflies.ai Business (Zoom, Teams, or in-person)
• Extraction: Fireflies AskFred feature or Zapier multi-step Zap to structured output
• CRM write-back: Zapier → Redtail Create Note + Create Task, or Wealthbox equivalent via API
• Email draft: LLM API call (with a signed data processing agreement) → Advisor approval queue in CRM

What the CSA supervises: Does the email reflect the actual conversation? Are action items complete? Is anything missing that the agent wouldn't know to include?

Agent 2: The Onboarding Sequencer

What it does: When a new engagement is signed, the sequencer fires: DocuSign new account envelope sends, welcome email drafts for advisor approval, 30-day check-in task queues, 90-day review task queues — all from a single CRM trigger. The CSA receives a review task rather than a setup task.

What the CSA does: Reviews sequence outputs, handles exceptions (DocuSign bounce, incorrect template, client requests a change). Time: 10 minutes of review per new client versus 45-90 minutes of manual setup.

Tool configuration:

• Trigger: Redtail Automation (Contact Status → New Client) or Wealthbox Workflow Template (on status change)
• DocuSign: Native integration in Redtail (Apps → DocuSign) and Wealthbox (Wealthbox Connect)
• Email draft: CRM template with field substitution → advisor approval queue
• Tasks: Created automatically by the CRM trigger with +30 and +90 day due dates, assigned to relationship CSA

What the CSA supervises: Correct template for this client type? DocuSign confirmed sent? Tasks assigned to the right team members?

Agent 3: The Account Monitor

What it does: Monitors Schwab and Fidelity data feeds for account activity warranting advisor awareness — large transactions, unusual allocation shifts, transfer activity. Surfaces flagged items to the CSA review queue before the advisor is notified.

What the CSA does: Reviews the flag, determines whether it warrants immediate advisor escalation or standard workflow routing, acts accordingly. Time: 3-5 minutes per flag review.

Tool configuration:

• Data feed: Schwab Advisor Services nightly feed to Redtail or Wealthbox (or via Orion/Black Diamond, which normalize the data to the CRM)
• Alert rule: CRM Automation triggered on account field changes exceeding a defined threshold (transaction value, account status change)
• Surfacing: CRM task in CSA queue with account context and client record link populated automatically

What the CSA supervises: Is this flag worth escalating? What context does the advisor need before the client calls?

What the CSA's New Job Looks Like

In practices running the Three-Agent CSA Stack, the CSA's daily work shifts not in intensity but in type.

Before: Attention split between executing sequences (drafting, triggering, queuing) and managing the client relationship. A significant portion of each day spent on work the CSA could do without thinking.

After: Attention focused on supervision, exception handling, and the 35-40% of tasks that require judgment. The execution layer is covered. The work remaining is the work the CSA is actually better at.

The transition also changes scope. With execution overhead reduced, a CSA running three agents can support more advisor-client relationships — not by working more hours, but because each hour goes to higher-leverage work.

How to Build It in Sequence

Month 1 — Follow-up Agent: Start here. Every firm has meetings. Fireflies for transcription, Zapier for CRM write-back, a simple LLM API call for the email draft. The CSA begins supervising output immediately and can calibrate quality from the first week.

Month 2 — Onboarding Sequencer: Configure the CRM automation (Redtail) or workflow template (Wealthbox) for new client status change. Activate the DocuSign integration if not live. Test with the next 2-3 new clients before relying on it for the full pipeline.

Month 3 — Account Monitor: Confirm the Schwab or Fidelity data feed writes to CRM account fields. Build the CRM alert rule. Start with one threshold — a single transaction over a defined dollar amount — before adding complexity.

---

Frequently Asked Questions

Does supervising agents require technical skills from the CSA?

No. Agent outputs land in the CRM as tasks, notes, and email drafts — interfaces the CSA already uses daily. Supervision means reviewing output and handling exceptions, not configuring the agent. Configuration is a one-time setup project; ongoing use requires no technical knowledge beyond current CRM proficiency.

What happens when an agent makes a mistake?

Every output from the three agents routes through a human review step before client contact. The follow-up email is approved before it sends. The onboarding sequence is reviewed before the CSA considers it complete. The account flag is reviewed before the advisor is notified. Mistakes surface in the review step, not in client experience. The agent handles the predictable path; the human handles variance.

Is this compliant with SEC supervision requirements?

Yes, when built correctly. The advisor approves all client-facing communications generated by agents before they send. The CSA reviews all account activity flags before escalation decisions are made. The review step is logged in the CRM as a dated activity. The supervision structure is equivalent to reviewing any other staff-generated communication.

---

Key Takeaways

• The Three-Agent CSA Stack: Follow-up Agent (transcript → CRM note + email draft), Onboarding Sequencer (trigger → DocuSign + tasks), Account Monitor (data feed → CSA review queue)
• Task-tier analysis: roughly 60-65% of CSA task volume is agent-appropriate execution; 35-40% requires human judgment — agents handle the first tier, CSAs own the second
• All three agents use tools most RIA firms already have: Fireflies, Zapier, Redtail or Wealthbox, Schwab or Fidelity data feeds, DocuSign
• Build sequence: Follow-up Agent (Month 1) → Onboarding Sequencer (Month 2) → Account Monitor (Month 3)
• The CSA managing three agents isn't doing less — they're doing higher-leverage work and can cover more client relationships without the execution bottleneck

Ready to build the Three-Agent CSA Stack for your firm? Book a discovery call with the Systemaic team.