Context
A wealth-advisory firm with a growing AUM book and a fixed advisor team was hitting a response-time ceiling. Advisors were fielding a mix of routine client messages (account status, statement requests, market-commentary queries) and substantive ones (planning questions, tax-implication discussions, product inquiries). The routine messages consumed enough advisor time that advisor-capacity was being used on work that did not require advisory judgment, and the substantive messages were getting slower response because of the volume.
The firm had considered outsourcing routine response to a service center; the compliance and relationship-quality risks made that a non-starter for a wealth-advisory model. What the firm wanted was an AI copilot that drafted responses advisors could review and send in seconds instead of minutes — with the compliance posture required to operate in a FINRA- and SEC-regulated environment.
Challenge
The requirements:
- Advisor-facing, not client-facing. AI draft, human review, human send. No autonomous client communication.
- Compliance-aware retrieval. Every response had to reference the firm's current compliance guidance and product documentation — not yesterday's, not generic industry material.
- Per-response audit trail. The audit log had to meet FINRA books-and-records and SEC supervision requirements on its own merits, without additional engineering to satisfy examiners.
- CRM-native. The draft had to appear in the existing CRM workflow; advisors would not adopt a separate app.
Approach
Thoughtwave deployed TWSS Finance AI/ML — our finance-tuned variant of the TWSS CS Agent — connected to the firm's Salesforce Financial Services Cloud instance and to its compliance and product-documentation knowledge stores. The platform runs the same five-stage pipeline as the base CS Agent, with finance-specific adjustments:
- Ingest. New client messages land in the case record in Salesforce Financial Services Cloud; the platform's CRM connector receives a webhook.
- Understand. A finance-tuned intent classifier identifies the case type with higher accuracy on advisory-domain intents than a general model.
- Retrieve. Grounded retrieval across compliance sources, product documentation, and relevant market-commentary assets — with every cited source captured.
- Generate. A compliance-aware draft response is created with per-source citations attached. The tone policy matches the firm's client-communication standard.
- Audit and sync. The draft lands in the case record; the advisor reviews, edits, and sends; every step is logged for supervision.
Engagement arc:
- Weeks 1-3. Inventoried the client-message intent taxonomy; cataloged compliance and product-doc knowledge sources; agreed the high-risk intent categories that would route straight to an advisor.
- Weeks 4-7. Deployed the platform in shadow mode against a sample advisor cohort; compliance and supervision reviewers evaluated the audit log.
- Weeks 8-10. Enabled production drafts in the broader advisor team; tuned confidence thresholds and the tone policy based on advisor feedback.
What we built
The production system has five components:
- Salesforce connector. Webhook-driven intake and write-back into the Financial Services Cloud case record.
- Finance-tuned classification. Intent, urgency, and risk-category detection adapted to advisory-domain messages.
- Compliance retrieval layer. MCP-based providers for the firm's current compliance guidance and product-documentation stores; version-aware so a draft always reflects the current policy.
- Draft and citation engine. LLM draft with source citations attached; tone policy matched to the firm's communication standard.
- Supervision-grade audit. Append-only log, advisor attribution, and retention matched to FINRA/SEC obligations.
Outcomes
- Faster client response time across the routine message mix. Advisors spend time sending instead of drafting.
- Reduced compliance risk per message through grounded retrieval and per-source citation. The draft reflects current policy; the advisor sees the sources.
- Higher advisor productivity. The ceiling on client volume per advisor shifts upward without changing the advisor team.
- FINRA/SEC-ready audit trail on 100% of responses, supporting supervision workflow out of the box.
What's next
The next phase extends the platform to internal advisor-to-operations messaging and to the firm's shared-inbox workflows (using TWSS AI Email Assistant as the companion module). A future phase evaluates client-portal copilot access for low-risk informational queries, with a separate guardrail configuration and a more conservative confidence threshold.
For broader context, see the TWSS CS Agent retail-safety case study, our Cybersecurity Solutions service, and the accelerators portfolio.
Why this architecture works for regulated industries
The core insight is that the compliance posture is a feature, not a constraint. An AI response that cites its sources, that is reviewed before it is sent, and that produces a complete audit log is a better response than one that does not — even outside regulated contexts. In regulated industries, it is the precondition for deployment. Wealth advisory, banking, insurance, and healthcare all have the same shape of requirement: grounded retrieval, human-in-the-loop, and supervision-grade audit. The TWSS Finance AI/ML and TWSS CS Agent architecture is designed for that shape, not retrofitted to it.