There is no such thing as a generic "advisor AI strategy." A solo RIA charging flat fees out of a converted carriage house in Asheville and a 200-FA wirehouse complex on the 38th floor of an office tower in midtown Manhattan share roughly two things — both must satisfy the SEC Marketing Rule and both have clients who care about Roth conversion math — and almost nothing else about how AI should be selected, deployed, supervised, and disclosed. This lesson installs the practice-type lens: five archetypes (Solo RIA, Ensemble RIA, Multi-Custodian RIA, Wirehouse FA, OSJ / BD supervisor), the operating constraint each faces, the AI strategy that fits, and the regulatory, vendor, budget, and adoption posture each archetype owns. By the end you can put your own practice in one of the five buckets and read the rest of L4 — vendor scorecard, WSPs, cyber, M&A — through the correct lens.

Why Practice Type Is the Primary Segmentation for AI Strategy

The wealth-management AI literature loves to segment by client niche — UHNW, mass affluent, business owners, pre-retirees, women in transition, equity-comp executives — and that segmentation drives the planning content (L2 Ch4, L3 Ch5, L3 Ch6). It is the wrong segmentation for AI strategy. The reason is mechanical: the AI strategy question is not "what does the client need?" but "what does my operating structure permit?" A solo RIA can deploy Jump on Monday and have it archived to Smarsh by Friday. A wirehouse FA cannot deploy Jump at all unless the home office has whitelisted it, even if every one of the FA's clients would benefit. The constraint set is operational, not clinical.

The five archetypes below are the practical operating-structure buckets every US advisor falls into in May 2026. They were validated against the Schwab 2026 RIA Benchmarking Study, the Cerulli channel-share data, and the FINRA-registered-firm count. The archetypes are exhaustive (every practicing US advisor fits one) and mutually exclusive at the firm level (an individual advisor can be dual-hatted, but the firm registration drives the AI constraint).

The decision logic for the rest of this lesson is the same regardless of archetype. For each archetype we will name (1) the operating constraint, (2) the AI strategy that fits, (3) the toolset that is realistically deployable, (4) the regulatory posture (Reg BI, Marketing Rule 206(4)-1, FINRA Rules 2210 / 3110 / 4511, Reg S-P 17 CFR Part 248, NY DFS 23 NYCRR 500, NAIC Model #275), (5) the budget anchor per advisor seat, and (6) the failure mode that kills the strategy. The L4 Ch1 L3 roadmap lesson stitches the archetype-specific strategies into the three-year budget; the L4 Ch2 vendor scorecard reads differently for each archetype.

Archetype 1 — The Solo RIA

The solo RIA is the canonical "one advisor plus maybe a CSA" practice. Roughly 12,000-14,000 US RIA firms have a single registered advisor; thousands more have an advisor plus one or two support staff. The solo RIA is registered with the SEC (if AUM is over $100M, or with a state if under) under the Investment Advisers Act of 1940 and is its own CCO under SEC Compliance Rule 206(4)-7. There is no FINRA registration unless the advisor also holds a BD license through an affiliated firm.

The operating constraint for the solo is time. The solo advisor has 200-400 households, no associate to delegate to, no compliance department to ask, and no in-house IT. Every hour spent on meeting prep, follow-up admin, IPS drafting, Reg BI memo writing, or NIGO rework is an hour not spent on prospecting or higher-leverage planning. The Kitces time-and-task research suggests the solo spends roughly 60-65% of working hours on workflow that produces no billable judgment. The AI strategy that fits is unambiguous: time leverage. Buy AI that compresses workflow. Do not buy AI that requires a deployment project, a vendor-management committee, or a CCO sign-off process the solo doesn't have.

The deployable toolset for the solo in 2026: Jump or Zocks for meeting AI (pick one, not both); Holistiplan for tax extraction; FP Alpha or Wealth.com for estate extraction (pick one); RightCapital or eMoney or MoneyGuidePro for planning (whichever the practice already uses — do not switch); Wealthbox or Redtail for CRM (whichever the practice already uses); Smarsh or Global Relay for archiving; one enterprise LLM (Microsoft Copilot, OpenAI Enterprise, Google Gemini Enterprise, or Anthropic Claude for Work) for general-purpose drafting. That stack is roughly $400-700/month all-in per advisor seat for the AI layer, on top of the existing planning + CRM + archive base.

The regulatory posture for the solo is the most exposed: the solo IS the CCO, the supervisor, the principal reviewer under Marketing Rule pre-use review, and the registered person making the recommendation. There is no separation of duties. The solo's L4 capstone deliverable is the WSP that explicitly names the solo's dual-hatting, the Marketing Rule review queue (which is the solo reviewing the solo's own AI-drafted client communication), the Rule 4511 retention pipeline, and the Reg S-P incident response plan. The solo's ADV Part 2A must disclose AI tool use and update on each tool change.

The failure mode for the solo: tool sprawl. Buying Jump, then trying Zocks, then adding FinMate, then adding Sybill, then layering Pulse360 on top — and never fully committing to a single workflow. The solo's strategy is "ruthless consolidation": one meeting AI, one tax AI, one estate AI, one planning, one CRM, one archive, one general LLM. Seven tools, mastered, retired-and-replaced only on a documented annual review.

Archetype 2 — The Ensemble RIA

The ensemble RIA is the 5-30 advisor firm with shared infrastructure, multiple advisor pods, an ops team, a dedicated CCO (sometimes outsourced to ACA Group or NRS), and usually a managing partner. Roughly 3,000-4,000 US RIA firms sit in this band. AUM typically $200M-$3B. SEC-registered. Self-clearing through Schwab, Fidelity, Pershing, or BNY Mellon Pershing X. Many also hold a state insurance license for annuity sales, which brings the NAIC AI Model Bulletin and Model #275 into the picture.

The operating constraint for the ensemble is advisor consistency. The same client meeting handled by Advisor A and Advisor B should produce comparable IPS quality, comparable Reg BI documentation, comparable follow-up timing, comparable Marketing Rule disclosure language. Without that, the partner who founded the firm has to spend their week being the de facto QA layer — which is exactly the bottleneck the ensemble was supposed to solve. The AI strategy that fits is institutionalization: pick the firm's prompts, the firm's templates, the firm's Marketing Rule disclosures, the firm's WSP, the firm's verification protocol — and bake them into the stack so a new advisor walking in on day 30 produces output indistinguishable from the founding partner's.

The deployable toolset for the ensemble is the solo stack plus: an enterprise LLM with a custom system prompt and a RAG layer pointed at the firm's compliance-approved disclosure library and Reg BI memo templates (L3 Ch9 lesson on RAG-to-firm-vault is the implementation); a CRM with AI-enabled workflow (Salesforce Financial Services Cloud + Einstein, Practifi, or Wealthbox AI); Pulse360 for meeting prep / follow-up routing if the firm has 10+ advisors; Catchlight for lead scoring at the marketing layer; and a vendor-managed prompt library that lives in Notion, SharePoint, or a dedicated tool. The all-in AI layer cost for the ensemble runs $800-1,500 per advisor seat per month at scale.

The regulatory posture for the ensemble is more sophisticated: the CCO (in-house or outsourced) owns the WSP under SEC Compliance Rule 206(4)-7, the principal review queue under Rule 2210 for any FINRA-registered staff (rare in pure RIAs), the Rule 4511 retention pipeline, the Marketing Rule pre-use review queue, the ADV Part 2A AI disclosure, and the Reg S-P 30-day breach response. The ensemble's vendor due diligence (L4 Ch2 L2) is the question set — anchored to SOC 2 Type II, NY DFS 23 NYCRR 500, and the May 2024 Reg S-P amendments — every new tool must clear before any advisor seat is provisioned.

The failure mode for the ensemble: shadow IT. Individual advisors signing up for free ChatGPT accounts, FreePlan Jump, or whatever the latest LinkedIn-promoted tool is — and pasting client NPI into it. The ensemble's WSP must include a prohibited-tools list, a mandatory tool-approval workflow, and an enforcement mechanism (Microsoft Defender for Cloud Apps, Netskope, or equivalent) that prevents shadow access. The SEC Division of Examinations Risk Alerts have flagged shadow AI as a recurring exam finding through 2025-2026.

Archetype 3 — The Multi-Custodian RIA

The multi-custodian RIA is the $1B-$10B AUM firm that holds client assets at Schwab, Fidelity, Pershing, BNY Mellon, and sometimes a fourth (Altruist, TradePMR, Apex). The operating profile looks like an ensemble but the data layer is materially more complex: portfolio data lives in four custodial pipes, billing reconciliation runs across four feeds, the trading desk has to handle four order management systems (Orion Eclipse, 55ip, or a custom built layer), and the AI tools that touch portfolio data — Orion's AI features, BlackRock Aladdin Wealth where applicable, and the planning software's portfolio sync — all have to be fed cleanly.

The operating constraint for the multi-custodian RIA is data integration. The AI strategy that fits is data plumbing first: invest disproportionately in the integration layer (the IBKR/Orion/Tamarac/Black Diamond reconciliation, the custodian portal SSO, the planning software single-source-of-truth) before layering AI on top. AI that runs on bad data is worse than no AI; the hallucination rate compounds with data drift. The L1 Ch2 Cardinal Rule lesson on source-system verification becomes load-bearing here — the verification step is not optional, and the only way to scale it is to have one canonical data source per data domain.

The deployable toolset is the ensemble stack plus: Orion Eclipse with the AI rebalance features turned on, integrated to the planning software; 55ip for tax-aware sleeves with their AI optimization; a dedicated data integration platform (Beacon, Hubly, or a custom internal solution); and an enterprise LLM with a RAG layer pointed at the firm's IPS templates AND the firm's portfolio-data dictionary so the model can correctly translate "VTI" into "Vanguard Total Stock Market ETF" without hallucinating the share class. Budget anchor runs $1,200-2,000 per advisor seat per month for the AI layer.

The regulatory posture adds the cross-custodian recordkeeping wrinkle under SEC Rule 204-2: the firm has to maintain books and records aggregated across all custodians, and the Smarsh / Global Relay archive must capture communications regardless of which custodian's portal the advisor was on when the email went out. The Reg S-P vendor oversight obligation under the May 2024 amendments applies to each custodian PLUS each AI vendor PLUS each integration vendor — the vendor inventory (L4 Ch4 L1) is the deliverable that survives an SEC exam.

The failure mode: integration debt. The firm buys a new AI tool that promises a Schwab integration, deploys it, then discovers it only supports Schwab Advisor Services for accounts opened after 2018 — so 30% of the book is silently outside the AI's scope, and the advisors learn to "just remember" which accounts the tool doesn't see. The remediation is the L4 Ch2 L3 POC pilot design: never deploy across the full book without a parallel-run validation against every custodian feed.

Archetype 4 — The Wirehouse FA

The wirehouse FA practices at Morgan Stanley, Merrill Lynch, UBS, Wells Fargo Advisors, Raymond James (mixed-channel), or LPL (technically independent but with home-office tool gating that operates similarly). The FA holds Series 7 and 66, is FINRA-registered, and is dual-hatted under Reg BI (broker-dealer side) and the Marketing Rule (investment-adviser side) for the same client conversation. The home office owns the technology stack, the WSPs, the Marketing Rule pre-use review queue, the Smarsh archive, the prohibited-tools list, and the supervisory architecture under FINRA Rule 3110.

The operating constraint for the wirehouse FA is home-office gating. The FA cannot deploy Jump or Zocks unless Morgan Stanley's tech committee has whitelisted them. The FA cannot paste a client's tax return into a public LLM under any reading of the firm's WSP. The FA cannot use a personal ChatGPT subscription on a firm device. The AI strategy that fits is operate within the whitelist and optimize the workflow inside it. That sounds limiting, and it is — but every major wirehouse by May 2026 has at least one approved meeting AI (Morgan Stanley with the OpenAI partnership; Merrill with the Bank of America-internal LLM; UBS with the Microsoft Copilot deployment), a planning sync layer, a CRM with AI workflow features (Salesforce FSC + Einstein on the AML/BSA / Reg BI workflow side), and an enterprise LLM (Microsoft Copilot is the dominant choice).

The deployable toolset for the wirehouse FA is whatever the home office has approved. Period. The FA's L4 capstone deliverable is not a vendor selection — it is a written workflow that uses the approved tools to compress the FA's hours-of-admin into hours-of-prospecting-and-planning. The FA's Marketing Rule disclosure language is set by the home office. The FA's Reg BI memo template is set by the home office. The FA's role is to be the best operator of a constrained toolset and to feed the home office's product team feedback on what the next round of approved tools should be.

The regulatory posture is the most layered: FINRA Rule 2210 principal review of every client communication, Rule 3110 supervision through the OSJ and complex manager, Rule 4511 retention through the firm's Smarsh / Global Relay implementation, Reg BI documentation flowing into the firm's product-specific Reg BI templates, Marketing Rule pre-use review for any FA-customized content. The FA's personal accountability under Form U4 (the Disclosure Reporting Page lessons in L5 Ch7) and the firm's Compliance program under SEC Rule 206(4)-7 (for the IA side) are both in play.

The failure mode: workaround creep. The FA who decides "the home office's tool is slow, let me just use ChatGPT from my phone for the prep" is the FA who shows up in a 2026 enforcement action. The strategy is to advocate for the FA team's needs through the home office's product feedback channel, not to circumvent the gating.

Archetype 5 — The OSJ and BD Supervisor

The OSJ (Office of Supervisory Jurisdiction) is the FINRA-required supervisory structure for an independent broker-dealer's branches. The OSJ principal is the registered person who supervises the producing reps — handling Rule 3110 supervisory obligations, Rule 2210 principal review, Reg BI compliance documentation, and the FINRA examination interface. The BD supervisor at LPL, Raymond James Financial Services, Cetera, Cambridge, Commonwealth, and the other independent BDs typically supervises 10-50 producing advisors across multiple offices.

The operating constraint for the OSJ is supervision at scale. The OSJ cannot personally read every email every producing rep sends, every Reg BI memo every rep drafts, every Marketing Rule communication every rep distributes. The FINRA 2026 Annual Regulatory Oversight Report's framing of agentic AI under Rule 3110 reasonable-design makes the supervisory architecture obligation explicit: the OSJ has to be able to demonstrate, to a FINRA examiner, that the supervisory system is "reasonably designed" to catch the violations it is supposed to catch. AI is the only way to do that at the scale the OSJ operates at — the supervisor's AI strategy is to use AI to supervise AI.

The deployable toolset for the OSJ: Smarsh or Global Relay with the AI search and policy-detection layers turned on (the L3 Ch10 archive pipeline lesson is the operational predicate); a Marketing Rule pre-use review queue that uses an AI first-pass to flag the 1-in-100 communication needing human review (L4 Ch3 L2 design); a Reg BI memo audit tool that samples and analyzes rep memos for the four documented alternatives; a CRM with AI-enabled supervisory dashboards; and a dedicated supervisory log that captures the human-reviewer signoff under Rule 4511. Budget anchor for the OSJ layer runs $1,500-3,000 per supervised advisor per month, justified by the alternative (an enforcement action) being orders of magnitude more expensive.

The regulatory posture is the densest: FINRA Rules 2210 / 3110 / 4511 are core; SEC Rule 204-2 applies if the firm is dually-registered as an IA; Compliance Rule 206(4)-7 governs the IA program; Reg S-P and the May 2024 amendments govern the cyber program; NY DFS 23 NYCRR 500 applies if the firm has any NY-licensed business; the NAIC AI Model Bulletin and Model #275 apply to any annuity-licensed reps. The OSJ's L4 capstone is the WSP that ties all of this together (L4 Ch3 L1 is the model).

The failure mode: AI-supervises-AI hallucination. The OSJ that trusts the AI first-pass review without spot-checking it is the OSJ whose system FINRA characterizes as "not reasonably designed" in the next examination. The mitigation is the L4 Ch3 L2 exception-handling protocol: sample the AI's "clean" outputs, not just its "flagged" outputs, and validate the false-negative rate.

Cross-Cutting Decisions and the Handoff to the Three-Year Roadmap

Regardless of archetype, four decisions cut across every AI strategy and feed into the L4 Ch1 L3 roadmap and the L4 Ch2 vendor scorecard.

First, the enterprise LLM choice. Microsoft Copilot, OpenAI Enterprise, Google Gemini Enterprise, or Anthropic Claude for Work. The wirehouse FA does not have this choice. The solo, ensemble, multi-custodian, and OSJ do. The choice is driven by (a) the firm's existing productivity stack (Microsoft 365 → Copilot; Google Workspace → Gemini Enterprise), (b) the SOC 2 Type II report quality (L4 Ch2 L1), (c) the data-residency and Reg S-P posture, and (d) the integration with the planning + CRM + archive layer. None of the four is "wrong"; all four have advisor-deployable enterprise tiers in 2026.

Second, the meeting AI choice. Jump and Zocks dominate the category — Schwab's 2026 RIA study and the Kitces March 2026 AdvisorTech map both confirm. FinMate, Sybill, Zeplyn, and Pulse360 are credible second-tier options. The solo should pick one and master it; the ensemble should standardize firm-wide; the multi-custodian should validate the integration to each planning software; the wirehouse FA uses whatever the home office approved.

Third, the archive architecture. Smarsh or Global Relay (the solo can sometimes use ACA Group's archive or NRS for smaller scale). The archive is the foundation of the Rule 4511 retention pipeline and the AI-supervised review queue. The L3 Ch10 Zocks-to-Wealthbox-to-Smarsh pipeline lesson is the operational predicate; the L4 Ch3 L3 agentic-AI WSP lesson extends the architecture to action-taking AI.

Fourth, the ADV Part 2A disclosure trigger. When AI tool use changes materially — new vendor, new data category, new client-facing application — the ADV is updated promptly under the SEC's prompt-amendment rules. The L5 Ch7 ADV-amendment lesson is the operational workflow; the practice-type lens determines who in the firm owns it (the solo is their own filer; the ensemble has a CCO; the wirehouse has the home office).

By the end of L4 Ch1 you will have placed your practice in one of the five archetypes, you will have run the readiness audit (Ch1 L2), and you will have a three-year roadmap and budget (Ch1 L3) anchored to the archetype-specific strategy. The L4 Ch2 vendor scorecard, the L4 Ch3 WSPs, the L4 Ch4 cyber playbook, and the L4 Ch5-Ch8 chapters (training, risk, Marketing Rule, M&A) all read differently for each archetype — but they all share the same regulatory floor (Reg BI, Marketing Rule, FINRA 2210/3110/4511, Reg S-P, NY DFS 500, NAIC #275) and the same Cardinal Rule discipline (L1 Ch2.3).

Key Takeaways

• Practice type, not client niche, drives AI strategy. The five archetypes — Solo RIA, Ensemble RIA, Multi-Custodian RIA, Wirehouse FA, OSJ / BD Supervisor — each have a different operating constraint and a different deployable toolset.
• Solo RIA strategy is time leverage: ruthless tool consolidation, $400-700/seat/month AI layer, one tool per category, the solo as their own CCO under SEC Compliance Rule 206(4)-7.
• Ensemble RIA strategy is institutionalization: firm-wide prompts, RAG to firm vault, shadow-IT prevention, $800-1,500/seat/month, vendor due diligence (L4 Ch2) anchored to SOC 2 Type II and Reg S-P May 2024 amendments.
• Multi-Custodian RIA strategy is data plumbing first: $1,200-2,000/seat/month with disproportionate spend on the integration layer and Orion Eclipse / 55ip portfolio AI.
• Wirehouse FA strategy is operate within the whitelist: home office gates the toolset; the FA optimizes the workflow inside it and advocates through product feedback.
• OSJ / BD Supervisor strategy is AI-to-supervise-AI: Smarsh / Global Relay with AI search, AI first-pass on Marketing Rule review (FINRA Rule 2210), Reg BI memo audit tooling, $1,500-3,000 per supervised advisor per month — justified by the cost of a Rule 3110 reasonable-design exam finding.
• Four cross-cutting choices apply to all archetypes: enterprise LLM (Microsoft Copilot / OpenAI Enterprise / Google Gemini Enterprise / Anthropic Claude for Work), meeting AI (Jump or Zocks), archive (Smarsh / Global Relay), and ADV Part 2A disclosure trigger.
• The archetype lens flows into the rest of L4: the readiness audit (Ch1 L2), the three-year roadmap and budget (Ch1 L3), the vendor scorecard (Ch2 L1), the WSPs (Ch3), the cyber playbook (Ch4), and the M&A defensibility framing (Ch8 — top-quartile RIAs trade at 8x-10x adjusted EBITDA per Mercer Capital / ECHELON Q3-Q4 2025, with AI maturity adding 0.5-1.5x to the multiple).