The single largest dollar of value a senior advisor produces in a fifteen-year career is almost certainly delivered in a two-minute conversation on a Tuesday afternoon in a drawdown — the one where a 67-year-old retired client, watching the S&P down 22% off its peak on CNBC, calls and asks whether they should "go to cash for a while." Get that conversation right and the household's plan holds. Get it wrong, or get it generic, and the next five years of Monte Carlo gets re-anchored to a behavioral mistake no spreadsheet can ever recover. This lesson installs the AI-guided drawdown-coaching protocol: pull the household's current position on its own success-probability curve from eMoney / RightCapital / MoneyGuidePro, fuse the curve with today's market level, and generate a personalized 60-second talking script the advisor can deliver from memory — Reg BI-suitable, Smarsh-archived, CRM-logged, and unlike the firmwide "everyone gets the same drawdown email" blast that defines the industry's failure mode in every crisis since 2008.

Why the Generic Drawdown Email Is the Industry's Quiet Failure Mode

Walk into any wirehouse or sizable RIA on the third red Friday of a correction and you will find the same artifact in everyone's outbox: a 400-word "perspective on market volatility" letter, drafted by a CIO office or a marketing team, blast-emailed to the entire client base, signed by the advisor. It is a category of communication that has survived every regulatory regime since the Investment Company Act because it is so anodyne that nothing in it can be called a recommendation under §240.15l-1. That same anodyne quality is the problem. A 38-year-old accumulator with $180,000 in a Roth IRA and 30 years to retirement is getting the same letter as the 72-year-old retiree drawing $9,200 a month from a $1.6M portfolio whose Monte Carlo plan was built on a 4.2% real return assumption that is no longer holding. One household needs to be reminded to keep contributing; the other needs to be told, very specifically, whether the dynamic-spending rule has been triggered and what the trigger means in plain English. The generic email serves neither and burns advisor credibility with both.

The senior-advisor failure mode the AI-guided protocol replaces is older than any large language model. It is the failure to translate the household's own plan — the eMoney scenario, the RightCapital success-probability heatmap, the MoneyGuidePro confidence-zone diagram — into a conversation the client can actually hear in the moment the conversation is needed. The plan has always contained the answer. The advisor has never had the working memory to surface it at 2:47 p.m. on a Tuesday for fifty different households in the same afternoon. That is the problem AI solves.

The Coaching Protocol, End-to-End

The drawdown-coaching protocol has six stages, each tied to a system of record and a regulatory artifact. The whole protocol runs in roughly four minutes per household once it is configured, which is what allows a single senior advisor to deliver a personalized coaching conversation to seventy retiree households in a single afternoon during a peak-VIX week.

Stage 1 — Pull the Household's Current Monte Carlo Position

The first input is the household's last-saved planning scenario from whichever planning system the firm uses — eMoney Advisor Pro, RightCapital, or MoneyGuidePro Elite. The output the protocol needs is not the headline probability of success (the "91%" or "78%" number the client remembers from their last review). The output the protocol needs is the household's location on the curve: the slope, the probability-of-success at today's market level, the distribution of terminal portfolio values, the year-by-year withdrawal pattern the plan assumed, and the dynamic-spending rule (if any) embedded in the plan. eMoney's plan-export API and RightCapital's scenario-export feed structured JSON the AI layer can consume directly. MoneyGuidePro's confidence-zone export is structurally similar. The advisor's CRM (Wealthbox, Redtail, Salesforce Financial Services Cloud) holds the link between the household record and the latest-saved scenario; the integration pulls the most recent plan rather than a stale one.

Stage 2 — Fuse the Plan Position With Today's Market Level

The household's plan was built on a portfolio value that no longer exists. Today's value is in Orion Eclipse, BlackRock Aladdin Wealth, or directly at the custodian (Schwab, Fidelity, Pershing). The AI fuses the two: given the plan's success-probability surface and today's portfolio value, where on the surface does the household actually sit right now? In RightCapital and eMoney terms, this is the equivalent of running the plan again with today's starting balance and computing the new probability. The protocol's output is a single number — the today-Monte-Carlo — and a single delta — the difference between today and the last-saved review. A household that was at 91% in February and sits at 86% in May after a 17% peak-to-trough drawdown has lost five points of success probability and a great deal of working memory in the client's head; the advisor's job is to translate the five points into language the client can act on.

Stage 3 — Locate the Dynamic-Spending Trigger

If the plan's withdrawal policy already embeds a dynamic-spending rule — Guyton-Klinger guardrails, the Kitces dynamic-spending range, a Bengen 4.7% revisited, or a ratcheting strategy — the AI checks whether the current portfolio value has crossed any of the rule's triggers. (The next lesson in this chapter teaches the rule selection itself.) If the plan is on a static 4% rule and has no dynamic rule, the AI computes what a Guyton-Klinger capital-preservation rule (a 10% downward spending adjustment when current withdrawal rate exceeds initial by more than 20%) would recommend in this drawdown, and flags it as an option for the conversation. The point is not to change the policy in the middle of the drawdown — that is a behavioral failure of its own. The point is to know, before the client calls, whether the plan's own rule has anything to say.

Stage 4 — Generate the 60-Second Talking Script

The AI now produces a personalized, household-specific script of roughly 180-220 words the advisor can deliver in 55-65 seconds. The script has four components in fixed order: (1) where this household sits today on its own probability curve and the delta from the last review; (2) what the plan already absorbs — the bear-case sequence the Monte Carlo already ran 1,000 times and survived; (3) what the dynamic-spending rule (existing or proposed) says about today's level; and (4) what (if anything) needs to change. The script is written in plain English, names a specific dollar amount the client will recognize (their monthly withdrawal, their portfolio value, their target Social Security claim age), and ends with an action — either "no change, our policy already handles this" or "let's reduce the discretionary withdrawal by $X for the next two quarters and revisit." It does not use the words "long-term" or "stay the course" without a specific number behind them.

Stage 5 — Reg BI and Marketing Rule Classification

The script is a communication. Whether it is also a recommendation under Reg BI §240.15l-1 depends on whether it counts as a "call to action regarding a securities transaction or investment strategy." A script that says "no change, the plan handles it" is a confirmation of the existing policy, not a new recommendation. A script that says "let's reduce the withdrawal by $1,200/month and revisit in October" is, in the SEC's reading, a recommendation about an investment strategy (the withdrawal strategy is part of the broader allocation/decumulation strategy that 2025-2026 enforcement has treated as in-scope). The AI tags each script with its classification and, for the recommendation-class scripts, attaches the Reg BI documentation: the alternatives considered (no change, the proposed change, an intermediate change), the client-specific rationale tied to the plan's probability surface, and the placeholder for the registered person's signoff. The Marketing Rule under 206(4)-1 governs scripts that get repurposed — a script written for one household that the advisor wants to send as a templated email to ten more becomes "advertising" the moment it is repurposed, and the January 2026 SEC staff FAQs on third-party ratings, testimonials, and templated communications apply.

Stage 6 — Smarsh Archive and CRM Activity Log

Every script generated, every script delivered (live, by phone, by Zoom, by video voicemail, or by email), every advisor edit, every supervisory signoff, and every client response gets archived. The Smarsh or Global Relay pipeline picks up the email and the Zocks/Jump capture of the call and writes the package to the firm's retention vault under SEC Rule 204-2 (advisers) and FINRA Rule 4511 (BDs). The CRM activity log carries the structured record: which household, which script version, which advisor, which time, which classification (recommendation vs. confirmation), and which signoff. This is the artifact a 2026 SEC exam asks for when the question is "show me how you communicated with clients during the May drawdown." It is also the artifact that survives the May 2024 Reg S-P 17 CFR Part 248 amendments' breach-notification regime — the prompts, outputs, and edits live in the firm's controlled environment, not in a free public LLM where a paste of "Hendersons' withdrawal" would be an NPI-sharing event with an unvetted third party.

Three Real Households, Three Different Scripts

The protocol earns its keep when the same drawdown produces three meaningfully different scripts for three meaningfully different households. The contrived "model client" gallery is useless. The real test is whether the AI surfaces the right conversation for an actual book of business.

Household A — The 92%-Probability Retiree, Plan Built in 2022

The Hendersons: 67 and 65, $1.8M portfolio (60/40), drawing $7,400 a month, plan built in eMoney in 2022 at a 92% success probability with a 4.1% initial withdrawal rate and no dynamic-spending rule. May 2026 drawdown takes the portfolio to $1.49M, an 18% drop from the rebalance reference. The today-Monte-Carlo, re-run with current value, is 87%. The five-point drop is not the point — the point is that the 92% plan was already running a thousand sequences that included drawdowns worse than this one. The script: "You're at 87% today versus 92% in March. Of the thousand sequences your plan ran, hundreds included a market drop bigger than this one — and the plan kept paying you. Our withdrawal stays at $7,400. If we get another 10% from here, we'll talk about a temporary reduction. Nothing needs to change today." Reg BI classification: confirmation of existing policy, not a new recommendation. Archive: Smarsh email + Wealthbox activity log. Total elapsed advisor time once the protocol is configured: ~4 minutes.

Household B — The 78%-Probability Retiree With Guyton-Klinger Guardrails

The Martinezes: 71 and 69, $1.1M portfolio (50/50), drawing $5,600 a month, plan in RightCapital at a 78% success probability with Guyton-Klinger guardrails embedded — capital-preservation rule triggers a 10% withdrawal cut if the current withdrawal rate exceeds the initial rate by 20%. May drawdown takes the portfolio to $890,000. The current withdrawal rate has risen from 6.1% initial to 7.5% — a 23% relative increase that has crossed the Guyton-Klinger capital-preservation threshold. The plan's own rule is now active. The script: "Your plan has a built-in rule we agreed to in 2024: if a market drop pushes your withdrawal rate too high relative to where we started, we step the withdrawal down by 10% for a year and revisit. The drop in April-May has crossed that line. We'll move your monthly withdrawal from $5,600 to $5,040 starting next month and re-check in February. This is the plan doing its job, not me changing the plan." Reg BI classification: recommendation under §240.15l-1 — the script proposes a change to the withdrawal strategy. Documentation: alternatives considered (no change, the proposed 10% cut, a smaller 5% cut, a temporary discretionary-only cut), client-specific rationale tied to the Guyton-Klinger capital-preservation rule embedded in the plan, signoff queue. Archive: Smarsh + Wealthbox + Reg BI memo PDF attached to the household record.

Household C — The 38-Year-Old Accumulator Who Should Not Get a Drawdown Script At All

The Pattersons: 38 and 36, $480,000 in 401(k) and Roth IRAs, contributing $46,000 a year, retirement in 27 years. The plan's Monte Carlo at today's market level is 96%, basically unchanged because the household is a net buyer for the next two decades and the drawdown helps the future contributions. The AI's correct output for this household is not a drawdown script. The AI's correct output is a flag: "Patterson household — no coaching call recommended. Suggested action: reconfirm 401(k) contribution rate of 18% for the 2026 raise." The advisor's protocol distinguishes the households for whom a drawdown call is value-additive (retired, drawing, on the curve) from those for whom it is value-destructive (young, accumulating, contributing) and saves the cost of seven hundred unnecessary calls.

Prompt Architecture and the Tool Stack That Makes This Run

The protocol is not a vendor product as of May 2026 — no single tool ships the end-to-end pipeline — but every component exists. The senior advisor or the firm's AI lead assembles it.

Data Sources

The household's planning scenario lives in eMoney, RightCapital, or MoneyGuidePro. Today's portfolio value lives in Orion Eclipse, BlackRock Aladdin Wealth, 55ip's tax-aware overlay (for clients with that service), or the custodian's data feed. The CRM linking household to scenario to portfolio lives in Wealthbox, Redtail, or Salesforce FSC. The IPS — which the script must respect — lives in the firm's document vault (Wealth.com if estate-adjacent, the firm's own drive otherwise). The IPS allocation policy ranges and any client-specified prohibitions are inputs to the script generator.

LLM Layer

The LLM layer can be Microsoft Copilot for Business (with the firm's tenant-isolated data), Anthropic Claude through a vendor-hosted wrapper, OpenAI Enterprise, or Salesforce Einstein for FSC running on the household record. The system prompt is locked: senior-advisor persona, 180-220 word output, four-component structure, no use of "long-term" or "stay the course" without specifics, mandatory Reg BI classification tag, mandatory plain-English style aimed at a non-financial spouse. The system prompt is itself a record under FINRA Rule 4511 and SEC Rule 204-2; it is version-controlled, change-logged, and supervised under FINRA Rule 3110.

Supervisory Layer

The CCO or OSJ runs the principal-review queue on recommendation-class scripts under FINRA Rule 2210 and the SEC's analogous Marketing Rule pre-use review. The supervisory architecture is the agentic-AI-adjacent design pattern the FINRA 2026 Annual Regulatory Oversight Report calls out under Rule 3110 reasonable design: AI drafts, advisor edits, principal reviews on sample plus exception, all archived. The day the protocol runs against fifty households in an afternoon, the principal-review queue gets, at most, the four or five scripts the AI's own confidence scoring flagged as the highest-risk recommendations — not all fifty. Volume management is the point.

The Regulatory Spine of the Protocol

Three regulatory regimes intersect on this lesson; none is optional.

Reg BI §240.15l-1. The Care Obligation requires that when a recommendation is made to a retail customer, the registered person exercises reasonable diligence, care, and skill to understand the recommendation and to have a reasonable basis to believe it is in the customer's best interest. A withdrawal-strategy change is, in the SEC's 2025-2026 reading reflected in the rollover-enforcement AWCs, an investment-strategy recommendation in scope. The protocol documents the reasonable basis by anchoring every recommendation-class script to the household's own probability surface and the alternatives considered — the documented file the May 2026 enforcement pattern looks for.

SEC Marketing Rule 206(4)-1. A script written for one household is not advertising. A script template repurposed across a segment becomes advertising the moment it is sent to more than one client in identical form. The January 2026 SEC staff FAQs on hypothetical performance, third-party ratings, and templated communications govern the line. The protocol enforces single-household personalization by construction — the prompt requires household-specific dollar amounts, plan probabilities, and the IPS reference — but the firm's Marketing Rule compliance officer still reviews the system prompt and the output templates on the same cadence as any other client-facing content.

FINRA Rule 3110, Rule 2210, Rule 4511, and Reg Notice 24-09. Rule 3110 requires a supervisory system reasonably designed to achieve compliance; the FINRA 2026 Annual Regulatory Oversight Report makes the agentic-AI implication explicit. Rule 2210 governs the principal review of communications, including AI-drafted communications. Rule 4511 requires retention of communications and the supporting books and records — for this protocol, that means the prompts, the outputs, the advisor edits, the signoffs, and the client responses. Reg Notice 24-09 confirms FINRA's framing of GenAI under existing rules; no new rule is required to make this scope plain. Reg S-P 17 CFR Part 248 and the May 2024 amendments govern the NPI handling — the protocol runs on tenant-isolated infrastructure with a written incident response program and 30-day breach notification readiness, not on a free public LLM. NY DFS 23 NYCRR 500 layers on for NY-domiciled firms and adds the 72-hour cybersecurity event notification.

What This Protocol Replaces, and What It Cannot

The protocol replaces the firmwide drawdown email blast, the advisor's panic-driven generic "stay the course" voicemail, the post-drawdown CRM scramble to figure out who got what, and the audit-time problem of explaining to an examiner how communications were tailored to households. It collapses what used to be a week of inconsistent, manual, mostly-bad communications into a single afternoon of consistent, archived, household-specific coaching.

What the protocol cannot replace is the fiduciary judgment the registered human still owns. The AI does not know whether the Martinezes' daughter just lost her job and they're already nervous about the discretionary cut. The AI does not know the Hendersons' last review featured an off-mic comment from a spouse that they want to spend more, not less. The AI does not know the firm's CIO has just shifted house view on duration and the script's "no change" language is about to conflict with a model trade. The advisor's final read of the AI-drafted script — the thirty seconds of reading before delivery — is the judgment the protocol cannot automate and the regulators will continue to require. The Cardinal Rule (L1 Ch2.3) is the operational form; this lesson is the application at the highest-stakes moment in the advisor's year.

Key Takeaways

• The generic firmwide drawdown email is the industry's quiet failure mode. The AI-guided coaching protocol replaces it with per-household scripts anchored to the household's own Monte Carlo position.
• The protocol has six stages: pull the plan scenario (eMoney / RightCapital / MoneyGuidePro), fuse with today's portfolio value (Orion Eclipse / Aladdin / custodian), locate the dynamic-spending trigger if any, generate the 60-second script (180-220 words, four-component structure), classify under Reg BI §240.15l-1, and archive to Smarsh / Global Relay with the CRM activity log.
• Three different households produce three different scripts — the 92% retiree gets a "no change, plan already handles this" confirmation; the 78% retiree on Guyton-Klinger gets a recommendation-class script implementing the capital-preservation rule; the 38-year-old accumulator gets no script at all and a 401(k) contribution-rate flag instead.
• Recommendation-class scripts trigger Reg BI documentation: alternatives considered, client-specific rationale tied to the probability surface, registered-person signoff, and Smarsh archive — the May 2026 enforcement pattern looks for exactly this file.
• Marketing Rule 206(4)-1 governs the moment a script gets repurposed across more than one household; the January 2026 SEC staff FAQs apply. Single-household scripts are not advertising; templated segment-wide blasts are.
• The supervisory architecture under FINRA Rule 3110, Rule 2210, Rule 4511, and Reg Notice 24-09 handles fifty scripts in an afternoon via AI-flagged exception sampling rather than all-fifty principal review. Reg S-P (May 2024 amendments) requires tenant-isolated infrastructure, a written IRP, and 30-day breach readiness.
• The protocol cannot replace the advisor's final judgment. The thirty-second read before delivery — does this fit the household I actually know? — is the fiduciary act the LLM cannot perform.