Richard the Marbella Family Office Manager

## Background

Richard is the Director / COO of Marbella Interests, the family office of principal Bryan. Small team — Richard + an analyst, a controller, a part-time bookkeeper, plus outside CPA and tax counsel. Portfolio is the usual single-family-office mix (public securities, PE/VC commitments, direct private investments, real estate, insurance / trust structures).

Richard's *long-term* job is what every FO COO does: tie out the GL, ship Bryan a monthly pack, manage compliance, run a clean close. The *immediate* job — and the one Marbella is using amble + Claude Cowork to attack first — is **AI-driven operational transformation of the family office itself**:

> What does this office actually do every week / month / quarter? Which of those things can AI take off the team's plate? Which ones first, and what's it worth?

That's a scoping problem, not a portfolio problem. Marbella is using amble as the **AI-transformation operating system**: a structured whiteboard that lives next to Claude Cowork, where every workstream, SOP, recurring task, and observation gets captured, scored quantitatively, prioritized, and turned into a delegation. Phase 2 — the actual portfolio data layer (holdings, commitments, monthly pack) — comes once Phase 1 is delivering compounding value and the team has bandwidth to wire it up.

Richard came up through PE ops, has lived in QuickBooks / Sage Intacct / Addepar / Juniper Square, and has zero patience for software that promises "AI" without showing him what it would actually do, in numbers, before he commits a week of the team's time to it.

## Mindset & Behavior (Phase 1)

- I'm scoping AI's role in this office. I am not buying a portfolio system today.
- Every meeting, every pain-point conversation, every "we always have to do X" comment is a capture opportunity. Friction between hearing it and recording it is friction I won't tolerate.
- I work in Claude Cowork. amble must meet me there — capture from Cowork should land in amble without me copy-pasting.
- I want every captured opportunity to have four numbers: AI-readiness, complexity, FTE-impact, ROI. Don't make me guess the priority. Compute it.
- I will ship a monthly "transformation report" to Bryan during Phase 1: how many workstreams captured, how many automated, hours saved, $ saved.
- I do not want a chat-with-AI sidebar shouting at me. I want AI to *do* — score the opportunity, draft the SOP, propose the priority order. Get out of the way.
- Once Phase 1 is humming, I'll come back to amble for the portfolio layer. Don't make me build it now.

## Surface map (where Richard spends time — Phase 1)

| Surface                                              | When opened                              | Job                                                                                                                             |
| ---------------------------------------------------- | ---------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------- |
| `/t/marbella/dashboard` (transformation snapshot)    | Daily / weekly                           | Live view of: opportunities captured this week, scored, prioritized, in-flight SOPs, hours saved trend, $ value delivered.       |
| Capture inbox                                        | Multiple times a day                     | Anything Cowork pushes (voice memo, meeting summary, paste) lands here as a draft Opportunity. Richard triages: keep / merge / drop. |
| Opportunity record                                   | After capture, when scoring              | Title, description, frequency, current owner. Four scores: AI-readiness, complexity, FTE-impact, ROI. Suggested next step.       |
| Priority matrix view                                 | Weekly planning                          | 2×2 (or quadrant) of value × complexity. Top-right quadrant = do first. Drag to reprioritize.                                    |
| Process / SOP record                                 | When scoping an opportunity              | Step-by-step current process, hand-off points, system touch points, automation candidates per step.                              |
| Delegation tracker                                   | Continuous                               | SOPs in flight: built / piloting / shipped. Hours saved per SOP. ROI realized.                                                   |
| Monthly transformation report (for Bryan)            | First week of every month                | One-screen Phase 1 update: captured / scored / prioritized / shipped / hours saved / $ saved / next quarter's plan.              |
| Claude Cowork integration settings                   | Setup + occasional review                | Where Cowork captures land. Which Cowork projects map to which Marbella workstream area. Auth state.                             |
| People / team                                        | When a workstream surfaces a name        | Who owns this process today, who would own it post-automation, who needs training.                                               |
| Phase 2 placeholder (Holdings / Commitments / etc.)  | Eventually                               | Out of scope for Phase 1. Surface labeled as such — don't pretend it's coming next week.                                         |

## What Richard does NOT want (Phase 1)

- A pre-built family-office dashboard with empty NAV / holdings / commitments tiles. That's Phase 2 and showing it now is noise.
- A chat-with-AI sidebar that interrupts capture or scoring. AI works *inside* the artifact (auto-score, draft SOP, suggest priority).
- Manual forms for every capture. Voice / paste / drop should work.
- "Entity", "criteria-set", "response score", "session_id" in any UI Richard or his team see.
- A "tasks" widget that mixes operational work-items with the captured opportunities. Operational ≠ scoping.
- A Phase 1 dashboard re-arranging itself because the AI thought it could improve it. The transformation pulse is templated for trust.
- Any feature that requires Richard to leave amble to capture or score. If it needs a separate tool, it's not built yet.

## What Richard wants from Claude Cowork integration

Cowork is where Richard does the actual thinking — long meetings, deep document review, multi-step research. amble is where the *artifacts* live structurally. The integration must:

- **Inbound from Cowork:** When Richard finishes a Cowork session that surfaced an opportunity, Cowork should be able to push it to amble as a draft Opportunity record (with the original Cowork context attached / linked).
- **Outbound to Cowork:** When Richard wants to deep-dive on an Opportunity (scope the SOP, simulate the automation), he should be able to one-click "open in Cowork with this context."
- **Source-of-truth split:** Cowork = ephemeral working memory + drafts; amble = structured persistent record + scoring + prioritization.
- **No double-entry:** anything captured in Cowork that reaches amble should not need to be re-typed.

## The four scores (Phase 1 quantification)

Every captured Opportunity needs these populated, ideally AI-suggested with one-click human override:

| Score             | Scale | What it captures                                                                                              | Tool / source                                  |
| ----------------- | ----- | ------------------------------------------------------------------------------------------------------------- | ---------------------------------------------- |
| **AI-readiness**  | 1–5   | How tractable is this for current AI (clear inputs/outputs, has examples, low ambiguity)                      | `assessAiReadiness` MCP tool              |
| **Complexity**    | 1–5   | Build effort to automate (data wiring, integrations, edge cases)                                              | `estimateComplexity` MCP tool                 |
| **FTE-impact**    | hrs/wk| Hours/week saved across the team if automated                                                                 | `calculateFteImpact` / `calculateTimeSavings` |
| **ROI**           | $/yr  | Annualized dollar value (FTE-impact × loaded cost + risk-reduction value)                                     | `calculateRoi`                                |

Priority = ROI ÷ Complexity (with AI-readiness as a gate — anything under 2 gets parked). Surfaced via `buildPriorityMatrix` and `quickScore`.

## Phase 1 → Phase 2 transition

Phase 1 is "use amble + Cowork to figure out where AI helps." It runs on Process / Opportunity / SOP / Task entity types and the AI-scoring tools that already exist in the Marbella tenant.

Phase 2 is "use amble for the actual family-office data layer" — and only kicks off when:

- Phase 1 has shipped at least 5 automations
- Hours/week saved is materially > Phase 1 setup cost
- Bryan and the team trust amble enough to put real GL / holdings / commitment data in it
- Or: Bryan asks for the monthly pack to be assembled in amble

When Phase 2 begins, the surface map in the original v1 of this persona file (Holdings / Commitments / Compliance / Cash / Bryan's Pack / Close Calendar) becomes the active scope. Until then, those surfaces should be either absent or clearly labeled "Phase 2 — coming when Phase 1 has paid for itself."

## Demo-day signals (Phase 1)

When Richard walks Bryan or the team through the Phase 1 use of amble, these are the moments that earn or lose trust:

- **Capture-to-record is under 60 sec live.** Voice memo or pasted Cowork summary becomes a draft Opportunity record on screen.
- **Every Opportunity has the four scores populated** — none are blank, none say "Score: —". (The platform "Score" column from generic entity views must not surface here; this is a different scoring concept and needs its own block.)
- **Priority matrix view is glance-readable.** Top-right quadrant is visually obvious.
- **Drill into one Opportunity → see the auto-generated SOP draft, the four scores, the suggested next step.**
- **Monthly transformation report renders with real numbers** — not "Not Tracked" placeholders.
- **Cowork integration visibly works** — at least one captured-from-Cowork record on the demo dashboard.
- **No portfolio jargon** (NAV, capital call, K-1) anywhere in Phase 1 UI. Save it.
- **No platform jargon** (entity, criteria-set, scored items) visible to Richard or the team.

## Related personas (later, after Phase 1)

- **Bryan the Principal** — Phase 1 consumer of the monthly transformation report; Phase 2 consumer of the portfolio pack.
- **Marbella Analyst / Controller / Bookkeeper** — derivative roles; in Phase 1 they capture opportunities; in Phase 2 they own GL / holdings / AP-AR.
- **Jordan the PE Analyst** (`pe-analyst.mdx`) — adjacent fund-side analyst; not the Marbella use case.

## Cross-references

- AI-scoring MCP tools (Marbella tenant): `assessAiReadiness`, `estimateComplexity`, `calculateFteImpact`, `buildPriorityMatrix`, `scopeProcessAutomation`, `quickScore`, `calculateRoi`, `calculateTimeSavings`, `scoreDelegationReadinessSweep` (old snake/kebab-case names remain as deprecated aliases until 2026-09-30)
- Claude Cowork integration spec: TBD (this audit will produce the gap)
- Phase 2 family-office data model gaps: `documents/work/2026-05-03-personas-and-gap-audit/audit-report.md` §3 (deferred until Phase 2 begins)