Product Memo

Decade Advisory Tool

A memo on the product, the user, and the architecture.
Author: Lucas Previato · Date: June 2026 · Status: Confidential
In brief

An AI-powered advisory tool for the underserved middle.

The product is human-led advisory with AI as the infrastructure that makes the advisor's work institutional-grade and scalable, built for the US$1M–10M client who has an advisor, pays for advice, and still receives product distribution instead. Seven parts follow; each links below.

Part I

The Gap.

The Brazilian advisory market serves most clients poorly, in three different ways, depending on where the client sits.

Retail

No advisory at all

The individual is on their own: whatever the branch offers, the algorithm surfaces, or they learn themselves.

US$1M–10M · the entry point

Product, dressed as advisory

Has an advisor. Pays for advice. Does not receive it. A book of ~250 clients, under an hour each per month.

Ultra-high-net-worth

The family office

Works, but is expensive and infrastructurally complex, and demands dedicated staff most would rather not own.

Why the middle is broken

At ~250 clients per advisor, even at full capacity there is less than one hour per client per month: no time to model credit risk, stress-test concentration, or decide whether an instrument belongs in a portfolio. The result is a distribution machine with advisory language layered on top.

The Banco Master case: the most concrete recent illustration. Across all platforms, R$52 billion in Banco Master-linked products were placed with clients; XP alone accounted for R$30 billion. Not a story about bad actors. A system with no structural incentive to ask hard questions: the product had a high yield, it fit the month's sales narrative, and the client's full financial picture was neither visible nor required to complete the transaction.

How AI changes the economics

AI makes institutional-grade customization economically viable at a segment that could not support it before. The work that used to need several junior analysts at a family office (reconciling statements, building a balance sheet, mapping concentration, modeling liquidity, drafting theses) can now be done by a model, reviewed by a senior advisor, and delivered at a quality that previously required a US$50M minimum. The advisor is no longer the constraint on customization. The constraint moves to what genuinely needs human judgment: building trust and holding the relationship through volatility.

The Brazilian high-income reality

A client holding ~90% in CDI-tracking instruments is not low-risk; it is a concentrated emerging-market credit bet, in a single issuer and a single currency, where high nominal yields say little about preserved purchasing power. Brazil is roughly 1% of global market capitalization (~340 of 55,000+ listed companies), so a 90%-Brazilian portfolio is a concentrated position in 1% of the investable universe. It only looks like safety.

Entry point and trajectory

The entry point is where the pain is sharpest. The same architecture then extends in both directions.

Upward · > US$50M

Thicken the human layer

Same tool, more service density: succession, alternatives, life-encompassing advisory.

US$1M–10M · entry

The validation segment

Where the model is proven: advisor-led, the improvement over the status quo most dramatic.

Downward · < US$1M

Thin the human layer

AI-first by default, a personal financial operating system.

Part II

Design Principles.

Four principles govern every decision in the tool.

1

Two stakeholders only: the advisor and the client.

The discipline is to refuse stakeholder creep. Compliance is a byproduct of good practice, not a module; the investment team operates behind the universe, not on top of the advisor. What remains is two interfaces: the Engine (dense, for the advisor) and the View (calm, for the client).

2

The client must feel heard before they are advised.

The tool earns the right to recommend. It comes after the client has told their story, seen it restated in plain language, been asked questions that prove the advisor listened, and read a Plan Document that opens with their words. The advisor never says "the model says."

3

Fluidity: friction must earn its place.

Some friction is legitimate (regulation, confirmations before irreversible actions). The principle is not to eliminate friction but to ensure it earns its place. The system suggests; the advisor decides; the client experiences care, not bureaucracy.

4

The investment team as backbone.

The infrastructure that makes recommendations defensible, via four functions: product curation (the approved universe), philosophy backbone, active opportunity surfacing, and AI training for portfolio construction.

Part III

The Client Journey.

Seven stages. The AI is embedded throughout (from Stage 0), visible to the client only when its presence makes the advisor's work better.

0
Pre-call preparation
The AI builds a plain-language brief from public data. Scaffolding, never a conclusion; never shared with the client.
AI
1
The first conversation
A human conversation. AI transcribes in the background and silently extracts the nine inputs, captured through natural flow, never a questionnaire.
Human-led
2
The Mirror Summary
Within 24 hours: one page of plain prose restating the client's situation, proof the advisor was listening. Drafted by AI, edited and signed by the advisor.
AI + Advisor
3
Documents: collection & validation
Validation against statements, discovery of the undisclosed, and construction of three artifacts: balance sheet, historical P&L, forward projection.
AI + Advisor
4
Profile synthesis & construction
Four parallel analyses → recommendation on the approved universe → advisor review & override → a staged transition plan.
AI + Advisor
5
The Plan Document
The structured deliverable, opening with a one-page Hero that tells the client's story back to them before any allocation is shown.
AI + Advisor
6
The ongoing relationship
Continuous AI monitoring → structured advisor touchpoints (2–8/year by complexity) → fully human escalation at high-stakes moments. The loop closes.
AI + Advisor

The first conversation: nine inputs

The advisor leaves with these, captured through natural conversation: seven core, plus tax, plus the mental benchmark.

1
Wealth source & trajectory

Where it came from, where it is going

2
Time horizon

When the money must start working

3
Income stability

How predictable the cashflow is

4
Concentration

Any position large enough to dominate the balance sheet

5
Loss tolerance

The number at which they would act

6
Life events

Exit, inheritance, purchase ahead

7
Near-term liquidity

Cash the portfolio must supply in 6–24 months

+
Tax exposure

Regimes, structures, offshore in place

+
Mental benchmark

What "doing well" is measured against

Stage 4: how a recommendation is built

Step 1

Four parallel analyses

Risk, liquidity scenarios, concentration, and cost & tax, run before any recommendation.

Step 2

Built on the approved universe

Position by position, each with a thesis. Anything outside it escalates to the investment team.

Step 3

Advisor review & override

Accept, modify, or reject: every override logged with reasoning, the AI's original kept.

Step 4

Transition planning

A phased roadmap around illiquidity and psychological barriers, paired with the Behavioral Covenant.

Part IV

Tool Architecture.

One Source of Truth, two interfaces, and an investment-team backbone. Everything else is derived.

The Source of Truth is the centralized, version-controlled client profile: the structured inputs, the financial artifacts, every transcript and document, the current and recommended portfolio, the Behavioral Covenant, and a full audit trail. It updates at every meaningful touchpoint.

The Engine: the advisor's interface (six surfaces)

Workbench

The default surface: profile, conversations, artifacts, and the discrepancy queue.

Recommendation Builder

Where the portfolio is constructed, reviewed, and overridden, with the four analyses behind it.

Pre-call Brief

What changed since last touch, prioritized questions, read in the last fifteen minutes before a call.

IT Feed

The approved universe, active views, escalation queue, and philosophy, live, no refresh needed.

Client Thread

A mirror of what the client is doing in their View: scenarios run, sections re-read, behavioral flags.

Audit Trail

Every change, recommendation, override, and version, timestamped, with authorship and reasoning.

The View: the client's interface (four surfaces)

Plan Document

The full deliverable, and the default surface, where most clients spend their time.

Live Dashboard

Current state at a glance, no KPI without context. Read-only.

Scenario Builder

The most consequential surface: "what if" made answerable, live. Curated by the advisor; never executes.

Communication Layer

Deliberately minimal: the channel is the client's choice (often WhatsApp); the record is the tool's responsibility.

Three roles, defined handoffs

AI handles

Pre-call briefs, transcript extraction, profile drafting, document validation, balance-sheet & P&L construction, the four analyses, recommendation drafting, and continuous monitoring.

The advisor handles

The first conversation, judgment on what the inputs mean for this client, the override, delivery of the Plan Document, the high-stakes Crossroads, and the relationship.

The investment team

The approved universe, the firm's philosophy, active opportunity views, and the training signal that improves the AI.

Part V

The Recommendation Logic.

Built from a small set of principles, applied consistently, visible at every step to the advisor.

Risk: what the client does, not what they say

1
Behavioral history
What they actually did in 2008, March 2020, 2022
Heaviest
2
Capacity for loss
The drawdown the plan can absorb, from the balance sheet
High
3
Mental benchmark
What "doing well" is measured against, often CDI
Medium
4
Stated tolerance
What they say: retained, but not driving when it diverges
Lightest
The benchmark, reframed. Not CDI. IPCA + 4–5% per year over a rolling 10-year window, the real-return profile of NTN-B. CDI is the cost of short-term liquidity in an emerging-market sovereign, not a long-term target. We do not change the portfolio to chase CDI; we change the comparison.

The portfolio, in four layers

Liquidity reserve
Layer 1

6–12 months of obligations, same-day liquidity. Sized up for variable income, high illiquid concentration, near-term events.

Stability
Layer 2

Extends protection from months to years: NTN-B (intermediate duration) and high-grade credit.

Growth
Layer 3

The residual after layers 1–2. Globally diversified by default; Brazilian equity a deliberate active position.

Opportunistic
Layer 4

Bounded, only on an active investment-team view, sized so being wrong never damages the plan. Most clients have none.

The waterfall: where the next R$50,000 goes

Gross income / inflow
Non-discretionary recurring expenses
Tax obligations
Pre-committed cash flows
=Investable surplus
Funded in order
1Reserve: fund first, to target
2Stability: fund second
3Growth: once protected
4Opportunistic: only when warranted

Time-horizon allocation

HorizonGrowth starting pointConstruction direction
< 3 years100% fixed-income biasedAdd equity only with strong capacity + tolerance
3–7 years30–50% equity-biasedCalibrate equity weighting against constraints
7–10 years50–70% equity-biasedEquity-biased default; reduce on binding constraint
10+ years70%+ equity-biasedBinding constraint reduces from this base

Concentration: multidimensional

A single Brazilian bank stock is concentrated three ways at once. Each dimension is measured independently and calibrated to its own nature.

Security

The tightest line: single-company risk does not diversify away.

Sector

More room: names within a sector retain some internal diversification.

Illiquid

Judged against the liquidity reserve, not punished outright.

Geographic

The most permissive line, and the one almost every Brazilian client already sits well past.

When a line is crossed, the response escalates, graduated, not binary:

Awareness flag

Documented and raised at the next touchpoint. Baseline unchanged.

Hard flag

Allocation modified to reflect the exposure. Advisor action logged.

Pre-condition

The portfolio is built around the position, not despite it.

The exact thresholds live in the tool's configuration, calibrated per profile, not in the client conversation. A calibration detail, kept off the table so the conversation stays on the risk.

Diagnostic lenses: where to probe, by background

Pattern-recognition priors: certain wealth origins reliably carry certain hidden risks. The lens guides attention, then recedes once the real numbers are in.

Equity-compensated executive

Employer-stock concentration, income-portfolio correlation, vesting and lock-ups, behavioral attachment.

Business owner before exit

The business as dominant illiquid asset, unquantified guarantees, liquidity bridge, sale timeline.

Post-exit founder

Idle cash without a plan, reinvestment-cadence risk, unsettled tax events, builder-to-steward shift.

Inheritor

Decision confidence, emotionally-weighted legacy positions, family governance, first-volatility overreaction.

Real-estate-heavy

Illiquid concentration, embedded leverage, rental-income instability, compounded geographic exposure.

Scattered accounts

No consolidated picture, unseen fees and tax drag, distributed-not-chosen positions, no architecture.

Cost- and tax-aware positioning

Not tax advice, but cost and tax are built into construction from the first recommendation, not added as an afterthought.

  • Embedded fees. Surfaces what the client actually pays (distribution fees, management fees, hidden spreads) and judges every position net of them.
  • Asset location. Models the tax drag of the same exposure held different ways (BDR vs direct US equity).
  • Tax-loss harvesting. Flags opportunities, scheduled around the Brazilian tax calendar.
  • Structural cost analysis. Models holding-company (PJ) versus direct, escalated to the advisor, validated with the client's tax counsel.
  • BACEN-CBE compliance. Tracks the reporting calendar for offshore assets above the threshold.
Part VI

Client Deliverables.

What the client experiences is a sequence of deliverables, each calibrated to a stage of the relationship.

Within 24 hours

The Mirror Summary

One page of plain prose, in the client's own language, signed by the advisor. They open it expecting a generic email and find proof that someone was paying attention, and never see the AI's hand.

Stage 3

The contextual document request

Framed against the client's own words, never a checklist: "Because you mentioned equity compensation…" Honest about the tedium; the 360° view is the goal.

The Plan Document: eight sections

The central deliverable, regenerated after every material update and version-controlled. The structure is the same for every client; the content is entirely specific to them.

The Hero

One page: the client's story, the headline numbers, and a preview of every section.

The Snapshot

Where they stand today and the trajectory that produced it, concentrations made explicit.

The Audit

Embedded fees, dead positions, tax inefficiencies, and concentration they did not know they had.

The Blueprint

The recommendation, position by position with rationale, and the transition path.

First Moves

The immediate actions, sequenced: the bridge from today to the Blueprint.

The Flow

The waterfall, made explicit, where the next R$50,000 of savings goes.

The Crossroads

Analytical frameworks for the major decisions the client is navigating.

The Horizon

Where the plan leads under a base case and two scenarios.

The Crossroads: decisions, framed before they arrive

The closing principle: the client decides; the plan adapts.

Business exit

Full sale, partial sale, secondary, IPO: each with its own portfolio, tax, and deployment timeline.

Education funding

Funding paths modeled against realistic cost scenarios, well in advance of the need.

Inheritance & structure

Structural alternatives, their tax implications, and what each signals to the next generation.

Liquidity events

The post-event balance sheet and deployment path, built before the event, not under time pressure.

Real estate decisions

Financing alternatives (cash, mortgage, offshore credit line, consórcio) in expected-value terms.

The Behavioral Covenant

Most portfolio failure in this segment is behavioral, not analytical. A plain-language agreement, signed while it is calm, for how the portfolio is held through volatility.

The covenant contains
What drawdowns may feel like

A calibrated description of what the portfolio could lose in historical-event scenarios, in BRL and USD, not just percentages. The worst case is anticipated, not hidden.

What the client has agreed not to do

No position sold in a drawdown without an advisor call first; no major change without 48 hours of waiting; no media-driven decisions.

What the advisor commits to do

Outreach at defined drawdown thresholds and concentrated-position events. The cadence increases under stress; it is not optional.

The benchmark reframe

IPCA + 4–5% over a rolling 10-year window, the floor the plan clears, not the ceiling. A year of CDI outperformance is not a signal to act.

Pre-agreed responses

What we do if the portfolio drops 20% in a year; if a concentrated position falls 40%; if a recession is declared; if the real strengthens mid-diversification.

Part VII

Why This Works.

The case rests on three claims.

1 · Advisor capacity scales without compromise

To serve clients with any real quality, the strongest high-income (alta renda) advisor today caps the book near 150, and even there the depth is limited. Push toward the ~250 the segment treats as normal and the advisor stops advising: reconciliation, document gathering, spreadsheet construction, and position monitoring consume the hours that judgment and relationship require. The Decade model places that entire workload on the AI, so the advisor carries 150 or more and gives each client materially more depth, breaking the trade-off between book size and quality.

Book size
150+
clients, higher quality, not fewer
Conversations
~600
150 clients × ~4 structured/year
Scheduled time
~900 h
of an advisor's ~1,700 productive hours
Headroom
~800 h
prep, escalations, the high-stakes moments

2 · Compliance as a natural byproduct

Most platforms bolt on compliance modules that slow the advisor down. Here, compliance falls out of the workflow: every recommendation has a documented thesis, every override is logged with reasoning, every conversation produces a transcript, and the Source of Truth is version-controlled by architecture. When a regulator asks why a recommendation was made and who approved it, the answer already exists in the form it was created. There is no separate compliance workflow because the requirement is already met by the work itself.

3 · The AI training flywheel

More clients → more profiles → more recommendations reviewed by the investment team → more advisor-labeled overrides → better-calibrated AI → stronger recommendations → better outcomes → easier acquisition → more clients. A new entrant cannot replicate this with better prompts or better engineers; the advantage compounds with operating history, and the gap widens every quarter.