Industry · Wealth Management
Two transformer backbones — one over investor event histories, one over advisor production timelines — pre-trained self-supervised on millions of accounts and tens of thousands of advisors. Downstream heads predict withdrawals, transfers, closures, and advisor departure months in advance — and agents open the retention play before the advisor calls.
01 · SharpAI
Predictive models
11 models · foundational + downstream
02 · AUTONOMY
Agents
20 agent patterns in this vertical
03 · FDE
Engineering services
3 anonymized engagements
01 · SharpAI · Wealth
Two transformer-based foundations: one over investor event sequences (transactions, holdings, advisor touches), one over advisor production timelines. Both pre-trained self-supervised across millions of accounts and tens of thousands of advisors. Downstream heads attach to the frozen embeddings and fine-tune to each firm's product catalogue and segment definitions.
Foundation · what the backbone tokenizes
Millions of investor accounts, multi-year event histories. Each dot is a raw signal the Investor Behavior Transformer tokenizes into the household's sequence — hover or tap to see what it is.
Downstream · what the head consumes
The attrition head reads the frozen IBT embedding plus a focused set of decision-time features. Hover any node — these are the inputs that drive a single retention score and the suggested play.
Foundation models · self-supervised pre-training, frozen backbone reused downstream
Backbone for investor intelligence. Tokenizes a household's investment event history — trades, contributions, withdrawals, portfolio drift, advisor interactions, life-stage transitions — into a sequence the encoder reads. The output embedding feeds every investor-facing head.
Transformer encoder · multi-head attention over event sequences · 256-dim shared embedding · learned tokenizer for portfolio + transaction tuples
Pre-training objectives
Corpus · Millions of investor accounts across wealth and brokerage portfolios · 18+ months of event history · multi-firm joint pre-training, federated where data residency requires it
Backbone for advisor intelligence. Tokenizes an advisor's production timeline — book composition, AUM evolution, manager interactions, tenure transitions — to surface the patterns that precede a resignation.
Transformer encoder · weekly + monthly aggregation tokens · 192-dim advisor embedding · attention with manager-context cross-attention
Pre-training objectives
Corpus · Tens of thousands of advisors across multi-affiliate US wealth firms · multi-year production histories · departure-labeled examples for contrastive pre-training
Downstream heads · fine-tuned on the frozen foundation embedding per institution and use case
Predicts significant withdrawals, transfers-out, and full closures over 30/90/180-day horizons.
Multi-horizon survival head over the IBT embedding
Predicts advisor departure and AUM-at-risk; surfaces a retention plan to the regional manager.
Fine-tuned head over the ALT embedding
Bidirectional matching — given a household, recommend investments; given an investment, surface matching households.
Two-tower retrieval · IBT embedding as query
Predicts SBL pay-down and turnover; routes retention plays before the line closes.
Fine-tuned survival head over IBT + product context
Encodes investor risk profile, behavioral tolerance, and regulatory constraints; flags or blocks mismatched recommendations pre-action.
Calibrated classifier · policy-gated output
Surfaces winning patterns across the advisor population; feeds the coaching engine.
Peer-cohort regression head over ALT
Predicts cash balances likely to be deployed; routes the cash-sweep or deposit recommendation.
Time-to-event head with cash-trajectory features
Classifies the household lifecycle stage (accumulation, transition, retirement); surfaces the right product cohort.
Classification head · life-stage taxonomy
Detects accounts likely belonging to the same household across the book; suggests household merges with confidence.
Embedding similarity + relational features
92%
Holdout performance on investor attrition prediction
$4.3B
Mid- and high-risk retention opportunity identified
$83B
AUM at risk identified via advisor attrition · top 10 NA bank
8,000+
Advisors covered at a US wealth firm with $350B AUM
02 · AUTONOMY · Wealth
Wealth is a relationship business — but the relationship can't scale without a system that surfaces the right action to the right advisor at the right time. These agents are that system.
20 agent patterns · ordered by deployment maturity
Predicts significant cash withdrawals, transfers out, and account closures with months of advance warning; opens the retention ticket and drafts the advisor outreach.
Trigger · Score threshold crossed
↳ 92% holdout · $4.3B retention identified
Predicts advisor departure and the AUM at risk; surfaces a retention plan to the regional manager.
Trigger · Departure score + AUM threshold
↳ $83B AUM at risk surfaced · top 10 NA bank
Given an investor cash balance, recommends specific investments; given an investment, surfaces the matching investor cohort.
Trigger · Cash threshold or new-product event
↳ Bidirectional matching
Predicts SBL turnover; routes retention plays to the borrower with personalized messaging through the advisor.
Trigger · Pay-down velocity shift
↳ $51M high-risk + $134M mid-risk identified · top US issuer
Surfaces winning patterns across the advisor population to managers; drafts coaching content for under-performers.
Trigger · Weekly performance window
↳ +300% engagement on advisor recommendations · US wealth firm 8,000+ advisors
Pre-checks every investment recommendation against the client's risk profile and regulatory framework — flags or blocks before submission.
Trigger · Pre-recommendation gate
↳ Pre-action compliance
Detects idle cash balances above threshold and recommends the right deposit or sweep product through the advisor.
Trigger · Cash balance + idle threshold
Monitors portfolios for drift outside the household's policy bands; surfaces a rebalancing recommendation to the advisor.
Trigger · Drift + policy threshold
Detects accounts likely belonging to the same household across the book; suggests household merges to the advisor for relationship view.
Trigger · Pattern match + threshold
Drives onboarding velocity for new accounts — funding milestones, advisor introduction, first-90-day engagement.
Trigger · Account opened
Detects estate-distribution and inheritance events; surfaces the next-generation conversation to the advisor.
Trigger · Inheritance event signal
Identifies clients with children at the firm — and clients with children elsewhere — and surfaces hand-off opportunities.
Trigger · Lifecycle + relationship signal
Generates weekly book-of-business briefings for advisors and regional managers — surfaces risks, opportunities, and coaching cues.
Trigger · Weekly cadence
Surfaces tax-lot harvesting and gain/loss matching opportunities at year-end and on rebalance events.
Trigger · Year-end + rebalance
Detects households entering retirement; surfaces income-planning and decumulation conversations to the advisor.
Trigger · Lifecycle stage transition
Schedules personalized outbound contact across email and advisor calls — frequency-capped per household.
Trigger · Score + campaign event
Reads inbound client documents (statements, tax forms, beneficiary forms), extracts structured fields, and routes to the advisor team.
Trigger · Document arrival
Monitors transaction patterns for AML signal and pre-checks suitability on each material recommendation; flags or escalates per policy.
Trigger · Pattern + threshold
↳ Continuous compliance surface
Inside the advisor's CRM — surfaces next-best-action across the book, drafts client notes, generates meeting prep.
Trigger · CRM session / pre-meeting
Coordinates attrition + suitability + outbound + advisor-briefing agents on a single high-risk household — one audit trail across the play.
Trigger · High-risk attrition score
↳ Coordinated retention play
92%
Holdout on investor attrition prediction
$4.3B
Retention opportunity identified
+300%
Engagement lift on advisor recommendations
$185M
SBL retention opportunity identified
03 · FDE · Wealth
Senior engineers embedded with the firm's wealth, advisory, compliance, and platform teams — building models against custodian data, integrating with advisor desktops, and operating under SEC / FINRA / regional regulator expectations.
Read-only access into custodian data (Pershing, Schwab, Fidelity, proprietary), advisor CRM (Salesforce / proprietary), and household aggregation systems — via API, signed URLs, or direct database.
Models and agents pre-check against the firm's suitability framework and the applicable supervisory regime — SEC, FINRA, regional. Explainability per recommendation is a default.
Insight lands where advisors already work — Salesforce, proprietary advisor desktop, household 360 views. No separate dashboard to log into.
For multi-affiliate or multi-jurisdiction wealth firms — foundation models train across affiliates without raw client data leaving each affiliate's perimeter.
Every score, prompt, tool call, and resulting action captured with lineage — same registry as model recommendations — for supervisory examinations and internal audit.
Behavioral AML monitoring layered on the foundation — surfaces pattern shifts the rule-based system misses, with case-level rationale logged.
Anonymized engagements · same engineering team end-to-end
US wealth firm · $350B AUM
Multi-affiliate broker-dealer · 8,000+ advisors
Investor attrition prediction across millions of accounts, advisor performance coaching, and next-best-investment matching. Engineering team embedded with the firm's wealth analytics and advisor platform teams.
92% holdout on attrition · $4.3B retention opportunity surfaced · +300% engagement lift on advisor recommendations.
Top 10 NA bank · wealth division
Bank-owned wealth platform
Advisor attrition and AUM-at-risk modeling, cross-sell from retail into wealth, and household aggregation across the firm's books.
$83B AUM at risk identified via advisor attrition signal.
Top US securities-based loan issuer
Wealth + brokerage lending
SBL turnover prediction and retention play across the borrower portfolio — integrated with the firm's advisor desktop.
$51M high-risk + $134M mid-risk SBL retention opportunity identified.
Team shape — typical FDE composition for this vertical
Embedded senior engineers, predictive models in production, and agents that execute on every score.
