CastleOak × DcisionAI

AI-Enabled Optimization for CastleOak Securities

A use case map identifying where DcisionAI's decision intelligence creates immediate, measurable value across CastleOak's trading, capital markets, and advisory businesses.

Prepared for CastleOak Meeting · May 2026

Firm Profile

CastleOak Securities — Who They Are

At a Glance

Boutique investment bank and broker-dealer, founded 2006 in New York City. 100% independently owned, 92% minority-owned. ~75 professionals across 5 regional offices.

Offices: NYC · ATL · CHI · SF · LA · PDX

Leadership: David R. Jones (CEO), Patrick de Catalogne (EVP, Fixed Income), Robert Bacon (EVP, Capital Markets)

$5T+

Public Offerings

Total assisted across equity and fixed income markets

$4.3B

Avg Daily Volume

Fixed income trading across all product types

670+

Institutional Accounts

Tier I–IV counterparties across the platform

$79B+

M&A Completed

Middle-market advisory transactions closed

Business Lines

Six Distinct Business Verticals — All Optimization-Ready

Fixed Income Sales & Trading

30+ professionals. Corporate, MBS, ABS, GSE, hybrid, CP. $4.3B average daily volume.

DirectPool

Proprietary electronic fixed income trading platform with Bloomberg BOLT integration. Institutional liquidity pool.

Investment Banking & Capital Markets

Underwriting, distribution, and structuring across fixed income products for institutional issuers.

Equity Sales & Trading

60+ global markets. Algorithmic strategies. ~$10B notional share repurchase execution per year.

Financial Advisory / M&A

Middle-market M&A under $1B. Fortune 500 divestitures. $79B+ in completed transactions.

Private Capital Advisory

Secondaries advisory, LP/GP liquidity solutions, fund management, and capital raising for alt asset managers.

Pilot #1 — Highest Probability of Sale

Use Case #1: New Issue Allocation Optimization

The Problem

When CastleOak co-manages or underwrites a bond offering, it must allocate bonds across 670+ institutional accounts — simultaneously balancing underwriter economics, relationship fairness, regulatory concentration limits, and syndicate desk preferences. Today, this is done in spreadsheets, introducing both errors and opportunity cost.

Decision Variables

Allocation amount per account per tranche — a classic multi-objective linear/mixed-integer program with hundreds of decision variables and hard constraints on diversification, fill rate, and revenue-weighted priority.

Optimization Objectives

Revenue-Weighted Priority

Higher-revenue accounts receive preferential allocation

Order Fill Rate

Minimize dissatisfaction from unfilled institutional orders

Diversification

No single account exceeds defined % of deal size

Regulatory Compliance

Syndicate and regulatory rules enforced as hard constraints

Pilot #2 — Clearest ROI

Use Case #2: Share Repurchase Execution Optimization

CastleOak executed ~$10B in notional share repurchases last year for 25+ corporate clients. Each program requires multi-period scheduling to minimize market impact while remaining inside SEC Rule 10b-18 safe harbor provisions — exactly the problem structure DcisionAI has already proven.

Regulatory Constraints (10b-18 Safe Harbor)

Volume Limit

≤25% of ADTV per day, excluding block trades

Timing Windows

Not in last 10 minutes for active NMS stocks

Price Condition

No bid above highest independent bid in the market

Single Broker Rule

One broker per day for open-market purchases

Financial Impact

$2M–$5M/year

Client savings from 2–5 bps execution improvement on $10B notional

Proven Architecture

Multi-period optimization already demonstrated: 7-year Roth conversion, 14 variables, solved optimally

Mandate Differentiator

"Mathematically optimal execution" — a compelling story for winning corporate mandates

Pilot #3 — Greenfield Opportunity

Use Case #3: LP Secondary Portfolio Selection

The Problem

CastleOak's Private Capital Advisory (PCA) practice, launched March 2025, advises LPs on secondary market sales of fund interests. Each mandate requires selecting the optimal subset of a client's portfolio to sell — maximizing proceeds while maintaining diversification, respecting NAV discount tolerance, buyer capacity limits, and GP consent requirements.

What DcisionAI Solves

Binary sell/hold decision per fund interest — a classic mixed-integer program. DcisionAI finds the mathematically optimal subset, with full constraint transparency for LP clients who want to see the reasoning behind every recommendation.

Why This Wins

Greenfield — No Incumbent

PCA launched March 2025. No legacy tools, no system to displace.

Binary Selection = Proven Model

Sell/hold per fund interest is the exact structure DcisionAI's PE models solve.

Differentiator for LPs

"Mathematically optimal, not judgment-based" — a transparency story that resonates with institutional LPs.

Every Mandate is Custom

DcisionAI makes each PCA engagement faster and more rigorous.

Use Case #4 — Honorable Mention

Use Case #4: Money Market Fund Portfolio Construction

CastleOak Shares (COSXX, CASXX) are branded money market products that must construct portfolios maximizing yield while maintaining full SEC Rule 2a-7 compliance. DcisionAI solves this as a constrained security selection problem — automatically enforcing all regulatory parameters as hard constraints while optimizing for yield.

SEC Rule 2a-7 Hard Constraints

WAM ≤ 60 Days

Weighted average maturity limit enforced at every rebalance

WAL ≤ 120 Days

Weighted average life constraint across the full portfolio

Daily Liquidity

Minimum daily liquid asset percentage maintained

Issuer Concentration

No single issuer exceeds defined % of fund NAV

Why It's Compelling

Branded Products

COSXX and CASXX — any yield improvement is directly attributable to CastleOak

Removes Manual Compliance

Eliminates manual 2a-7 checking; enables dynamic rebalancing as rates shift

State Street Platform

DcisionAI optimizes the portfolio that trades on State Street Fund Connect

Why DcisionAI

The Fit Is Structural, Not Generic

DcisionAI's architecture was built for exactly the problems CastleOak faces every day — constrained optimization at speed, with regulatory hard constraints that can never be violated.

Speed at Scale

Solver runs in <100ms for inventory-size problems. $4.3B/day of trading decisions demand real-time answers, not overnight batch runs.

Regulatory Hard Constraints

Domain cards with cited regulatory references — SEC 15c3-1, 10b-18, Rule 2a-7. Hard constraints are never relaxed, never violated.

No OR PhD Required

Boutique firms can't hire quant optimization teams. DcisionAI's natural language interface lets traders describe problems — the engine formulates and solves them.

Proven Multi-Period Engine

7-year Roth conversion with 14 variables solved optimally — the same architecture handles 30-day repurchase schedules with 10b-18 compliance windows.

"Our execution is mathematically optimal, not rule-of-thumb." — The story CastleOak tells institutional clients with DcisionAI powering the desk.

Recommended Pilot

Start Here: New Issue Allocation Pilot

Why This Use Case First

1

Weekly Frequency, Immediate Data

40+ new issues in the last 6 months — CastleOak can run a retroactive test against real allocations immediately.

2

Spreadsheet-Based Today

No allocation optimization tool in place. Zero incumbent to displace.

3

670 Accounts = Classic MILP

Revenue-weighted priority, fill rate, diversification, and regulatory constraints — a textbook LP/MILP problem.

4

Low Risk, High Signal

Wrong allocation doesn't lose money — it just leaves relationship value on the table. Safe to test.

Pilot Structure

Scope

Run DcisionAI against CastleOak's last 3 actual new issue allocations — same orders, same account tiers, same constraints.

Measurement

Fill rate improvement · account satisfaction delta · relationship value left on the table

Duration

2–3 weeks to model + results presentation