Optimizing investment decisions with a governed AI agent

Starting point

As the family office’s portfolio grew through real estate gains, inheritance, and consolidation, they wanted to professionalize how investment decisions were sourced, justified, and governed. A meaningful portion of liquid capital was allocated to public markets. Reviews were disciplined, but trading decisions still relied on fragmented tools, market narratives, generic AI searches, and individual judgment. That created three recurring problems.

First, asset selection was noisy. Themes produced long lists of topical assets, but few were truly investable once liquidity, listing structure, FX friction, and data coverage were applied. Second, valuation work was inconsistent. Different assets demanded different lenses, and it was hard to translate analysis into an action-ready conclusion with clear entry logic. Third, risk decisions were not systematized. Sizing, drawdown vulnerability, and concentration effects were often evaluated late, after conviction had already formed.

Rather than outsource decisions, the family office wanted a trusted system that respected confidentiality, enforced policy constraints, improved downside discipline through a risk engine, and remained simple enough to learn and execute consistently. The client engaged Strathen Group to build a governed AI agent that converts selected themes into investable shortlists, shortlists into valuations, and valuations into risk-approved actions.

‍

Approach

Strathen Group designed an agent as a decision workflow with three linked modules: Market Intelligence, Asset Valuation, and a Risk Engine. Each module used structured prompts, clear inputs, and an output contract so the system behaves consistently across themes, asset types, and market regimes.

I. Market Intelligence

The Market Intelligence module turns a theme into a vetted and ranked shortlist of candidates. The user starts by selecting a theme they would like to invest in. The agent then filters out anything that is not practical to own, such as illiquid names, structurally messy products, or assets with weak data coverage. Each remaining candidate is tagged by asset type (stock, ETF, REIT, etc.), primary exposure, and the best listing to use (CAD vs USD), with FX and spread friction called out.

Next, the agent builds a list of 10 to 20 candidates and groups them into simple sub-segments, so the theme is easy to navigate. It scores candidates using asset-type appropriate metrics, plus a technical regime check, then ranks them into score bands. Hard veto flags (provided by the client) prevent unsuitable candidates from reaching the final shortlist.

The output starts with a one-line Theme Verdict (Attractive, Neutral, or Avoid), then provides a ranked list and a Top 5 shortlist, each with a clear reason, biggest risk, and a copy-ready block to feed directly into the Asset Valuation step.

II. Asset Valuation

The Asset Valuation module answers one question: is this asset worth owning at today’s price for a 12-to-24-month horizon? The investment horizon can be customized by the client. The module starts by identifying the asset type (operating company, bank, REIT, ETF, etc.), then states data coverage and confidence so the reader knows how hard to lean on the output.

Valuation is then triangulated using two to three appropriate lenses. Depending on the asset, this can include a fair value model (DCF where meaningful, or comps where not), peer and history bands, and an implied expectations check that makes clear what must be true for the current price to be justified. The output is deliberately decision-focused: a clear verdict (buy or don’t buy, or hold or sell if already owned), a bear/base/bull fair value range, and a reward-to-risk view.

Finally, it turns valuation into an entry plan. Instead of pretending precision, it provides Good, Better, and Best entry zones, key catalysts and risks, and veto flags that override excitement. A regime check (trend, setup, key levels) is included to ensure the entry plan matches market conditions, not just the valuation thesis.

III. Risk Engine

The Risk Engine evaluates downside vulnerability and diversification efficiency before a change is approved. It takes a holdings list and weights, and optionally proposed trades and sizing options. It reports a clear Risk Verdict: approve, approve with resize, or reject.

At the portfolio level, it provides a best-effort risk snapshot, including concentration flags, diversification efficiency, and liquidity sensitivity. It then produces holding-level risk scores and identifies the top risk contributors. Stress tests are designed to be interpretable: a broad correction, a correlated shock, a sector leader shock, a volatility spike, and rate sensitivity where relevant. The output ends with practical sizing guidance and the top risk fixes, expressed in plain English.

The agent turns deep financial and market analysis into a governed workflow with explicit handoffs, score bands, and veto rules.

‍

Outcome

The family office moved from fragmented research to a repeatable, governed workflow. Market Intelligence produced investable shortlists with clear listing choices. Valuation translated candidates into decision-ready views with ranges and entry planning. The Risk Engine evaluated every proposed change for concentration and drawdown before it reached the portfolio.

Over time, the family office reported consistent double-digit returns, driven by structured market and financial analysis and disciplined risk controls, not ad hoc judgment. Recommendations also became easier to defend. Each output included a clear rationale, the constraints that shaped it, and what would change the view, so investment committee discussions shifted from rebuilding analysis to testing assumptions and tradeoffs.

AI earns trust in investing when constraints and oversight are enforced before recommendations are generated.

Strathen Group can design such governed AI agents for family offices and investment advisory teams, spanning theme sourcing, valuation ranges, portfolio construction, rebalancing, and multi-portfolio reporting, deployed inside your perimeter with human-in-the-loop approvals. Contact Strathen Group to discuss a diagnostic and a timeboxed pilot.