Build and own an Investment Prioritization Specialist Model.
Given a set of candidate internal investments (projects, capital initiatives, product bets), which should we fund first?
Overview
Ranks competing internal investments on a consistent, comparable basis instead of ad hoc committee debate.
Who owns this decision
Chief Financial Officer, Chief Strategy Officer
Why owned AI may fit
What counts as a good investment ranking depends on a company's own capital constraints, strategic priorities, and historical outcome data, none of which a general frontier model has.
What this Specialist Model actually does
Decision: Given a set of candidate internal investments (projects, capital initiatives, product bets), which should we fund first?
This is a larger system with 3 specialist model candidates and 1 supporting or external component.
Who uses this Specialist Model
- Chief Financial Officer
- Chief Strategy Officer
- VP Corporate Development
- Head of FP&A
Where the specialist model sits
This is a larger system with 3 specialist model candidates and 1 supporting or external component.
Structured ranking/scoring model over candidate investments using historical outcome data
Specialist Model
Long-document reasoning LLM
Specialist Model
Long-document reasoning LLM to extract structured features from unstructured business cases
Specialist LLM
Specialist Model
Specialist LLM for rationale generation tying each ranking to strategy and precedent
Portfolio optimization layer applying budget, headcount, and risk constraints
Supporting component
Read the Portfolio Optimization guide →
How to evaluate and improve it
Evaluation
- Ranking agreement with investment committee decisions
- Realized ROI of top-ranked investments vs. lower-ranked ones
- Override rate by the investment committee
Feedback signals
- Investment committee override of the ranking
- Actual return/outcome of funded investments vs. predicted
- Outcome of investments that were deprioritized but later funded elsewhere
- Budget reallocation events that change relative priority
- Post-mortem findings on completed initiatives
Model candidates to evaluate
- qwen3.5-9b
- glm-5.2
These are candidates, not performance claims. Run them against your evaluation set.


