How a U.S. Investment Manager Turned Its Decision Process into a Firm Asset
Executive Summary
A U.S. investment manager wanted AI to help its advisors build client proposals, but its recommendation logic lived in expert memory and scattered documents. In 35 weeks, Andersen Consulting built an AI control plane and an investment brain in the firm’s tenant and put two use cases into production. Advisors now start each proposal from traceable evidence, every model call follows one governed path, and the firm can switch models without code changes. The program showed the value of Andersen Consulting’s applied AI approach and gave the firm a reusable foundation for each new use case.
Introduction: Context and Challenge
Across the Industry
Fee pressure and client expectations push investment managers to give faster, more personal advice without more staff. Many investment managers test AI one tool at a time, with no view of spend, no common policy, and outputs that compliance teams cannot trace.
The Client’s Challenge
The firm’s best recommendations depended on the judgment of a few senior professionals, with the Chief Investment Officer at the center. That judgment sat in historical proposals, custodial documents, CRM records, and correspondence, and no system recorded the rationale behind each proposal. Advisors rebuilt each proposal by hand, and that effort limited the number of clients per advisor. At the same time, teams across the firm called AI models through separate paths, with no common policy, cost attribution, or audit trace. In a regulated business, the firm needed AI under its own control, with a full audit trail and no unreviewed output to clients.
Solution: LLM-Powered Proposal and Client Intelligence Built on an AI Control Plane
Approach Overview
Andersen Consulting mobilized a joint team in the first week, with the firm’s product owner, project manager, and business analyst. The work ran in three phases, and each phase ended at a Steering Committee gate. The team started with one question: how does the firm make a recommendation? The Chief Investment Officer and senior advisors walked the team through real decisions, and those sessions defined the decision record that the system uses. The control plane went live first, so each use case plugged into a governed foundation from the start.
Advisors now build proposals from the firm’s own decision history, and leadership controls every model call from one place.
Data Integration and Foundation
The team brought the firm’s priority sources into its Azure tenant, with provenance on every record. The pipeline extracts facts from custodial statements, CRM records, and correspondence, and identity resolution links each fact to the correct client.
The investment brain is a decision record that holds past proposals, their win or loss outcomes, and the rationale behind each recommendation. Every element carries two dates, when a fact was true and when the firm learned it, so a review shows what the advisor knew then. Retrieval filters results by entitlement, so each advisor sees only authorized data.
AI and Insight Generation
Language models extract client facts, map each proposal to a factor taxonomy, and assemble the evidence behind a recommendation. Hybrid retrieval returns relevant past proposals, and every result links to the exact source passage with its provenance. Guardrails enforce a regulated perimeter that the firm’s compliance and legal teams defined, and no output reaches a client without a professional’s review. An evaluation harness scores each release against a gold set of historical proposals, split by date so that each test uses only earlier data.
For example, an advisor prepares a proposal for a new client who moved assets from another custodian. The system extracts the client’s facts from custodial statements and CRM notes and retrieves past proposals for similar clients, with their win or loss outcomes. The advisor edits the draft, records the rationale, and sends it to the CIO review desk for approval. That approved rationale goes back into the investment brain, so the next advisor starts from a better record.
User Experience and Workflow Integration
Advisors see the facts, the precedents, and the review steps for each proposal on one screen. Review and approval are structured fields, so the audit trail stays complete with no extra work. The workflow mirrors the review process that advisors already follow, so adoption needed little change in habit. The second use case, a quarterly health check of each account, detects changes over time and alerts the advisor when an event needs attention. The advisor reviews each alert before any client contact. Andersen Consulting set up this use case through the admin configuration, with no new code.
Rapid Implementation
Andersen Consulting delivered the program in 35 weeks, with its own staff and no subcontractors. The team combined a delivery lead, a solution architect, an infrastructure engineer, and three AI engineers for evaluation, retrieval, and guardrails.
The investment brain and the proposal use case went live in the second phase. The third phase added the health check, moved the firm’s applications onto the gateway, and proved a model switch.
Each use case passed user acceptance against targets that the firm approved in writing, then entered hypercare. Knowledge transfer started at the close of the first phase, and the firm’s named operators ran the platform through a shadow period before handover.
Results and Impact
The firm turned its decision history into a governed asset that advisors use every day.
- Faster, Better-Supported Proposals: Advisors start each proposal from extracted facts and relevant precedents, which cuts assembly time and frees capacity for client work.
- Full Traceability for Compliance: Every retrieved result links to a source passage, and every recommendation carries a professional’s approval before it leaves the firm.
- One Governed Path for AI: Every confirmed application calls models only through the gateway, and spend reports attribute cost to each application and model call.
- Model Independence: A model switch passed the full evaluation suite with no code changes, and every component stays in the firm’s tenant in exportable form.
- Faster Time to New Use Cases: An operator at the firm now sets up a use case through configuration, so each new use case costs less and starts sooner.
The firm’s decisions now build a record that improves each new proposal, under controls that its compliance team trusts.
A New Standard for Client Recommendations
The collaboration gave the firm a way to keep its institutional judgment and use it in every proposal. AI assembles the evidence, and the advisor decides. The CIO review desk now spends its time on the reasoning in each proposal.
“Our best judgment lived in a few people’s heads and in old proposals,” said the firm’s Chief Investment Officer. “Now every advisor starts from it, and every recommendation still carries a professional’s name.”
Looking Ahead
Andersen Consulting will continue to support and improve the platform under a second agreement. Next candidates include institutional knowledge preservation and advisor similarity scoring, and each one reuses the same control plane and investment brain. Andersen Consulting will also support AI governance and change management, so that adoption grows with the platform. Each step moves the firm toward an operating model in which every client recommendation builds on the firm’s full experience.
Deliver Value to Your Organization with AI in Weeks, not Years
Andersen Consulting is a global consulting practice providing a comprehensive suite of services spanning corporate strategy, business, technology, AI transformation, and human capital solutions. Andersen Consulting integrates with the multidimensional service model of Andersen’s global platform, delivering world-class consulting, tax, legal, valuation, global mobility, and advisory expertise.