Analyzing the Analysis: How Do AI Portfolio Recommendations Hold Up?

Why It Matters

AI tools like Claude can boost advisor productivity by quickly parsing complex statements, yet their numerical errors highlight the need for oversight, making the balance between automation and human review critical for reliable financial advice.

Key Takeaways

• Claude summarized $1.37 M portfolio with 0.03% expense ratio
• AI omitted $200 K of taxable assets in its totals
• AI recommended selling Treasury yielding 0.125% despite discount purchase
• Human review essential to catch arithmetic errors in LLM outputs
• Claude Code built near‑commercial planner in five to six days

Pulse Analysis

The rise of large language models (LLMs) in wealth management is reshaping how advisors process client data. In Roth’s test, Claude ingested four February 2026 statements, identified tax‑efficient holdings, and proposed a straightforward cash deployment strategy. By converting the $80,000 infusion into a split between VTI and VXUS, the model demonstrated its ability to translate high‑level allocation targets into actionable trades, a task that traditionally consumes hours of manual spreadsheet work.

Despite the speed and clarity of Claude’s narrative, the experiment exposed a critical weakness: arithmetic precision. The model excluded roughly $200,000 of taxable assets and mis‑read bond yields, errors that could materially affect client outcomes. This underscores a broader industry lesson—LLMs excel at pattern recognition and recommendation framing, but they are not calculators. Integrating code execution, such as Claude Code’s Python backend, can mitigate these gaps, allowing advisors to retain the AI’s analytical insight while ensuring numerical fidelity.

Looking ahead, the technology’s potential is evident. Financial planner Mike Piper built a near‑commercial planning application in five to six days using Claude Code, dramatically compressing development timelines that once required thousands of hours. As AI continues to mature, advisors who blend model‑driven insight with rigorous validation will gain a decisive competitive edge, delivering higher‑value, error‑free advice at scale.