金融从业者终于有了可审计的 AI 建模方案——Primer 解决了 Excel 模型难以验证的痛点，做财务分析或审计的团队值得关注这个方向。
WallStreetPrep 进行了一项针对真实金融场景的 AI 基准测试，任务要求 AI 构建苹果公司历史和预测财务报表，包括引用来源、链接假设、添加附表并确保工作簿可审计。Primer 作为 AI 金融建模工具胜出，其优势在于将 Excel 视为最终输出格式而非工作语言，先构建强健的三表模型再转换为可审计的电子表格。Primer 将工作簿表示为结构化记录（如收入、成本、现金、债务、假设、公式、来源链接、注释和依赖检查），使 AI 能直接查询和验证财务逻辑，例如“显示所有影响现金流的公式”或“查找资产负债表平衡项”，而非逐格编辑脆弱的单元格引用。这反映了专业 AI 代理的评判标准正从聊天质量转向产出物能否通过审计。
WallStreetPrep did a very practical AI benchmarkin…
WallStreetPrep did a very practical AI benchmarking exercise for real-world finance.
It tested financial modeling agents on a real analyst assignment, not a toy prompt with a neat answer key.
The task was a serious analyst job: build Apple’s historical and forecast financial statements, cite sources, link assumptions, add schedules, and make the workbook auditable.
Primer, an AI financial modeling tool, came out ahead in this test, but the more useful point is why: its output looked less like a spreadsheet patched together cell by cell and more like a connected financial system that could be audited.
Primer treats Excel as the final output format, not the agent’s working language, so the AI can build a stronger 3-statement financial model first and then convert it into an auditable spreadsheet.
Primer represents the workbook as structured records such as revenue, cost of sales, cash, debt, assumptions, formulas, source links, comments, and dependency checks.
That means the AI can query and validate the finance logic directly, for example “show me every formula feeding cash flow” or “find balance sheet plugs,” instead of visually navigating Excel and editing fragile cell references one by one.
This is what I am seeing in many areas, that professional AI agents will be judged less by chat quality and more by whether their artifacts survive audit