Metal

The interesting thing about Metal is not that it does AI search over your files. Several products do that. It is that it takes a position on which number is true. Private capital runs on documents that disagree with each other. A CIM says revenue was $40.1M. An audited statement says $82,412K consolidated revenue with $39,226K EBITDA. A CFO's email adds $0.6M of one-time legal and $0.3M of owner comp to get to an adjusted figure. DealCloud says net retention is 118%. Any of those could be right depending on what you are underwriting. Metal's metric detection engine surfaces the discovered values, ranks them by source authority and recency, and retains the losing candidates alongside the winner — its own homepage example shows a reconciled GAAP EBITDA of $39.2M with three discovered values retained, an average confidence of 0.95, and a source path back to page 6 of the audited financials. That is the feature that matters. It is also the feature that is hardest to fake, because it requires a data model for private capital concepts rather than a vector index over PDFs. The second design decision worth noting is the canonical company record. Firms accumulate the same company across DealCloud, Salesforce, and Affinity under slightly different names, with relationship history split across them. Metal resolves those into one record — the homepage example resolves 'Halden Industrials' with 14 years of relationship history and a latest interaction dated Aug 12. If your CRM hygiene is poor, this is either the reason you buy or the reason the rollout takes a quarter. The third is the MCP server. Metal is explicit that it is a context layer rather than a connector. That distinction matters to a CTO: a connector pipes raw files into whatever assistant is asking, a context layer answers with reconciled, permission-aware firm knowledge. Practically, it means the same governed answer shows up whether an associate asks Claude, ChatGPT, Gemini, or Copilot, and you are not betting the firm on one model vendor. Metal's July 2026 post makes this argument directly. Where it fits: funds with real scale and real document sprawl — PE and private credit teams with more than a handful of portfolio companies, an existing SharePoint or Egnyte standard, and a CRM the deal team actually updates. The named references (Clearlake, Berkshire Partners, Valesco) are all institutional, and Berkshire's CTO frames the value as firm-wide adoption across investment and support staff, not a single-team tool. Where it does not fit: anyone outside institutional private capital. The data model, the metric vocabulary, the DDQ and VDR workflows are all shaped by that domain, and a general enterprise wanting search over its handbook will find it a poor match. Individual investors are out of scope entirely. And firms whose documents still live on personal drives will spend their first month on plumbing before they see an answer. Weaknesses worth naming. This is not a tool you switch on.