A general assistant is not a domain decision system.
Microsoft Copilot is excellent at generic productivity work — drafting, summarizing, retrieving across your tenant. Capital Refinery is a domain-specific decision integrity layer for private markets. They sit at completely different levels of the stack and answer completely different questions.
What Copilot does well, and where the line is.
What Copilot is built for
Generic AI assistance across the Microsoft 365 surface.
Drafting in Word, summarizing in Outlook, formula assistance in Excel, retrieval across your SharePoint tenant. For any team that lives in Microsoft 365 — which is most of them — it is a real productivity layer. The category is mature, the tenant integration is deep, and the cost of adding it is low.
What it is not built for
Holding the IC decision as a structured testable record.
Copilot is a horizontal generic-purpose tool. It has no domain model for private capital. It does not know what a covenant is. It does not know that a KPI on a slide is supposed to be tested against an entry assumption from the IC memo. It can summarize a CIM; it cannot tell you whether the IC decision the team made on the basis of that CIM is still defensible eighteen months later.
Where Capital Refinery sits next to it
Underneath the workflow, with a domain data model.
Most teams will keep Copilot. Capital Refinery is not a Copilot replacement — it is a different layer of the stack. The structured investment record, the deterministic-first extraction kernel, the decision validity scoring, the time-to-consequence ranking. None of these are things a generic assistant can produce, because none of them is a generic productivity task.
Where the lines fall.
| Capability | Microsoft Copilot | Capital Refinery |
|---|---|---|
| Generic drafting and summarization | Strong — core capability | No — not the goal |
| Microsoft 365 tenant integration | Strong | No — separate platform |
| Domain model for private capital | No — generic LLM | Yes — first-class objects: thesis, covenant, condition, decision |
| Deterministic extraction with provenance | No | Yes — every figure has a candidate ID |
| Structured IC decision record | No | Yes — bound at approval |
| Continuous re-test of investment thesis | No | Yes |
| Covenant + breach probability | No | Yes — for credit positions |
| Time-to-consequence ranking | No | Yes |
Your team will keep Copilot. The question is what sits underneath it.
Bring us a deal from your portfolio. We will show you the structured decision layer Copilot is not designed to produce.