Alkymi

Machine learning and language models turn capital calls, loan notices and statements into structured data, and Patterns automates the workflow for each document type.

Output is structured data for investment teams rather than a decision, and validation is its own workflow step. How confidence in a single extracted value is shown, or what review happens before figures move downstream, isn't described.

Each fund's expected documents are tracked for completeness, which catches the report that never arrived, but no field level accuracy or error rate is published.

Northwestern Mutual, Principal Asset Management and HESTA are named customers, with case studies listed, but the resource pages attach no outcome figure to any of them.

Firms control where documents go by choosing the model host, yet whether their documents train the extraction models, or whether learning crosses clients, isn't stated.

The privacy policy, effective December 13, 2020, covers the services and casts Alkymi as a service provider for customers' end users. It names no providers or locations, and the documents it governs carry the full private markets positions of pension funds and insurers.

An annual SOC 2 Type 2 examination, multi factor authentication on systems holding customer data and staff background checks are described. There's no public trust portal.

Private credit output feeds valuation and credit monitoring at regulated lenders, and Alkymi cites no rule or reporting standard it's built to support.

A smaller fund's nonstandard reports may be read less reliably than a large manager's, and borrowers flagged as weakening in private credit monitoring never see the flag. Alkymi offers no analysis by document source.

When a wrong extracted value has already flowed into reporting, nothing describes how Alkymi corrects it or tells the client, and no accuracy commitment backs the output.

GPT models through Azure OpenAI, Amazon Bedrock and Groq are the named model options, alongside a customer's own model, though there's no subprocessor list.

SimCorp Dimension, FINBOURNE and Snowflake Secure Data Sharing are named integrations, and documents are pulled automatically from investor portals.

Customer data can be hosted in the EU where GDPR requires it, in data centers holding SOC 2 and ISO 27001 attestations, though no cloud provider or region is named.

The platform, Alts and Private Credit products carry no published price, and nothing says whether cost follows document volume or funds monitored.

Allocators, private equity firms and wealth managers are named buyers, with separate products for alternatives and private credit and document coverage listed type by type.