Keye

Proprietary models read deal files of any format and produce structured analysis along a chain of thought modeled on how a deal team works, while deterministic pipelines handle the arithmetic. Remove the models and a folder of documents and a spreadsheet are left.

The product analyzes and never acts, and an analyst can change assumptions, reload the export and have the system update its work. Confidence per analysis, flags on low certainty extractions and guidance on what to verify before a number reaches an investment committee are not described.

Live formulas rather than pasted values come in the spreadsheet export, so a reviewer can audit the calculation chain inside their own tool without the vendor. Neither accuracy measurement nor error analysis by document type is published, and model documentation is absent.

Testimonials name roles and fund sizes, such as a principal at a fund managing more than $45 billion, but never a firm or a person. Aggregates cover funds with $1.4 trillion under management, five days saved per deal and 40 percent more deals at one firm.

Training is the unsettled point: the homepage rules out training on customer data, while the privacy policy refers to training models with an opt out. Its cross client boundary is asserted in one document and qualified in another.

Uploaded files fall under a privacy policy updated February 3, 2025, which names service provider categories. The homepage promises zero data retention, yet the policy keeps information as long as legitimate business purposes require, and the two do not match.

SOC 2 Type 2 and end to end encryption are named on the homepage, though a SOC 2 examination is an attestation rather than a certification and the privacy policy speaks only of commercially reasonable safeguards. Audit period, auditor and trust portal are unpublished.

Registered advisers among the customers carry recordkeeping and marketing rules tied to the analysis and valuations they produce. No supervisory instrument is named, nothing places generated diligence inside supervised records, and Keye itself is not licensed, nor does it need to be.

Analysis described as 100 percent accurate and free of human error sits uneasily with quality of earnings work, which turns on judgment about unusual accounting and inconsistent formats. Accuracy measurement and an account of where the system degrades are both unpublished.

Every output carries its sources and supporting quotes, so a wrong figure can be traced to the document it came from. Keye publishes no remedy for one that reaches an investment committee and shapes a bid, and the party harmed is usually the fund's own investors rather than anyone holding a contract.

Models described as proprietary and trained on private equity documents suggest an owned analytical layer. Infrastructure providers appear only as categories, and whether any external model is called during processing is unstated.

Ingestion takes raw deal files of any type, which suits an unstructured data room, and output follows one path, the spreadsheet export. Integrations with data rooms, customer relationship or portfolio monitoring systems are not named, and no developer documentation is published.

Processing and storage on American servers are committed in the privacy policy. Cloud provider and private hosting options are unnamed.

A booked demo is the only route to pricing, with no rates, tiers or seat cost published. Deal teams count cost in associate hours, and whether charging follows seats, deals or documents is left unsaid.

Private equity and private credit investors are the single target, with financial diligence, CIM analysis and deal flow covered and a Capital AI blog written for private investment teams. Material is not split by fund type, so growth or credit strategies get the same description.