Skip to content
BM2Capital

Platform

Research and technology for private markets

Research and technology for private markets: company data, documents, meeting intelligence, and quantitative methods in one research system.

BM² combines company data, documents, meeting intelligence, and quantitative methods in one research system. Filings, financial statements, market data, meeting transcripts, and relationship records enter through controlled pipelines and become a shared, analysis-ready research base.

The platform supports origination, forecasting, underwriting, peer analysis, and monitoring. Quantitative methods, including machine learning and AI, support that work. Analysts direct the research, and material outputs are reviewed before they enter the research record.

The Platform

One system, end to end.

20+ governed data pipelines feed a broad feature space, multiple model families and a large number of forecasting, scenario and research configurations.

20+
Governed data pipelines
Multiple
Model families across regression, tree ensembles and classification
Broad
Financial, operating, semantic and relationship features
Continuous
Forecasting, scenario and research configurations on time-based splits
01

Data Sources

  • Company and financial data
  • Market and transaction data
  • Operating and sector data
  • Documents and meetings
  • Relationship and entity data
02

Data Pipelines

20+governed pipelines

  • Ingestion
  • Normalization
  • Period alignment
  • Validation
  • Identity resolution
  • Data quality
  • Feature publication
  • Provenance
03

Research Data Platform

  • Entity-resolved company records
  • Financial and operating histories
  • Structured documents
  • Feature tables
  • Embeddings and vectors
  • Ontology and knowledge graph
  • Relationship data
  • Source provenance
04

Models and Methods

Forecasting and predictive models

Gradient-boosted trees, regularized regression, and random forests.

Language models and semantic extraction

Entities, themes, and relationships from documents and transcripts.

Similarity, graph and entity models

Hilbert-space similarity, ontology, and the knowledge graph.

Retrieval and research workflows

Vector retrieval and source-linked research synthesis.

05

Investment Workflows

  • Origination
  • Forecasting
  • Underwriting and scenarios
  • Comparable-company selection
  • Valuation
  • Meeting and relationship intelligence
  • Research synthesis
  • Monitoring

Testing, provenance and analyst review

  • Time-based validation
  • Data-quality tests
  • Source traceability
  • Feature and model versions
  • Analyst approval
  • Recorded corrections

What the Platform Supports

Three areas of research work.

Origination and Meeting Intelligence

Meeting transcripts, company materials, and relationship data enter the system as they arrive. Language models extract entities, operating themes, and possible opportunity signals, and each extraction keeps a reference to its source. Analysts review the results before they join the research record.

Forecasting and Underwriting

Normalized financial and operating histories feed predictive models, together with semantic features drawn from documents and transcripts. The models support forecasting, scenario analysis, and valuation work, and every model is evaluated against historical periods before the team relies on it.

Company, Peer and Relationship Analysis

An ontology and knowledge graph connect companies, sponsors, and executives, and entity resolution keeps those records consistent. Similarity models compare companies in a normalized feature space, so peer sets reflect how companies actually operate. Industry labels alone often miss that.

Methods

Selected technical methods.

Forecasting models

Gradient-boosted trees handle non-linear interactions across financial and operating features. Regularized regression provides an interpretable baseline, and random forests give a variance check on both.

Language models and semantic extraction

Language models read filings, company materials, and meeting transcripts. They extract entities, operating themes, and relationship changes, and map each result to a structured field with a reference back to the source text.

Embeddings, retrieval and similarity

Documents and companies are embedded into vector representations. Vector retrieval assembles evidence for research, and Hilbert-space similarity compares companies across many normalized dimensions to build peer sets.

Ontology and knowledge graph

An ontology defines the entity types and relationships the firm tracks. The knowledge graph records how companies, sponsors, executives, and markets connect, and entity resolution keeps names consistent across sources.

Validation

Forecasting models train on earlier periods and are tested on later ones. Random row splits are not used, because they leak future information into training.

Feature-ablation testing

New feature groups earn their place by comparison. A baseline specification is measured against specifications that add each semantic feature group, separately and combined.

Review and Controls

Nothing enters the research record unreviewed.

The platform is governed at each step. These controls are built into the system, not applied as a later check.

Data-quality tests
Pipelines validate structure, ranges, and continuity before data is published to the research platform.
Temporal validation
Models are measured on later periods than they trained on, against baseline comparisons.
Source provenance
Every extracted value and research output keeps a link to the document, filing, or transcript it came from.
Model and feature versions
Each model version is recorded with the feature definitions and data it was trained on.
Analyst approval
Material changes are reviewed by an analyst before they enter the research record.
Recorded corrections
Corrections are stored with their reasons and are available to later model and research runs.

Learn more about the platform.

Request the BM² overview presentation or contact the firm directly.