Semarchy

Overview

Semarchy's product is now the Semarchy Data Platform, launched in September 2025, of which its master data module is one part alongside integration and governance. Older descriptions naming only the master data module are out of date.

It is the most purely model-driven of the four. You define your own model, with pre-built accelerators offered as optional starting points rather than as the product, expressed in its own SQL-like modelling language.

Its real architectural distinction is hub style. It supports registry, consolidation, coexistence and centralised styles in parallel across domains, so you can start virtual and evolve to physical through configuration rather than a rebuild.

That makes it the only one of the four that genuinely offers both a virtual registry view and a physical golden record, which is a real fork in this category and usually forces a product choice.

When physical, the golden records are plainly available: it writes master records and golden records to their own tables in your database, queryable directly with SQL, with separate error and history tables.

Because the hub lives in a database you control, consumption can be as simple as a SQL query.

It also runs natively inside Snowflake as an application, the only one of the four to do so, and is the only one offering a genuinely air-gapped on-premises deployment.

Its engineering workflow is unusual for this category, with a design process built around the tools data engineers already use, including version control and continuous delivery pipelines.

Matching runs in two phases, binning records into small groups then comparing pairs within them, with match groups forming transitively, and rules written as either exact comparisons or similarity thresholds.

Survivorship has a two-part structure its rivals do not publish: a consolidation rule deciding the value from duplicates, and a separate override rule deciding whether a human-authored value beats it.

Key features and capabilities

The same headings are used for every master data management entry, so two tools can be read side by side.

Domains and data model

• Customer, product, supplier, location and reference data, plus employee, financial hierarchy, assets and materials
• The data model is user-defined, which is the core of the product, with optional accelerators as starting points
• Multidomain is native, and models are reusable across governance, master data and AI initiatives
• An accelerator marketplace launched in February 2026 supplies pre-built starting models
• All four hub styles run in parallel across domains, which none of its rivals offers

Matching and survivorship

• Matchers group similar records into duplicate clusters, merged or confirmed by confidence score
• Two phases, binning records into small groups then comparing pairs within each bin
• Match groups form transitively, so a matching b and b matching c puts all three together
• Rules are written in its own language, as exact comparisons or similarity thresholds, plus vector embeddings
• Survivorship pairs a consolidation rule with a separate override rule governing human-authored values

Stewardship and workflow

• AI-powered stewardship with explicit human-in-the-loop workflows
• Workflows enforce policy with dynamic task routing rather than fixed queues
• Stewards work in data applications generated from the model, with personalised interfaces
• Role-based access control and audit logging, plus policy-aware access logging on consumption
• Merge suggestions handle below-threshold groups; unmerge is not published as a named feature

Hierarchies, quality and governance

• Hierarchy management, with financial hierarchy published as a domain in its own right
• Reference data as a listed domain with its own model capability
• A continuous automated certification pipeline for enrichment, validation, matching and quality rules
• Quality rules can run in real time or batch as part of that pipeline
• Its own governance module harvests the master data module, so catalogue integration is first-party

How mastered data is consumed

• Golden records are queried directly with SQL from their own tables, unusual in this category
• REST APIs generated from the model, with documented query paths
• Agent protocol endpoints generated per data product, carrying semantic context and access rights
• Data applications give people a generated interface for direct consumption
• Reverse integration to sources runs through its integration module and publish tables

Integrations

• The deepest Snowflake integration here, running natively as an application inside it
• Microsoft Fabric integration announced in September 2025, with medallion architecture compatibility
• Red Hat support expanded for self-hosted deployment in September 2026
• Runs on or with AWS, Azure, Snowflake, on-premises or hybrid
• Its integration module supplies source connectivity; named application connectors are not enumerated

Where it runs and what it costs

• Four options, managed service, Snowflake-native, self-hosted cloud, and self-hosted on-premises
• The managed service publishes a 99.9% availability commitment with SOC 2 and ISO 27001
• On-premises supports OpenShift, Rancher and genuinely air-gapped environments, unique among the four
• An explicit portability promise, moving between options without re-platforming or redesigning models
• Available through the Azure, AWS and Google Cloud marketplaces; regions are not published

Pricing

Price on requestQuote only; 7 and 30-day trials published

Quote-only, and the least specific of the four on the unit: no pricing page, no amounts and no named metric at all, so it should not be described as priced per record, per domain or per user. What is published is the cost structure per deployment, an annual subscription for the managed service, a licence plus infrastructure when self-hosted in cloud, a licence plus hardware on-premises, and pay-as-you-go through Snowflake. Its trial terms are the most precise here though: seven days self-serve with no approval, or thirty days after email activation.

Vendor pricing page →

Demos and videos

• Introducing the Semarchy Data Platform Product demo · YouTube
• Semarchy Data Platform overview Tutorial · YouTube

About Semarchy

Semarchy was founded in 2011 in Lyon by a team of data experts, though it names no individual founders, and is now headquartered in Phoenix with offices in the UK, France and India. It is private, backed by PSG Equity since September 2020 on undisclosed terms, with no other rounds published and no acquisition in either direction; Ben Werth became chief executive in June 2024. It reports more than 400 clients including Fortune 500 companies, over a trillion consolidated master records managed and more than 190 partners, and holds ISO 27001 and SOC 2. Named customers include Chipotle, Sanofi, Elsevier and Pernod Ricard.

Founded 2011 · Phoenix, Arizona · semarchy.com