| Proof before you buy | Demo and proof of concept run with the vendor or an integrator[1][2] | Install and try it right away: GX Core and Soda Core are open source[3][4] | Demo or trial requested through a sales form[5] | Self-service free account, no card and no sales call, plus a public proof on open data: 30 files scanned in 32 s |
|---|
| Time-to-value | A project with an integrator and implementation services; Collibra DQ is installed in the customer's environment[1][2] | First check in hours; coverage and investigation are ongoing engineering work | Metadata monitors once the warehouse is connected, without querying the tables (Monte Carlo), as SaaS or with a collector in your environment; Metaplane documents exceptions for Postgres, MySQL and SQL Server[6][7][8] | First scan in under 15 minutes; metadata monitors on every dataset in the connection from the first scan, up to 500 per connection |
|---|
| Fixing | Cleansing, standardization and MDM that write to your systems (Informatica IDMC) — requires write credentials and a project[2] | Tests fail. Someone investigates and fixes by hand | Detects, alerts and helps investigate; the fix stays with your team[5][7] | Proposes the fix in SQL/PySpark or as a pull request, with € savings and risk — read-only by contract |
|---|
| Fixes where the code lives | Workflows inside the platform[1][2] | You write the PR | An alert linking to the platform; the PR is up to your team[5] | An accepted fix opens a pull request in GitHub or GitLab, syncs its state, and the next scan confirms it worked |
|---|
| Tests you already have | Suite-native rules; no announced import of dbt, Great Expectations or Soda[1][2] | They are your tests — and your maintenance | Monitors in their own YAML or in dbt meta; Monte Carlo generates monitors from dbt tests, singular tests included, and Sifflet opens incidents when a dbt test fails; no announced export back as dbt tests or import of Great Expectations or Soda[9][10][11][12][13] | Imports dbt tests, singular tests included, Great Expectations suites and SodaCL and exports back to dbt, SodaCL and GX; relationships becomes a referential integrity rule; what doesn't convert (macros and Jinja beyond ref, source and config), or wouldn't come back unchanged from the SodaCL and GX export, is reported, never silently dropped |
|---|
| Custom checks | Informatica: rule specifications built from graphical elements or IF-THEN-ELSE expressions; Collibra: custom SQL rules with joins and aggregations, hand-written or AI-suggested — pushdown jobs only[14][15][16] | SQL and Python without limits: singular tests and macros (dbt), custom expectations in Python or SQL (Great Expectations), SQL checks (Soda)[17][18][19][20] | Metadata monitors; Monte Carlo accepts custom SQL monitors, with manual or learned thresholds[6][21] | 16 rule types (nulls, uniqueness, patterns, ranges, freshness, volume, check-digit validity for tax IDs and IBANs, reference lists, opt-in accuracy against authoritative sources, streams (Kafka and Pub/Sub, not yet installed in the SaaS), cross-column conditions, custom SQL and referential integrity across datasets), industry packs and suggestions from the profile and from acceptance history; custom SQL is a single read-only SELECT, validated before it runs in the source (SQL sources only); no custom Python checks |
|---|
| Work signals and new rules | Stewardship workflows: issues are logged by hand and rules are written by stewards[1] | Complaints arrive as tickets; you write and maintain every test | Open tickets in Jira or ServiceNow and learn monitors; they don't read your teams' complaints[5] | Reads Jira, Confluence and Slack and turns complaints into issues linked to the right dataset; proposes rules from the profile, from the resolved issue and from those complaints — they only land when you accept them. The Teams connector is built, with its driver not yet installed in the SaaS |
|---|
| Memory of what already happened | Glossary and manual stewardship[1] | — | Incident history per table[5] | A knowledge base of reports and diagnoses, semantic search and a sources↔data↔knowledge graph; the assistant and the diagnoses cite it |
|---|
| Governance and stewardship | Catalog and business glossary are the suite's system of record; in Collibra, quality scores sit right on the data marketplace[22] | Out of scope: tests live in your repository and wiring them into the catalog is up to you | The focus is detecting and investigating incidents; formal governance and stewardship aren't the product[5][7] | Business glossary linked to columns, owner and steward per dataset, tags and an exportable data dictionary; publishes verdicts to your OpenMetadata or DataHub and reads owners, domains and tags from it |
|---|
| Multi-source view | A broad suite, licensed per module[1][2] | One test repository per source | Strong on warehouses; limited operational SaaS sources[5][7] | Warehouses, operational DBs and SaaS in a single view |
|---|
| Business-process integration (SAP, Salesforce) | Quality inside the process, with write-back: SAP MDG validates master-data changes against business rules before activating them; Informatica cleanses, standardizes and validates with agents and brings data quality into applications (Informatica Headless). Collibra connects to S/4HANA through Edge, for the catalog[23][24][25] | Tests run in your pipeline, on data already written; calling them from SAP or Salesforce is integration you write and maintain | Monte Carlo connects to Salesforce CRM and SAP HANA and monitors data already written (freshness, volume, schema, SQL rules); no announced validation at the point of entry[26][27] | Reads SAP S/4HANA and SAP Gateway over OData, Salesforce, Dynamics 365 and the ERP databases on SQL Server (Primavera, PHC, Sage). Validates records at the point of entry: the integrator calls the dataset's rules through the API before writing them and gets a verdict for each record, with nothing stored by BeryX. Never writes to the ERP or the CRM: the fix comes out as SQL/PySpark or a PR |
|---|
| Trust for AI agents | Catalog and AI governance; no announced readiness verdict per use case[1][28] | — | Monitor data and pipelines; no announced verdict on the dataset before an agent consumes it[5][7] | Explainable trust score and readiness per use case (RAG, training, agent, BI), a pre-ingestion gate for RAG and an MCP server so the agent asks before it uses |
|---|
| Where the work runs — and what leaves your perimeter | Vendor SaaS or installation in the customer's environment (Collibra DQ installs there): once installed, processing stays on your infrastructure[1][28] | Runs in your pipeline, on your warehouse or your Spark: nothing leaves — operations and scale are on you | Read warehouse metadata and query logs, as SaaS or with the collector in your environment (Monte Carlo)[7] | On SQL sources, profiling runs in the source itself and only aggregates and a few examples per column come back; row-level rules and duplicates use a sample of up to 5,000 rows per table, processed in the EU. We keep results and limited examples, never the table. With AI on, examples follow the workspace's AI policy |
|---|
| AI in your own account | AI built into the vendor's platform; no announced own key or model in your cloud account[1][2] | — | AI built into the vendor's platform; no announced model in your cloud account[5][7] | Your own key, Claude in your AWS or GCP account, a per-workspace model policy with an “AI off” mode, and samples anonymised before any budget provider |
|---|
| AI Act compliance | AI governance as a suite module (Collibra: ISO 42001)[28] | — | No announced Art. 50 labelling or Art. 10 dossier[5][29] | Art. 50 labelling on every AI output + exportable Art. 10 dossier |
|---|
| Security, residency and certifications | Informatica: ISO 27001 (BSI) and SOC 2 Type II; Collibra: SOC 2 and ISO 27001/42001, with a public trust center. SaaS or installed in the customer's environment[1][2][28] | Runs inside your perimeter: security and certification are those of your own infrastructure | Monte Carlo and Sifflet: SOC 2 Type II and ISO 27001; Monte Carlo as SaaS or hybrid, with the collector in your environment[5][7][29] | SaaS hosted in the EU (Google Cloud europe-west1), read-only enforced in code and optional AI with your own key or in your own AWS or GCP account (Bedrock or Vertex), with samples that can be anonymised before the model. A fixed egress IP (Cloud NAT) for your firewall allow-list: built, not yet active in production. SOC 2 Type I in progress, ISO 27001 next — no badges we don't have: we don't claim them, and we answer with the technical evidence for each control |
|---|
| Identity and administration | SSO and role-based access control, included in the suite[1][2] | Open source has no users; SSO and roles come with paid cloud plans (e.g. Soda Team)[4] | SSO and roles on higher-tier plans[5] | OIDC single sign-on with your IdP (Entra ID, Google Workspace, Keycloak…), TOTP MFA mandatory for organisation admins and enforceable across the workspace, per-workspace roles, teams, an audit log exportable to CSV and support access only with your consent, read-only and for 72 h at most; no SCIM yet |
|---|
| Support and SLA | Support and availability levels in the contract — Informatica IDMC: 99.5% with service credits; Collibra: a public status page[1][2][30] | Community support for open source; support with an SLA only on paid plans[3][4] | Support and SLA by plan, in the annual contract[5] | No contractual availability or support SLA yet: it only goes into the contract after a quarter of measured availability and a support policy by severity, still being defined. Today: daily backups with 7-day PITR and offline mode if AI fails |
|---|
| Data exit | End-of-contract data extraction, under the service terms[1][2] | Tests and results are already yours, in code and in your warehouse | Data exports only on the Scale and Enterprise plans (Monte Carlo)[31] | Self-service export of the workspace data (GDPR) and of the configuration as code (datavault.yaml) |
|---|
| Real cost | Public UK G-Cloud prices: Collibra DQ £122,919 per licence/year; Informatica IDMC £74,999/year, plus implementation[1][2] | Licence 0; Soda Team $750/month, plus warehouse compute and engineering hours[4] | Annual contracts or per monitored table (Metaplane, with 10 tables free); Sifflet ≈$48k/year at published pricing[5][32] | Free plan; paid plans from €490/mo, no implementation project |
|---|