
SnowConvert AI speeds up one important piece: assessing and converting legacy source code. This article covers what the tool is, how its conversion workflow works, how it fits into a full migration plan, and when outside implementation help makes sense. Features change often, so confirm specifics against current Snowflake documentation.
Key Takeaways
- SnowConvert AI accelerates code conversion, but code is only one part of a production migration.
- Plan for assessment, dependency mapping, data movement, testing, security review, and cutover.
- Business owners and engineers must still review flagged functional differences.
- Validate every important workload at structural, behavioral, and data levels before retiring the legacy system.
What Is SnowConvert AI?
Snowflake's documentation calls the tool SnowConvert AI (SCAI), and some pages shorten it to SnowConvert. Its core job is converting supported source-database code into Snowflake code and objects.
Interfaces you will encounter
- Desktop application: a downloadable local app. Snowflake says source and target code is not transmitted, while operational usage data is shared.
- CLI (
scai): scriptable, with GA and Preview build channels. Snowflake recommends the stable version for normal use. - VS Code Migration Assistant: Public Preview. It explains issue root causes and offers chat-based fix suggestions.
Preview conversion settings are for evaluation only. Snowflake says not to use them in production.
Supported source platforms and objects
Support varies by platform. A few examples from Snowflake's current matrix:
| Source | Objects converted | Extras listed |
|---|---|---|
| Teradata | Tables, views, procedures, functions, BTEQ, MLOAD, TPUMP | None |
| Oracle | Tables, views, procedures, functions, packages | None |
| SQL Server | Tables, views, procedures, functions | Data migration, deployment, AI code conversion |
| Amazon Redshift | Tables, views, procedures, functions | Data migration, deployment, AI code conversion |
| BigQuery, PostgreSQL | Tables, views | AI code conversion |
Other platforms, including Synapse, Netezza, Greenplum, Vertica, Hive, and DB2, are listed with their own object coverage. Check the matrix for your source before you scope anything.
Just as important is what the matrix does not list as converted objects:
- BI reports and semantic models
- General orchestration and security policies
- Application connections
The one documented ETL path turns SSIS and Informatica PowerCenter into a dbt project. It does not cover arbitrary ETL tools.
What "AI-assisted" means
Snowflake describes a transpiler built on an abstract syntax tree (AST) and symbol-table model, with agentic AI refining results. In practice, AI explains issues and suggests fixes. A person still approves each change.
Data handling differs by interface. The Migration Assistant sends an issue and surrounding code snippet to Snowflake Cortex, and the documentation says this happens within your Snowflake account. It does not state prompt-retention or model-training terms, so have your security team review them first.
How SnowConvert AI Supports Code Conversion
Assessment and source inventory
Inventory before you convert. List schemas, objects, scripts, procedures, dependencies, and unsupported constructs. The assessment report estimates a conversion rate from source lines of code and includes a top-level unit inventory with issues.
That rate measures code conversion, not behavioral equivalence. Use it to estimate effort and sequence work, not to declare victory.
Good input hygiene helps:
- Extract the latest production-aligned definitions
- Exclude system and backup objects
- Keep inputs and outputs in a versioned repository
Semantic parsing and translation
Plain text replacement breaks on real code. The AST and symbol table let the tool understand objects, references, variables, data types, and control flow. Platform-specific translation rules then map source syntax into Snowflake equivalents. Snowflake's product page says fully custom conversion rules are not supported.
Where no direct equivalent exists, the output carries three kinds of markers:
- EWI: error or incomplete conversion. Unresolved markers can prevent compilation.
- FDM: converted, but behavior may differ. It can compile and still need business review.
- PRF: a performance warning.
AI-assisted remediation and human review
A workable loop looks like this:
- Review unresolved EWIs and FDMs
- Use AI guidance to propose a fix
- Apply only an approved change
- Rerun tests
- Document decisions on data types, null handling, date logic, and procedural behavior

On automation, the CLI documents extraction, deterministic conversion, deployment, data migration, and validation commands. scai test covers seed, capture, and validate workflows plus stored-procedure test generation. Structured JSON output supports CI/CD scripting. The CLI documentation does not describe a separate packaged batch product or managed CI/CD service, so confirm what your source platform and interface support.
How to Use SnowConvert AI in a Complete Snowflake Migration
Snowflake's migration overview describes migration as moving schemas, code, data, pipelines, and consumers. It separates code conversion, historical loading with incremental catch-up, validation, and controlled cutover. SnowConvert AI mostly addresses the first.
Plan the target architecture first
Before generating target code in volume, settle:
- Account structure, databases, and schemas
- Roles, environments, and naming conventions
- Data classification rules
Then set migration waves by business criticality, dependency order, complexity, data freshness, and tolerable downtime. Converting everything at once is how teams end up with thousands of objects nobody can test.
Keep code conversion and data movement separate
Conversion produces code. Moving data is its own design problem. SnowConvert AI's data migration documentation covers SQL Server, Redshift, Teradata, Oracle, and PostgreSQL, using an orchestrator/worker model that stages data and loads with COPY INTO. Its sync strategies are:
none: default full extractionchecksum: compares partition hashes and re-extracts changed partitions (not supported for Oracle)watermark: incremental by marker
Snowflake's general loading options, such as bulk COPY INTO and continuous Snowpipe ingestion, are separate from conversion. Plan full and incremental loads, source-to-target mappings, retention, sensitive-data handling, and rollback or coexistence as their own workstream.

Map ETL, orchestration, BI, and application dependencies
Map these dependencies before renaming anything:
- Schedulers and ETL or ELT tools
- dbt projects and stored procedures
- APIs, semantic layers, and dashboards
- Downstream applications
Repointing them means separate code changes, credential updates, user acceptance testing, and coordinated releases.
Build layered validation into every wave
- Structural: objects, data types, keys, grants, dependency resolution
- Behavioral: query results, procedure outcomes, null handling, rounding, date logic, error behavior
- Data: row counts, column metrics such as min/max/average, sampling, row-level comparison
Test representative business queries and high-risk edge cases with business owners. Record exceptions and acceptance criteria. Compiling cleanly proves nothing about correctness.

Prepare for cutover and operations
Before go-live, rehearse cutover, synchronize source and target, set freeze windows, and define rollback triggers and incident ownership.
After go-live, budget for:
- Cost and performance tuning
- Observability and data-quality monitoring
- Access reviews
- Knowledge transfer to the team that will run the platform
Validation, Limitations, and Decision Criteria
What SnowConvert AI can and cannot prove
Successful translation or schema deployment doesn't prove:
- Data parity
- Equivalent business outcomes
- Complete dependency coverage
- Acceptable performance
- Production readiness
Treat vendor-documented capabilities separately from assumptions, especially in third-party write-ups that describe end-to-end migration, ETL conversion, or BI repointing loosely.
Risks that need additional engineering
- Dynamic SQL: SQL Server dynamic SQL is flagged as an EWI because runtime-built statements are hard to analyze.
- Teradata specifics: CHECK constraints using unsupported UDFs can be flagged, and
COMPRESSexpressions are removed. - Out-of-scope objects: encrypted or unreadable objects, unsupported languages such as Java, and Teradata triggers.
- Hidden risks: undocumented business rules, data-type differences, procedural side effects, late-arriving data, and changes in downstream reporting.

Keep a risk register with owners, severity, test evidence, unresolved conversion messages, and explicit business sign-off for accepted differences.
When self-service fits
Self-service suits teams with:
- Contained, well-understood workloads
- Strong in-house Snowflake skills
- Manageable validation scope
Get deeper assessment and implementation help when the environment is regulated, highly interdependent, high-impact, or poorly documented.
How AICG Can Support a SnowConvert AI Migration
AICG is a senior-led data and AI consultancy based in Charlotte, NC, and a Snowflake Select Partner. Its migration offering covers SQL Server, Oracle, and Teradata sources.
Assess and define the roadmap
AICG's fixed-scope Snowflake Architecture & Readiness Assessment reviews the data estate, architecture, governance, warehouse strategy, data sharing, and AI readiness.
The output is a target architecture, a sequenced roadmap, and a business case. From there, migrations start with an inventory of schemas, stored procedures, and ETL dependencies.
Build and validate production systems
Conversion is one input to the build. Around it, AICG's architects, engineers, and project managers handle:
- Ingestion with DLH.io, including historical and incremental loads
- dbt models with contracts, tests, and CI/CD through GitHub Actions
- Automated parity validation against the source
- Sequenced, reversible cutover waves

The goal is a working system your team can own, with accountability from architecture through go-live.
Continue with managed support
After cutover, AICG's run and optimize phase covers:
- Managed support
- FinOps reviews
- Warehouse right-sizing
- Resource monitors
- Continuous delivery
Managed Snowflake Support is available through AWS Marketplace.
Planning a move to Snowflake? Talk with AICG about your source platforms, workload complexity, target architecture, and validation needs before you commit to an approach.
Frequently Asked Questions
Is Snowflake AI free?
Not entirely. Snowflake's product page describes SnowConvert as free for listed code-conversion use, and training is free. But the Cortex pricing bills AI functions per million tokens, and warehouse compute and implementation services cost extra.
What is Snowflake migration?
It is the movement of data, code, workloads, security, and dependencies from a legacy platform into Snowflake. Code conversion is one part of it.
What are the best practices for migrating data to Snowflake?
Assess first, map dependencies, and migrate in phases. Transfer data securely, automate tests, reconcile source to target, rehearse cutover, and monitor after go-live.
Is Snowflake an ETL tool for data migration?
No. Snowflake is a cloud data platform that supports loading and SQL transformation (ELT). Extraction, orchestration, and migration automation often need separate tools or services.
What does Snowflake AI do?
Snowflake's broader AI services, such as Cortex, support analytics and AI workloads. SnowConvert AI is a migration tool that uses some of them for conversion help. Check current documentation for feature boundaries.
Is SnowSQL deprecated?
Snowflake labels SnowSQL a legacy client and recommends Snowflake CLI for new work. SnowSQL is documented as supported through April 16, 2028. Both differ from SnowConvert's scai command.