NTech IncSnowflake Partner

Home / Snowflake / Migration / Teradata to Snowflake

Teradata migration

Teradata to Snowflake migration partner for BTEQ, TPT and reconciled reports

For data platform teams moving a Teradata warehouse to Snowflake. NTech, a Snowflake Partner Network member, delivers a DBQL-based inventory and wave plan, conversion of DDL, SQL, BTEQ scripts, macros and stored procedures with Snowflake's SnowConvert AI plus engineer review, history loads with Teradata Parallel Transporter and COPY INTO, table- and report-level reconciliation, and a parallel run until each wave is signed off. A principal architect in the US or Europe owns the design; engineers in India work 14:00–23:00 IST, overlapping 09:00–13:30 US Eastern.

Quick answer

What does a Teradata to Snowflake migration partner do?

A Teradata to Snowflake migration partner inventories what is actually used from DBQL query logs, converts DDL, SQL, BTEQ scripts, macros and stored procedures with Snowflake's SnowConvert AI plus engineer review, exports history with Teradata Parallel Transporter and loads it with COPY INTO, then runs both platforms in parallel and reconciles row counts, checksums and reports before cutting over wave by wave.

Last updated · NTech Inc

Object mapping

Teradata objects and features mapped to Snowflake

Standard SQL converts easily. These rows are where Teradata behavior differs and most rework and testing time goes.

TeradataSnowflake equivalentWhat to watch
BTEQ scriptsSnowflake Scripting or Python (SnowConvert AI can output either), run by Airflow, dbt, Snowflake Tasks or the Snowflake CLI.IF ERRORCODE, .GOTO, .EXPORT and .IMPORT logic, restarts and alerts must be rebuilt
TPT, FastLoad, MultiLoad, TPumpFiles in cloud storage loaded with COPY INTO, or Snowpipe for continuous loads; MERGE for upsertsUpsert logic becomes MERGE; error tables and restart checkpoints need a new design
SET tablesStandard tables, which allow duplicate rowsTeradata silently drops duplicate rows; Snowflake keeps them, so loads must deduplicate (QUALIFY ROW_NUMBER() or MERGE)
MULTISET tablesStandard tablesDirect mapping; same duplicate behavior
Primary index (UPI, NUPI)No indexes; automatic micro-partitions. Primary and unique keys can be declared but are not enforcedUniqueness that a UPI enforced must be enforced in the pipeline or tested with dbt tests
Partitioned primary index (PPI)Natural load order, or a clustering key on large tablesAdd clustering keys only where query profiles show poor pruning; reclustering consumes credits
Secondary and join indexesSearch Optimization Service, materialized views or clustering keysRarely needed one-for-one; decide per query from DBQL evidence. Some need Enterprise Edition
MacrosViews for simple macros; stored procedures for parameterized onesMulti-statement macros become procedures with explicit transactions
Stored procedures (SPL)Snowflake Scripting procedures (or JavaScript/Python)Cursors, handlers and dynamic SQL need engineer review
QUALIFY, SAMPLE, TOPQUALIFY, SAMPLE and TOP are supportedUsually converts directly; check sampling semantics where results feed tests
Volatile tablesTemporary tables (session-scoped)Multi-step scripts must keep one session in the orchestrator
Global temporary tablesTemporary or transient tablesCreate them inside the job; definitions do not persist
COLLECT STATISTICSNot neededRemove from scripts; Snowflake maintains table metadata automatically
NOT CASESPECIFIC columnsComparisons are case-sensitive by defaultUse collation or UPPER(), or joins and filters return different rows
FORMAT clauses, date and PERIOD typesTO_CHAR / TO_DATE formats; PERIOD becomes two start and end columnsA common source of silent result differences
TASM workload managementSeparate virtual warehouses per workload, multi-cluster warehouses and resource monitorsSize from DBQL workload profiles; set auto-suspend and credit limits early
Fallback, journals, backupsTime Travel and Fail-safeLong retention on high-churn tables raises storage cost

Tooling

The Teradata to Snowflake toolchain

Snowflake's SnowConvert AI translates Teradata SQL, DDL, BTEQ and load scripts and stored procedures. In June 2026 Snowflake announced Snowflake AIM, a migration platform that builds on SnowConvert AI and Datometry virtualization. Names and licensing change, so we confirm current options at kickoff.

TaskToolsWhat engineers still do
Inventory and usageDBQL query logs, DBC dictionary views, scheduler exports, SnowConvert AI assessment reportsAgree what is used and what to retire
Code conversionSnowConvert AI (inside Snowflake AIM)Fix flagged items, rewrite procedures and BTEQ control flow, review every converted object
History exportTeradata Parallel Transporter (TPT) export to compressed, split filesPlan extract order and file sizes per wave
Staging and loadAmazon S3, Azure Blob Storage or Google Cloud Storage stages; COPY INTO; SnowpipeLoad in parallel, track rejects
Ongoing sync during parallel runDual-running converted loads, or change data capture with replication tools such as Fivetran or AWS DMSKeep both platforms on the same data cut so comparisons are fair
Transformations and orchestrationdbt, Airflow, Snowflake Tasks, Snowflake CLIRebuild schedules and dependencies from BTEQ wrappers
ValidationSnowflake's validation tooling (schema, metric and row-hash comparisons), dbt tests, Great Expectations, custom reconciliation SQLDefine tolerances, chase every difference to a root cause, collect business sign-off
Optional: run unchanged firstDatometry virtualization (Teradata workloads on Snowflake with minimal rewrites)Decide whether to lift and shift first, then refactor

Phased plan

How a Teradata to Snowflake migration runs, step by step

Each business domain moves as a wave. Teradata keeps running for that domain until its tables and reports reconcile and the business owner signs off.

  1. Inventory from DBQL

    List tables, views, macros, procedures, BTEQ and TPT scripts, scheduler jobs, and the reports and extracts that read from Teradata. DBQL logs show what is actually queried, so unused objects are retired, not migrated.

  2. Wave plan

    Group objects by business domain (for example finance, sales, operations) so each wave delivers something usable. Agree tolerances and sign-off owners per wave.

  3. Target design

    Schemas, role-based access, masking and row access policies, warehouse sizing and resource monitors. See Snowflake implementation for the platform setup.

  4. Convert and unit test

    Run SnowConvert AI on DDL, SQL, BTEQ and procedures. Engineers fix flagged items, deduplicate former SET-table loads, handle case sensitivity, and rebuild orchestration. A senior engineer reviews every pull request.

  5. Move history

    Export with TPT to compressed files, stage in cloud storage, and load with COPY INTO in parallel. Record row counts at both ends.

  6. Parallel run

    Run converted loads in Snowflake alongside Teradata, or replicate changes, so both hold the same data. Run the parallel period across at least one full business cycle, including a month-end close where finance reports depend on it.

  7. Reconcile and sign off

    Compare every table at several levels (counts, aggregates, row hashes), then compare key reports side by side. Each difference gets a root cause and a fix or an agreed explanation before sign-off.

  8. Cut over and decommission

    Switch BI tools, extracts and feeds, keep Teradata read-only for a short fallback window, then retire its jobs and, finally, the platform.

Reconciliation

How we prove the numbers match

LevelCheckCatches
1. SchemaColumn names, data types, precision, nullabilityTruncated decimals, wrong timestamp types, lost precision
2. Row countsPer table and per partition or load dateMissing files, rejected rows, duplicates from former SET tables
3. Column aggregatesSUM, MIN, MAX, COUNT DISTINCT and null counts per columnRounding, NULL handling, date conversion errors
4. Row hashesHash of each row or key range, compared across platformsIndividual changed values that aggregates hide
5. ReportsKey dashboards and extracts run on both platforms with the same parametersLogic differences such as case-insensitive joins or date arithmetic
6. Business sign-offReport owners confirm results for their domainAnything the tests did not anticipate

What to measure

Metrics that show a migration is on track

Data volume says little about progress. These show work done, risk left and whether credits are under control.

We report them in the weekly delivery review. See how we work.

  • Inventory coverage: share of DBQL-active objects assigned to a wave or marked for retirement
  • Conversion status: objects converted automatically, converted with manual fixes, and still open
  • Reconciliation pass rate: tables and reports passing each level, and open differences by root cause
  • Report sign-offs: key reports signed off by business owners, per wave
  • Query performance: run time of the most frequent and most expensive DBQL queries, Teradata versus Snowflake
  • Credit consumption: credits per warehouse against the agreed budget, including during the parallel run
  • Pipeline SLAs: load completion times against the batch window
  • Decommission progress: Teradata jobs and tables switched off

Case study

Parallel-run experience from a regulated migration

This was a mainframe core-banking migration, not a Teradata migration. It shows the discipline a Teradata move needs: phased waves, a parallel run and evidence an auditor accepts.

Financial services

Mainframe core banking to cloud-native

Duration
18 months
Team
12 senior engineers

The client moved its core banking system off a mainframe to a cloud-native architecture.

Approach

  • Phased migration
  • Parallel run of old and new systems
  • Compliance controls built in from the start

Outcomes

  • Zero-downtime migration
  • Regulatory audit with no findings

Client name withheld; the full version is available under NDA. Read the case studies.

Why move

Why companies leave Teradata for Snowflake

Cost model. Teradata capacity is bought ahead of demand. Snowflake bills compute for the time warehouses run, so quiet periods cost little, provided warehouses are sized and suspended well. See Snowflake cost optimization.

Workload isolation. Loads, transformations and BI each get their own warehouse, so a heavy batch job no longer slows dashboards.

Less administration. No index or statistics maintenance, no space management and no hardware refresh cycles.

Deliverables

  • Inventory and wave plan from DBQL usage
  • Object mapping and conversion report, including what was retired
  • Converted, tested SQL, procedures and scripts in your repository
  • Rebuilt orchestration (Airflow, dbt or Snowflake Tasks)
  • Reconciliation report for every table and key report
  • Snowflake role model, masking policies and warehouse plan
  • Runbooks, handover sessions and a decommission plan for Teradata

Decision guide

How to choose a Teradata to Snowflake migration partner

Check a firm's status in Snowflake's partner directory first. Then use these questions to see whether it has planned for Teradata's specific behavior.

Questions to ask

  • Will you base the inventory and estimate on DBQL usage, or on a raw object count?
  • How do you handle SET-table duplicates, NOT CASESPECIFIC columns and unenforced keys?
  • What will BTEQ scripts become, and who rebuilds error handling and restarts?
  • Which reconciliation levels do you run, and who signs off each wave?
  • How long is the parallel run, and how will you control credits during it?

Evidence to request

  • A sample object mapping and reconciliation report (anonymized)
  • Named architect and their hours of overlap with your team
  • Their Snowflake partner status, checked in Snowflake's directory
  • Security controls: access, data masking, background checks, offboarding
  • A wave plan with cutover and fallback criteria, not just a finish date
OptionWorks best whenWatch for
In-house team with Snowflake's toolsSmall estate, strong Teradata and Snowflake skills already on staffConversion tools leave procedures, BTEQ control flow and reconciliation to you
Migration partner (fixed scope)Many scripts and reports, a firm deadline, need for an owned planScope that counts objects but ignores reports and downstream extracts
Partner engineers added to your teamYou own the plan and need Teradata-experienced handsYour team must provide architecture and sign-off ownership

NTech works in all three ways: fixed-scope project, dedicated team or staff augmentation, or you can hire Snowflake developers with Teradata experience. Security practices are described on our security page.

Contact us

Plan your Teradata to Snowflake migration

Tell us roughly how big your Teradata estate is and how many scripts and reports depend on it. A senior data architect will reply with an approach.

  • A senior engineer reads every message
  • Reply within one business day
  • No obligation, and your details are used only to reply

Prefer the full form? Go to the contact page.

FAQ

Questions buyers ask us

What does a Teradata to Snowflake migration partner do?

It inventories what is actually used from DBQL query logs, converts DDL, SQL, BTEQ scripts, macros and stored procedures with Snowflake's SnowConvert AI plus engineer review, exports history with Teradata Parallel Transporter and loads it with COPY INTO, then runs both platforms in parallel and reconciles tables and reports before cutting over wave by wave.

Which tool converts Teradata code to Snowflake?

Snowflake's SnowConvert AI converts Teradata DDL, SQL, BTEQ and load scripts, and stored procedures. In June 2026 Snowflake announced Snowflake AIM, a migration platform that builds on it. Engineers still fix flagged items, rewrite complex procedures and rebuild orchestration.

Does Snowflake support Teradata SET tables?

No. Snowflake tables allow duplicate rows, so pipelines that relied on SET tables to drop duplicates must deduplicate explicitly, for example with QUALIFY ROW_NUMBER() or MERGE.

What replaces BTEQ in Snowflake?

Snowflake Scripting or Python scripts (SnowConvert AI can produce either), run by an orchestrator such as Airflow, dbt, Snowflake Tasks or the Snowflake CLI. BTEQ error handling such as .IF ERRORCODE and .GOTO must be rebuilt in the new tool.

How long does a Teradata to Snowflake migration take?

It depends on the number of scripts, procedures and reports more than on data volume. A DBQL-based inventory gives a reliable estimate and a wave plan, and the first wave can go live before later waves are converted.

How long should the parallel run last?

Long enough to cover at least one full business cycle for the wave, including a month-end close where finance reports depend on it, and until every table and key report reconciles and the business owner signs off.

How do you check that the data matches after migration?

We compare schemas, row counts, column aggregates and row hashes for every table, then run key reports on both platforms with the same parameters. Each difference is traced to a root cause and fixed or explained before sign-off.

Can we hire Teradata-experienced engineers instead of outsourcing the whole migration?

Yes. NTech offers staff augmentation and dedicated teams as well as fixed-scope projects. With augmentation, your team keeps ownership of the plan and sign-off while our engineers convert, load and reconcile.

Next step

Tell us what you need to build or who you need to hire.

A 30-minute call with a senior architect. You leave with a scoped plan or a role profile, whether or not you work with us.