Audience: CIOs, CTOs, data platform leaders, analytics directors and migration program owners
A warehouse migration can look straightforward on a platform diagram: extract data, translate SQL, rebuild pipelines and point reports to the new destination. The real difficulty is usually hidden between those boxes. A table may feed a scheduled transformation, a finance model, a spreadsheet extract and an application API. A nightly job may depend on a temporary object created by another job. A report that appears unused may still support the monthly close.
These dependencies determine migration sequencing, testing effort and business risk. Discovering them late turns a planned cutover into a series of emergency fixes. BigQuery migration lineage is useful because it can make part of that dependency network visible before engineering teams commit to a detailed delivery schedule. It is not, however, a complete inventory of every system or a guarantee that translated SQL will behave identically.
The executive question is not simply whether the new warehouse can run the queries. It is whether the organization knows what must keep working, who depends on it, and how to prove the replacement is correct.
What changed: lineage is becoming part of migration discovery
Google Cloud introduced its BigQuery migration lineage service in Preview on September 16, 2026. It visualizes data flow and relationships in a source warehouse to support migration planning. Google’s documented supported SQL dialects include Amazon Redshift, Snowflake, Teradata and GoogleSQL. The service uses extracted metadata and query logs, not a magical scan of every downstream business process.
The official Google Cloud migration lineage documentation describes the setup, permissions, visualization and current limitations. The release notes establish the September 2026 introduction. Because this is a Preview capability as of this writing, teams should verify regional availability, feature behavior and support conditions before treating it as a production-critical dependency.
The strategic change is in discovery discipline. Instead of starting with a list of tables and a rough count of SQL scripts, a migration team can examine which objects are connected through observed workloads. This can reveal shared upstream dependencies and help organize a migration around coherent business flows rather than arbitrary database schemas.
Lineage is not just a diagram of tables
A useful migration inventory has at least four layers. The physical layer contains databases, schemas, tables, views, columns and storage locations. The execution layer contains SQL statements, jobs, procedures, orchestration schedules and service accounts. The consumption layer contains BI reports, exports, applications, APIs and downstream data products. The business layer contains the decisions and processes that depend on those outputs.
Migration lineage can help with relationships visible through source metadata and SQL activity. It may show that an orders table feeds an enriched sales table, which feeds a regional margin model. It may show which users or pipelines touched the objects. But a dashboard title, a regulatory filing obligation, a manual CSV handoff or a contractually required SLA may live outside the source query logs.
Treat automated lineage as evidence to enrich the inventory, not as the inventory’s sole authority. Business owners, BI administrators, application engineers and operational teams still need to confirm what the relationships mean.`

Why conventional discovery often underestimates migration risk
1. Object counts disguise dependency concentration
Two thousand tables do not necessarily mean two thousand equally difficult migrations. A small group of shared dimensions, reference tables and financial transformations may underpin most of the important outputs. Conversely, hundreds of isolated staging tables may be low risk. Prioritization should reflect downstream criticality and the number of dependencies, not simply object volume.
2. Query frequency is not the same as business importance
A frequently queried exploration table may be less important than a monthly revenue recognition model executed only during close. Log-derived activity is a helpful signal, but it needs a business criticality overlay. Classify outputs by financial, operational, contractual and regulatory impact, then use activity to understand technical exposure.
3. SQL conversion is only one part of the workload
Syntax translation can address many dialect differences, but it cannot certify the intended business result. Date and time semantics, numeric precision, null handling, string comparisons, case sensitivity, window behavior, stored logic, permissions and execution order may require explicit review. A translated statement that runs without error is not necessarily a validated replacement.
4. Hidden consumers are discovered after cutover
BI extracts, emailed reports, embedded analytics, notebooks and external integrations can depend on warehouse objects without appearing as named downstream assets in the migration tool. A cutover plan must reconcile the lineage graph with BI catalogs, scheduler definitions, code repositories, access logs and interviews with report owners.
5. The source environment keeps changing
During a long migration, new models are added and existing ones are modified. A one-time discovery snapshot becomes stale. Establish a change freeze or a controlled change-capture process for the relevant workloads. Re-run or refresh discovery at planned milestones and explicitly account for new dependencies.
How BigQuery migration lineage works in practice
According to Google Cloud’s current workflow, teams run the dwh-migration-dumper utility to extract source metadata and query logs, place the generated files in Cloud Storage, and start a lineage job through the console or the BigQuery Migration API. Once processing finishes, they can inspect connected tables and columns and the scripts associated with observed relationships. The job may take hours, depending on input size.
The service distinguishes nodes, which represent data objects, from edges, which represent relationships inferred from statements. An edge can arise because a source object influences a target through clauses such as FROM, WHERE or GROUP BY. Not every displayed relationship means that rows were physically copied between two tables. That distinction matters when engineers estimate data movement or identify transformations that must be rebuilt.
The interface can show upstream and downstream connections, users, scripts and column-level lineage where available. These are useful starting points for dependency analysis, but the quality of the output depends on the input metadata, log coverage and the SQL the service can interpret.
Google’s current documentation states that the service processes the first 5 GB of the oldest source logs. That limit makes input selection and completeness checks important. A successful job should not be interpreted as proof that every recent or infrequently executed workload has been observed.

The first deliverable should be a migration dependency register
Before allocating engineering sprints, turn discovery into a controlled register. A lineage graph is excellent for exploration; a register is better for assigning owners, testing and release decisions. Each migration unit should have an identifier, source objects, producing jobs, consumers, business owner, technical owner, refresh requirement, data sensitivity, target design, validation method and cutover status.
Include an evidence field. For example: observed in query logs, found in scheduler configuration, referenced by BI metadata, confirmed by a business owner or inferred but not yet validated. Distinguishing verified from assumed dependencies prevents false confidence.
A practical register also captures open questions: Which transformations use dynamic SQL? Which reports query through an intermediate view? Which jobs run only quarterly? Which data feeds leave the organization? Which datasets require historical backfills? An unresolved dependency should be visible as a risk, not buried in meeting notes.
A concrete way to score migration risk
Use a small, explainable scoring model rather than a complex formula that nobody trusts. Rate each workload from 1 to 5 on business criticality, dependency complexity, transformation difficulty, data sensitivity and recovery difficulty. Add a separate confidence rating for inventory completeness. A high-criticality workload with low confidence in its dependencies should receive early investigation even if the SQL itself looks simple.
For example, a daily sales dashboard with moderate SQL complexity but many downstream users may require extensive reconciliation and communication. A complex internal experiment with no production consumers may be technically difficult but safer to postpone. The scoring should guide sequencing, not mechanically dictate it.
Ask the business owner to approve the criticality rating. Ask the platform team to own technical complexity and recovery ratings. Where the two disagree, investigate before scheduling a cutover.
Group migration waves around business flows, not just schemas
A warehouse often organizes data by source or team, while a business process spans multiple schemas. Migrating the CRM schema first may not deliver a usable sales reporting capability if its invoices, targets and product dimensions remain elsewhere. A better wave might be a complete sales pipeline reporting flow, including its ingestion, transformations, metrics and consuming dashboards.
Look for natural dependency clusters: finance close, order-to-cash, marketing attribution, inventory availability, customer support or workforce reporting. Each wave should have a clearly bounded set of source objects, transformations and consumers, with an explicit acceptance owner.
Some highly shared reference objects may need to be established early in the target environment. Avoid creating a new shared bottleneck by making every wave depend on an untested central model. Build stable contracts, define ownership and validate common dimensions before expanding the number of consumers.
How to validate the lineage itself
Start by comparing the objects discovered in the lineage output with the source warehouse catalog. Investigate major differences in table and view counts. Then sample known critical workflows and confirm that expected upstream and downstream relationships appear. Test a rarely executed monthly or quarterly workflow deliberately, rather than assuming it will be represented in the logs.
Review dynamic SQL, stored procedures, external stages, orchestration jobs and application queries separately. Some may require code inspection or manual mapping. Confirm whether log retention covers a representative business cycle, including close periods, seasonal reporting and exceptional backfills.
For each priority output, ask an engineer to trace a metric from the consuming report back through transformations to the original source. The resulting record should identify the transformation code, scheduling dependencies and business definition. Actiknow’s article on data lineage for executives explains why this traceability matters beyond a one-time migration.

Migration lineage versus migration assessment
These tools answer related but different questions. A migration assessment can help characterize source workloads and estimate feasibility or sizing considerations. Migration lineage focuses on relationships and data flow within observed source activity. Neither independently proves semantic equivalence, user acceptance or operational readiness.
Google Cloud’s migration assessment guidance describes a separate assessment workflow using source metadata and log inputs, with output stored in BigQuery. Use assessment findings to understand workload characteristics, lineage to map dependencies, and controlled test runs to validate the target. Combine these with cost modeling and security review before a platform commitment.
A good discovery report therefore has three parts: what exists and runs; what depends on what; and what must be true for the business to accept the replacement. Only the third part turns technical findings into a decision.
What should be reconciled before a report moves?
Start with a baseline period agreed with the business owner. Compare row counts, distinct business keys, null rates, duplicates, date boundaries and financial totals between source and target. Then compare the actual metrics at useful dimensions: month, region, customer segment, product, channel and status.
Do not rely on one grand total. Two incorrect segments can cancel each other out. Use tolerance rules only where the business has approved them, such as documented rounding or currency-conversion differences. Keep exact-match expectations for identifiers, statuses and other deterministic fields.
Where historical backfills are required, validate both recent incremental loads and older periods. Include late-arriving changes, deletes, corrections and reprocessing. If a reconciliation fails, retain the old report as authoritative until the difference is understood and signed off.
Security and governance belong in discovery
A migration changes not only where data is stored but often who can access it. Source roles, masking rules, row-level policies, service accounts and data-sharing arrangements need a target mapping. Lineage may help identify sensitive data propagation, but it does not replace a field-level classification and access review.
For each wave, record the data classification, lawful retention requirements, permitted consumers and owner of access approval. Test permissions using representative identities, including service accounts and users who should be denied access. Ensure the target does not accidentally expose raw data through a broad analytical role.
Treat the extracted source metadata and query logs as potentially sensitive. SQL text can contain object names, business logic and, in some environments, literal values. Apply appropriate storage controls, retention limits and access restrictions to the discovery artifacts themselves.
The migration plan needs an operational rollback
A rollback plan is not a sentence saying that the team can switch back. It should specify which reports, endpoints and jobs will be redirected; who authorizes the decision; how writes or incremental states are handled; and what happens to data created after the target goes live.
For analytical migrations, a period of parallel operation is often practical. Run source and target workloads side by side, reconcile important outputs and monitor freshness, cost and failure behavior. Do not shut down the source solely because the first target refresh succeeded.
Define cutover gates: agreed reconciliation thresholds, performance at realistic concurrency, security tests, operational alerting, named support owners, tested recovery and business sign-off. Where the new platform changes metric definitions intentionally, document those changes as a business decision rather than calling them migration variance.

A 60-day discovery and planning sequence
Days 1-10: establish scope and evidence
Agree the business outcomes that justify the migration. Identify the critical reports and processes, source platforms, accountable owners, historical periods and non-negotiable SLAs. Collect metadata, query logs, orchestration definitions, BI inventories and known integration contracts. Define security controls for extracted artifacts.
Days 11-20: generate and challenge lineage
Run a representative lineage exercise for a supported source. Compare discovered objects with the source catalog, inspect high-value dependency paths and record coverage gaps. Validate a sample of known pipelines manually. Capture dependencies that automated lineage cannot see.
Days 21-30: build migration units and risk scores
Create the dependency register, classify criticality, map owners and group workloads into coherent waves. Identify shared dimensions and cross-wave dependencies. List required SQL rewrites, ingestion changes, security mappings and historical backfills. Estimate effort as a range, with assumptions visible.
Days 31-45: prove one end-to-end vertical slice
Choose a bounded business flow with representative complexity. Rebuild its ingestion, transformations and consumption path in BigQuery. Test correctness, access, freshness, performance, cost and recovery. Measure the effort to diagnose differences, not just the time to translate SQL.
Days 46-60: decide the migration roadmap
Use pilot findings to revise wave estimates and identify work that should be deferred or redesigned. Publish a cutover checklist, reconciliation plan, operational runbook, owner matrix and decision log. Obtain business acceptance for the first wave before scaling delivery.
These phases are an illustrative planning framework, not a promise that every organization can complete discovery in 60 days. Large estates, limited source logs, complex compliance requirements or scarce business-owner availability can extend the timeline.
How executives should measure discovery quality
Useful measures include the share of critical reports with a named owner; the share of priority workloads with verified upstream and downstream dependencies; the number of unresolved high-risk integrations; the proportion of key metrics with approved reconciliation rules; and the percentage of first-wave objects covered by automated tests.
Also track discovery confidence. A graph covering only the observed SQL log period may be useful, but it is not equivalent to a verified map of all production behavior. Explicit confidence levels make uncertainty manageable and keep project status reports credible.
Do not use the number of nodes in a lineage diagram as a success metric. The goal is fewer unexpected dependencies at cutover and faster, evidence-based decisions about what to migrate next.

Common mistakes that make lineage less useful
- The first mistake is treating lineage generation as a one-click audit of the entire company. It is bounded by the source inputs and supported parsing.
- The second is prioritizing by table count instead of business dependency.
- The third is assuming every edge represents a physical copy of data.
- The fourth is ignoring infrequently run jobs and manual exports.
- The fifth is separating technical migration plans from report-owner acceptance.
Another mistake is postponing cost and security analysis until the target is built. A workload can be technically portable yet economically unattractive or operationally difficult under the new execution model. Test representative queries, concurrency, refresh patterns and access controls before committing to the full move.
Where external support can be useful
Migration programs often need coordination across data modeling, source integrations, reporting, validation and application dependencies. Actiknow’s business intelligence consulting and integration work covers data integration from databases and APIs, solution architecture, modeling and dashboards. That combination is relevant when the challenge is not merely translating SQL but preserving trusted reporting across systems.
The right engagement starts with a scoped discovery: identify critical business outputs, inspect available evidence, validate a few representative dependency chains and estimate the work needed for an end-to-end pilot. It should produce a decision-ready inventory and risk assessment, not a premature commitment to rebuild everything.
Frequently asked questions
What is BigQuery migration lineage?
It is a Google Cloud service that visualizes relationships and data flows inferred from supported warehouse metadata and query logs to help plan a migration to BigQuery. It is a discovery aid, not a full application dependency scanner.
Is BigQuery migration lineage generally available?
Google Cloud’s documentation currently labels the service Preview. Verify its current status, supported regions and terms before relying on it in a delivery plan.
Which source warehouses are supported?
The current documentation lists Amazon Redshift SQL, Snowflake SQL, Teradata SQL and GoogleSQL. Support for a dialect does not mean every statement or external workflow will be interpreted perfectly.
Can migration lineage find every dashboard and API dependency?
No. It can reveal relationships visible in the supplied metadata and SQL activity. BI metadata, application code, exports, schedules and business interviews are still needed to establish the complete consumer inventory.
Does lineage automatically convert SQL into BigQuery syntax?
No. Lineage and SQL translation are distinct capabilities. Even when translation tools are used, teams must validate semantics, performance and the resulting business metrics.
How much source history should be analyzed?
Use a period that includes normal daily operations, month-end close and relevant seasonal or exceptional processes. Account for the service’s documented log-processing limit and explicitly inventory workflows not represented in the captured logs.
Should we migrate the largest tables first?
Not necessarily. Sequence work around business value, dependency clusters, readiness and risk. A small shared dimension can be more important than a large isolated fact table.
What proves that a migration wave is ready to cut over?
Business-approved reconciliations, tested security, realistic performance and cost checks, operational monitoring, a recovery plan, owner sign-off and an explicit decision about any intentional metric changes.
The decision to make before committing to a cutover date
A credible BigQuery migration starts with evidence about dependencies, not a target architecture slide. Use migration lineage to expose observed source relationships, then challenge that picture with application inventories, business ownership and reconciliation. The result should be a migration sequence that protects important decisions and gives the team a way to demonstrate correctness.
If your team is evaluating a warehouse migration and needs help scoping data flows, reporting dependencies and a validation plan, contact Actiknow to discuss a focused discovery engagement. The objective should be a defensible roadmap and a measurable pilot before a large-scale commitment.
References and further reading
- Google Cloud BigQuery migration lineage documentation: feature behavior, supported dialects, permissions and limitations.
- Google Cloud BigQuery migration assessment guidance: a complementary assessment workflow.
- Google Cloud BigQuery release notes, September 16, 2026: Preview introduction of migration lineage.

