Capability / Migration assessment
Fixed-Scope Data Warehouse Migration Assessment
A data warehouse migration assessment is a fixed-scope engagement that establishes what a legacy warehouse actually does before anyone commits to moving it. Quorlytic inventories objects and consumers from query logs and lineage, scores translation complexity, defines reconciliation and acceptance rules, and delivers a costed, phased migration plan with the evidence behind every estimate.
By QuorlyticUpdated
Most migration estimates are estimates of the easy half. The assessment exists to find the stored procedures nobody owns, the extracts nobody documented and the reports that will break, while changing course is still cheap.
Evidence
- Query logsWho reads what
- SchedulesWhat runs when
- CatalogueWhat logic does
- LineageWhere data flows
Disposition
- MigrateMove as it is
- RedesignRebuild on target
- RetireNo reader, on evidence
Complexity score per workload
- SimpleSQL and views
- Moderatedialect and functions
- Complexprocedures and state
- Costed, phased migration planWaves · cost per phase · rollback position
Agreed before anything moves — what must match, at what tolerance, signed off by whom
The business problem
A migration estimated before its dependency set is known is a guess about the visible part of the estate. Query logs routinely reveal consumers the platform team has never met, and procedural logic routinely turns out to hold business rules nobody wrote down. The assessment converts those unknowns into a scoped, costed plan before budget is committed to the move.
What we build
The inventory is built from query history, job schedules, lineage and the warehouse catalogue rather than from interviews alone. Each object is classified as migrate, redesign or retire, and retirement is recommended only on evidence that nothing reads it.
The outcome
A costed, phased migration plan with the evidence behind each estimate
Assessment register
Every workload classified, scored and owned before the plan is written.
The assessment's working surface is a register: one row per workload, with the disposition the evidence supports, the complexity that drives its effort and the person who signs it off. The plan is built from this register, so every estimate can be traced back to a row.
- Legacy warehouse estate
- Disposition review
- Owners named
| Workload | State | Disposition | Complexity driver | Owner |
|---|---|---|---|---|
| Finance reporting marts | Scored | Migrate | Procedural logic in month-end jobs | Finance data owner |
| Customer extracts | Owner to confirm | Redesign | Consumers outside the platform team | CRM lead |
| Legacy sales history | Evidence attached | Retire | No reader in the measured period | Sales operations |
| ETL mappings | Being scored | Migrate | Workflow state and reject handling | Integration lead |
| Access roles | Decision needed | Migrate | Row-level policies per region | Security owner |
What it answers
- Which objects does anyone actually read?
- What makes this workload hard to move?
- Who signs off that the new numbers are right?
When this is the right engagement
Symptoms that bring teams here.
These are the situations in which a migration assessment pays for itself. If none of them describe your estate, a direct migration plan may be enough.
The migration estimate came from an object count
Tables are the easy half. The effort sits in procedural logic, extracts and consumers nobody listed.
A licence or support end date is driving the timeline
The deadline is fixed, so the plan has to be built from evidence rather than optimism.
Nobody can say who reads the warehouse
Without the consumer set from query logs, every cutover risks breaking a report someone depends on.
Stored procedures carry business rules nobody owns
Undocumented logic is the largest single source of migration risk and the hardest to estimate.
The board wants a number before the project starts
A defensible budget needs a complexity score per workload, not a figure from a vendor demonstration.
Two vendors gave estimates that differ wildly
Each estimated a different scope. An assessment fixes the scope so estimates become comparable.
Capability model
What the assessment establishes.
The assessment answers five questions before any budget is committed to the move. Select one to see what it involves and what it produces.
01 — Inventory
Establish what the warehouse actually contains and who actually uses it, from query history rather than from documentation.
What it produces
An object and consumer inventory with a named owner per critical consumer.
What usually goes wrong
- Extracts and applications reading tables directly
- Reports built outside the platform team
- Objects nobody has read in a year still counted as scope
Technologies in this stage
- Query log analysis
- Job schedules
- Catalogue metadata
- Lineage
02 — Classify
Decide, object by object, what is migrated as it is, what is redesigned and what is retired on evidence.
What it produces
A migrate, redesign or retire disposition for every object, with the evidence behind it.
What usually goes wrong
- Retiring objects on opinion rather than read history
- Redesigning everything because a move is happening anyway
- Disposition decisions with no owner
Technologies in this stage
- Usage evidence
- Disposition register
- Owner sign-off
03 — Score
Score translation complexity per workload against the source platform's known problem areas, so the estimate reflects the estate you actually run.
What it produces
A complexity score per workload, with the behaviours that drive it named.
What usually goes wrong
- Procedural logic with no direct equivalent on the target
- Dialect differences in dates, nulls and precision
- Physical design assumptions that do not carry over
Technologies in this stage
- Teradata
- Oracle
- Netezza
- SQL Server
- Informatica
04 — Accept
Write down what must match between old and new, at what tolerance, and who signs off, before anything moves.
What it produces
Reconciliation and acceptance rules agreed with the data owners.
What usually goes wrong
- Totals compared but distributions ignored
- Rounding and time-zone differences dismissed as noise
- Access rules discovered only at cutover
Technologies in this stage
- Reconciliation rules
- Metric definitions
- Access inventory
05 — Plan
Turn the evidence into waves ordered by dependency and risk, with the effort, cost and rollback position of each phase stated.
What it produces
A costed, phased migration plan with the evidence behind every estimate.
What usually goes wrong
- Waves ordered by convenience rather than dependency
- Costs stated without their assumptions
- Open decisions left without an owner
Technologies in this stage
- Wave plan
- Cost model
- Risk register
- Rollback approach
Reference architecture
How evidence becomes a plan
The assessment narrows a large, uncertain estate into a small number of decisions. Four kinds of evidence feed one disposition register, which feeds the scored, phased plan.
Query history
Who reads what, and how often.
- Consumers
- Read frequency
- Unused objects
Job schedules
What runs, when, and in what order.
- Dependencies
- Run windows
- Failure history
Catalogue and code
What the logic actually does.
- Procedures
- Views
- ETL mappings
Lineage and access
Where data flows and who may see it.
- Lineage
- Roles
- Row policies
Disposition and score
Every object classified and every workload scored.
- Migrate
- Redesign
- Retire
- Complexity
Costed, phased plan
Waves, effort, cost and rollback per phase, with acceptance rules attached.
- Waves
- Cost per phase
- Acceptance rules
- Open decisions
Agreed before anything moves
- Acceptance thresholds
- Data owners named
- Target recommendation
- Risk register
The approach
- 01
Why an assessment comes before the migration
A migration estimated before its dependency set is known is a guess about the visible part of the estate. Query logs routinely reveal consumers the platform team has never met, and procedural logic routinely turns out to hold business rules nobody wrote down. The assessment converts those unknowns into a scoped, costed plan before budget is committed to the move.
The sequencing logic is set out in the warehouse migration playbook, and the delivery itself is described on the Snowflake migration services page.
- 02
Discovery from evidence, not from documentation
The inventory is built from query history, job schedules, lineage and the warehouse catalogue rather than from interviews alone. Each object is classified as migrate, redesign or retire, and retirement is recommended only on evidence that nothing reads it.
- Consumers identified from query logs, including unknown ones
- Objects classified as migrate, redesign or retire
- Retirement backed by read history, not opinion
- 03
Source-specific complexity is scored, not assumed
Teradata, Oracle, Netezza, SQL Server and Informatica estates each fail in different places: procedural logic, proprietary functions, physical design assumptions, identity and sequence behaviour, and orchestration. The assessment scores each workload against the source's known translation problems so the estimate reflects the estate you actually run.
Source-specific detail is published for Teradata, Oracle, Netezza, Informatica PowerCenter to dbt and Cloudera to Databricks.
- 04
Acceptance is defined before anything moves
The assessment writes down what has to match between the old and new platforms, at what tolerance, and who signs off, so the later cutover decision is judged against an agreed standard rather than debated under time pressure. Access roles, row-level policies and service identities are inventoried at the same time, because they are the part most often discovered late.
- 05
The deliverable is a phased, costed migration plan
The output is a phased migration plan: waves ordered by dependency and risk, the effort and cost behind each phase, the reconciliation approach, the rollback position and the open decisions that still need an owner.
- Waves ordered by dependency and business risk
- Cost and effort stated per phase, with assumptions
- Open decisions listed with a named owner
- 06
Who the assessment is for, and who it is not for
It suits organisations with a legacy warehouse or ETL estate they intend to move, where the dependency set, the procedural logic or the reconciliation burden is not yet understood. It is not the right first step when the estate is small and well documented, or when the target platform has not yet been chosen; that choice belongs in a data and AI advisory engagement first.
Delivery model
How the work is sequenced.
Four steps, each ending in an artefact you keep, not a status update.
| Phase | Activities | Decisions made here | What you keep |
|---|---|---|---|
| InventoryFind every object and every consumer. |
|
| Object and consumer inventory |
| Score complexityMeasure how hard each workload is to move. |
|
| Disposition register and complexity scores |
| Define acceptanceDecide what proves the new platform is right. |
|
| Reconciliation and acceptance rules |
| Plan the wavesTurn evidence into an executable sequence. |
|
| Costed, phased migration plan |
Technology fit
What each technology is here to do.
What the assessment reads, by role. The source platform decides most of the effort, so the assessment is always specific to it.
Source warehouses
- Teradata
- Oracle
- Netezza
- SQL Server
- Redshift
Procedural logic, proprietary functions and physical design assumptions drive the translation effort.
Source ETL
- Informatica PowerCenter
- Stored procedures
- Scheduled scripts
Orchestration and state handling are often the hidden half of a warehouse estate.
Platform detailTypical targets
- Snowflake
- Databricks
- BigQuery
- dbt
The recommendation follows the workload, the team's skills and the cost model, and states the options set aside.
Platform detailEvidence
- Query logs
- Lineage
- Catalogue metadata
- Access logs
Scope decided from what is actually used rather than from what people remember.
Technology marks belong to their respective owners. No partnership, certification or endorsement is implied.
Engineering concerns
What has to be answered before it is trusted.
What the assessment has to settle before a migration budget is defensible.
- Scope from usage
- Scope is set from query history, so objects nobody reads are retired on evidence rather than migrated by default.
- Undocumented logic
- Every stored procedure and ETL mapping in scope has an owner or a recorded decision about it.
- Comparable estimates
- Effort and cost are stated per phase with their assumptions, so any delivery estimate can be compared against the same scope.
- Acceptance in advance
- What must match, and at what tolerance, is agreed before anything moves, not argued at cutover.
- Access parity
- Roles, row-level policies and service identities are inventoried early, because they are the part most often found late.
- Reversibility
- Each wave has a stated rollback position, so the programme stays reversible until it deliberately is not.
What the assessment includes, and what it does not
| Dimension | Included | Not included |
|---|---|---|
| Inventory | Objects, jobs, stored procedures, extracts and downstream consumers, built from query logs and lineage | Migrating any object or data |
| Complexity | A translation-complexity score per workload, with the source-specific behaviour that drives it | Rewriting or converting code |
| Target | A target architecture recommendation, with the options considered and why each was set aside | Buying or provisioning the target platform |
| Reconciliation | The counts, aggregates, samples and business metrics that must match, and the thresholds that block a cutover | Running the parallel period itself |
| Plan | A phased migration plan with waves, dependencies, risks, a rollback approach and a cost estimate per phase | Executing the migration waves |
Where it applies
Related work, and why it is related.
Where the assessment leads, and the guides behind it.
- Snowflake migration servicesThe delivery that follows an assessment when Snowflake is the recommended target.
- Data modernizationThe parent service: platform moves, phased cutover and a tested rollback.
- Teradata to SnowflakeBTEQ, macros, stored procedures and the physical design assumptions that do not carry over.
- Informatica PowerCenter to dbtSeparating transformation logic from orchestration and state before anything moves.
- Oracle to SnowflakePL/SQL, sequences and change capture: what the assessment scores for an Oracle estate.
- Data and AI advisoryWhen the target platform has not been chosen yet, that decision comes first.
Questions
Questions worth asking.
What is a data warehouse migration assessment?
A fixed-scope engagement that maps what a legacy warehouse does, who depends on it and how hard each workload is to translate, then turns that into a costed, phased migration plan with reconciliation and acceptance rules. It happens before the migration is committed to, so the budget is set against evidence.
What does the assessment produce?
An object and consumer inventory built from query logs and lineage, a translation-complexity score per workload, a target architecture recommendation, reconciliation and acceptance rules, and a phased migration plan with effort, cost, risks and a rollback approach per phase.
Which source platforms do you assess?
Teradata, Oracle, Netezza, SQL Server and Redshift warehouses, Informatica PowerCenter ETL estates, and Cloudera or Hadoop platforms, with Snowflake, Databricks, BigQuery or dbt as typical targets. The source platform decides most of the effort, so the assessment is always specific to it.
How is the assessment priced?
As a fixed scope agreed in writing before it starts, so the cost does not grow with the number of surprises found. The fee depends mainly on the number of source systems and the volume of procedural and ETL logic in scope; tell us the platform and the rough size of the estate and we will quote it.
What do you need from us to start?
Read access to query history, job schedules and the warehouse catalogue, a list of the main source systems and consumers you know about, and a business owner for each critical reporting area. Production data itself is rarely needed for the assessment.
Next step
Scope your migration assessment.
Tell us the source platform, the target you are considering and roughly how many jobs and reports depend on it. That is enough to scope the assessment.