For years, enterprises have been told that cloud modernization is a technology decision: choose the right platform, migrate the right workloads, and unlock scale, performance, and innovation.
But any CIO or CDO who has led a major data platform migration knows the harder truth. The technology is rarely the constraint. The bottleneck usually manifests at the level of delivery.
Before a legacy data warehouse can move to the cloud, teams must understand what they have: the sources, targets, dependencies, transformation logic, business rules, and undocumented decisions embedded across years of code.
They must decode procedures written by people who may no longer be with the organization, determine where data comes from and where it lands, and turn institutional memory into executable migration plans. Historically, that work has been manual, specialized, and slow. An AI-first mindset, however, changes the conversation.
Breaking Through the Delivery Bottleneck
An AI-first approach to projects and delivery means looking at the work itself and asking: which parts of this engagement are repetitive, knowledge-heavy, documentation-intensive, and constrained by human bandwidth?
Next, it means designing AI-enabled skills to perform those activities consistently, in the context of the client’s own environment, while leaving expert consultants focused on architecture, quality, risk, and business outcomes.
In cloud data migrations, that shift is already changing what is possible.
What This Looks Like in Practice
IBM recently applied this approach with one of the world’s largest internet services companies: a Fortune 500 digital platform serving tens of millions of small businesses globally and operating one of the most complex, high-volume e-commerce data environments in the industry.
The client’s legacy Teradata environment reflected years of growth and accumulated engineering decisions. Critical ETL logic was embedded in multi-step stored procedures. Individual processes drew from more than 20 source tables, applied multi-currency conversion across international markets, handled multi-company business rules, and implemented slowly changing dimension patterns. Much of that logic was undocumented, and some had been written by people no longer at the company.
In a traditional migration, the first phase would have required engineers to read procedures line by line, reconstruct business logic, manually map source-to-target relationships, and create lineage documentation before real modernization could begin. Discovery alone could take days per complex procedure; end-to-end planning could stretch across weeks.
Purpose-Built AI Skills, Developed with Snowflake Cortex
Using IBM-built AI skills developed with Snowflake Cortex, the team compressed manual discovery from days to hours and reduced the overall timeline from weeks to days. The skills did not simply summarize code, but rather operationalized key parts of the migration workflow.
One skill parsed Teradata and Snowflake procedures to identify source and target tables, procedure blocks, join conditions, filters, transformation logic, execution order, and call chains. It generated visual lineage diagrams in draw.io format with two views: full source-to-target data flow, and a transformation-detail view of how columns were derived. For Snowflake, the diagrams reflected the client’s medallion architecture:RAW, CLEAN, and CONFORMED layers, cross-layer procedure calls, lookup dependencies, and business-ready targets.
A second skill generated structured business-rule documentation and transformation workbooks. Rather than asking engineers to manually reconstruct every CASE statement, COALESCE default, hardcoded value, aggregation, filter, join, date rule, and FX conversion, it extracted that logic into reviewable Excel artifacts while separating parameter rules, source routing, step-level rules, delete and idempotency logic, key transformations, error handling, and post-processing. Where currency conversion existed, the skills produced FX matrices showing how currency-specific rates were applied.
A third skill supported migration from Teradata SQL to Snowflake by identifying platform-specific constructs and applying Snowflake-native patterns. To give an example: ACTIVITY_COUNT becoming SQLROWCOUNT, SYS_CALENDAR logic becoming DAYOFWEEK(), DELETE ALL becoming TRUNCATE TABLE, and COLLECT STATISTICS being removed because Snowflake manages optimization differently. It also organized the implementation into a RAW/CLEAN/CONFORMED architecture with thin RAW staging, CLEAN-layer transformation logic, and CONFORMED-layer orchestration and final loads.
From Manual Inspection to a Shared Evidence Base
The impact on delivery was significant. The team moved from manually inspecting legacy code to generating a consistent set of migration artifacts: lineage documentation, draw.io diagrams, transformation workbooks, object inventories, discrepancy reports, validation outputs, and Snowflake SQL modules. That gave architects, engineers, and business stakeholders a shared evidence base for decisions that otherwise depended on individual interpretation.
The important point is not simply that AI made the work faster. It is that the skills were designed for the client’s environment: their database names, schema conventions, design preferences, migration patterns, and operating model. The outputs looked familiar because they were tailored delivery assets, not generic templates.
Just as importantly, the engagement created a lasting capability. The AI skills could be shared with the client to help their teams understand their own data landscape, extend documentation, and continue applying the same discovery and modernization patterns beyond the initial effort.
That is the difference between using AI as a productivity tool and using AI as a delivery model.
The Migration Equation Has Changed
For many technology leaders, the business case for cloud modernization has never been the issue. The challenge has been execution risk.
Large-scale migrations often stall because teams cannot see the full dependency landscape quickly enough, confidently prioritize what to move first, or validate whether critical business rules have been captured. Because documentation is often incomplete or outdated, migration becomes a discovery exercise, a forensic investigation, and a transformation program all at once.
Traditional accelerators have helped but not fully solved this. Many are difficult to stand up, require extensive configuration, or impose assumptions that don’t match the client’s environment. AI changes the equation because it can operate directly on the messy reality of enterprise environments: stored procedures, SQL scripts, naming conventions, undocumented logic, and inconsistent patterns accumulated over time.
When designed well, AI skills parse those artifacts, extract structure, generate documentation, flag discrepancies, and accelerate the translation from legacy implementation to modern architecture.
Guardrails That Make Outputs Trustworthy
Trust matters, so the skills included guardrails to reduce hallucination risk: object references had to be traceable to source SQL, call chains had to match explicit CALL statements, and column counts in documentation had to align to the corresponding DDL and INSERT logic.
This does not eliminate the need for experienced migration teams, but it changes where their time is spent. Instead of manually reconstructing what a system does, teams focus on validating what should carry forward, identifying opportunities to simplify, and aligning the target architecture to business priorities. Consultants spend less time producing baseline documentation and more time advising on modernization strategy, governance, performance, and operating model design.
For CIOs and CDOs, that matters because migration success is not measured by code conversion alone. It is measured by whether the organization emerges with a data platform that is faster, more trusted, easier to govern, and better aligned to business outcomes.
Why AI Skills Must Be Built for Your Environment
The next era of AI-enabled consulting will not be defined by generic prompts or one-size-fits-all accelerators. It will be defined by skills: reusable, purpose-built AI capabilities that encode delivery knowledge and apply it within the specific context of a client’s environment.
Beyond Syntax: Patterns and Preferences
A useful skill must understand more than syntax. It must understand the migration patterns that matter, like how to trace column-level transformations, recognize embedded business rules, distinguish orchestration from transformation logic, and map legacy structures into the target architecture.
It must also reflect the client’s preferences: how schemas are named, how steps are numbered, where reusable logic lives, what documentation format architects and data stewards will actually use, and which validation checks are required before outputs can be trusted.
These details are the difference between an AI output that is interesting and one that is usable.
Quality, Built In
They also reinforce quality. The migration skills compared expected logic against the Snowflake implementation and classified findings as pass, warning, or fail. Discrepancy analysis caught missing filters and aggregations, wrong hardcoded values, missing CASE transformations, wrong join keys, and delete-scope mismatches (i.e., the kinds of issues that can materially affect business reporting if not caught before production).
That is why IBM’s approach combines platform capability with consulting expertise. Snowflake Cortex provides the foundation for AI-enabled analysis and generation. IBM Consulting brings the migration methodology, data engineering experience, industry context, and delivery discipline required to turn that foundation into client-ready capability while producing faster discovery, more complete lineage, better documentation, accelerated code modernization, and a clearer handoff to client teams.
From AI-First Strategy to AI-First Delivery
Many organizations are now defining AI-first strategies. The next question is whether their transformation programs are being delivered in an AI-first way.
For cloud modernization, that means rethinking the mechanics of delivery: using AI not only to generate ideas or summarize meetings, but to automate the manual work that has historically slowed enterprise change. It means creating client-specific skills that make complex systems understandable and easier to modernize, and leaving clients with new capabilities they can continue to apply after the engagement ends.
The opportunity is not simply to migrate faster. It is to modernize better, with more transparency, more confidence, and a stronger connection between technology execution and business value.
Ready to bring AI-first projects and delivery to your cloud modernization journey? IBM Consulting helps global enterprises modernize complex data environments by combining deep Teradata and Snowflake expertise with AI-powered delivery capabilities built for each client’s architecture, standards, and business outcomes.
Reach out today to explore how IBM can help accelerate your migration to Snowflake and build AI skills that continue creating value long after the initial project is complete.
