How a global cruise line modernized on Snowflake
Six to eight hours to close out daily incentive calculations, every single day, for a $9B+ global cruise line running three separate consumer brands on a decade-old Oracle system nobody could safely change. A legacy system modernization on Snowflake cut that time in half and unified all three brands onto one platform.
At a glance
| Industry | Travel & hospitality (cruise) |
| Scale | $9B+ revenue, three consumer brands |
| Challenge | A legacy Oracle and MS Access system calculated employee incentive payouts across three siloed brands, with daily batch runs taking six to eight hours. |
| Technology | Snowflake, CoCo, dbt, Streamlit, Claude Code (replacing Oracle and MS Access) |
| Timeline | Approximately 6 to 7 month rebuild and migration |
| Result | Batch processing time cut by 50%+, reconciled within less than 1% of legacy output |
| Phdata Service | Data engineering, Snowflake, Anthropic |
The problem
A $9B+ global cruise line’s employee incentive program, which calculates commission and bonus payouts across three distinct consumer brands, ran on a legacy Oracle and MS Access system built up over a decade of brand-specific rules that nobody could fully explain anymore. Each brand kept its own bookings and call logs in separate systems, and the whole program could only run once a day, so every calculation took hours of manual work to get right. Continuing to run the system meant accepting more risk over time, with no way to grow, see into the process, or run it faster.
What phData did
Five decisions shaped the delivery.
Rewrote the calculation logic in dbt instead of migrating it as-is
Decomposed a decade of undocumented Oracle business rules into modular, testable dbt models rather than porting the legacy code directly into Snowflake. In doing so, the team preserved the existing business outputs while changing the underlying architecture, making the incentive calculation logic auditable and maintainable for the first time.
Built automated validation checks directly into the pipeline
Rather than relying on manual QA after each run, phData embedded validation checks into the dbt pipeline so every run immediately confirmed output quality against the legacy system. This caught discrepancies early and gave the business confidence to cut over.
Delivered a Streamlit application in place of the legacy Access tool
Business users needed a way to manually adjust incentive data when source systems could not support edge cases, such as split commissions. phData built a Streamlit application so the business could manage exceptions directly instead of relying on ad hoc Access workarounds.
Consolidated three brands onto one Snowflake-based platform
Instead of maintaining three siloed calculation systems, phData unified all three brands’ incentive calculations into a single Snowflake-based model, aligning the platform with the client’s broader tri-brand strategy.
Paired the platform rebuild with hands-on Claude Code enablement for engineers
Alongside the Snowflake migration, phData ran a Claude Code hands-on lab with the client’s engineering team, covering codebase exploration, debugging, and reusable Claude Skills. The client requested follow-on sessions immediately after, extending the platform work into a broader push toward agentic development practices.
Streamlit is an open-source framework for building interactive, browser-based applications directly from Python code, without requiring a dedicated front-end development team.
Claude Code is Anthropic’s agentic coding tool that helps engineering teams explore codebases, fix bugs, and build features through natural-language prompts.
dbt (data build tool) is a framework that lets data teams transform and test data directly inside a warehouse like Snowflake using modular, version-controlled code.
The results
The rebuilt system cut daily batch processing time by more than half, while reconciling within less than 1% of legacy output during validation, as automated validation checks replaced manual reconciliation after every run. That gave the finance and operations teams confidence in the output without needing to audit it line by line, and is what allowed the business to retire the legacy Oracle and MS Access system entirely.
Run daily incentive calculations in half the time, removing an overnight processing bottleneck
Manage commission exceptions, like split payouts, directly through Streamlit instead of routing around system limitations
Trust automated validation checks that reconcile every run within less than 1% of legacy output
Operate one unified incentive model across three brands instead of three disconnected systems
Reduction in daily batch processing time, from 6-8 hours down to 3-4 hours, after legacy system modernization.
Deviation from legacy system output, confirmed during validation before cutover.
What the client said
The phData team brought a great solution to the table, and the business is really pleased with the simplicity and ease of the new Streamlit app.
Ready to modernize a legacy system your business depends on?
Why this matters beyond this project
Legacy calculation systems are common, especially in finance-adjacent functions like commissions, incentives, and payouts. Fixing one means rebuilding the underlying business logic into a testable, auditable set of rules that keeps pace with daily operations.Â
In this project a decade of undocumented, brand-specific rules got decomposed and consolidated onto one platform without disrupting the people who depend on it. Platform modernization and AI enablement ran side by side, and the client asked for follow-on Claude Code sessions immediately after the rebuild wrapped.Â
Any multi-brand organization running critical calculations on aging, siloed systems while building its team’s AI capabilities is looking at the same opportunity.
Related customer stories
How a fast-casual restaurant chain built an AI-powered sales assistant in eight weeks
phData built an AI-powered sales assistant on Snowflake Cortex for a fast-casual chain, translating 200+ BI measures in 8 weeks. Here's how.
Frequently asked questions
What does legacy system modernization typically involve?
Legacy system modernization usually means rewriting undocumented business logic into a modern, testable framework rather than simply migrating a database to new infrastructure. In this engagement, that meant decomposing a decade of Oracle-based commission rules into modular dbt models on Snowflake.
How long does a legacy incentive system modernization take?
This engagement took approximately six to seven months from rewrite to production validation. Timelines vary based on how much undocumented business logic needs to be reverse-engineered and how many source systems are involved.
Can multiple business units be consolidated onto one data platform?
Yes, three previously siloed brands were unified onto a single Snowflake-based platform without disrupting end users. Each brand’s distinct business rules were preserved as modular components within the same system.
What technologies are involved in this kind of legacy modernization?
This engagement used Snowflake as the target data warehouse, dbt to rebuild the business logic, and Streamlit to replace legacy admin tooling. The prior system ran on Oracle and MS Access.
How does phData ensure accuracy during a legacy system migration?
phData builds automated validation checks directly into the data pipeline so every run is reconciled against the legacy system’s output. In this engagement, that validation confirmed results within less than 1% of legacy output before cutover.
Did this engagement include Claude Code enablement?
Yes, alongside the Snowflake-based data platform rebuild, phData also ran a Claude Code hands-on lab for the client’s engineering team. The session covered codebase exploration, debugging, and reusable Claude Skills, and prompted the client to request follow-on training immediately afterward.
Ready to modernize a legacy system your business depends on?
Tell us about your challenge. A phData expert will follow up to discuss how to accelerate AI in your organization.
- Free scoping conversation — no commitment
- Certified data and AI engineers
- Proven delivery