Financial services · benchmark
Data platform modernization in financial services: published results
Banks, insurers and fintechs publish more data platform results than any other industry, and most of the claims are about money. We collected 38 public case studies from Databricks and Snowflake customers in financial services and pulled out every number they report, so you can see what a realistic outcome looks like before you build a business case.
- median cost reduction (19 results)
- 60%
- median speed-up where reported as a multiplier (8 results)
- 5x
- of case studies report a cost outcome
- 47%
- published case studies (27 Databricks and 11 Snowflake)
- 38
Analysis
What the results have in common
Our reading of the 38 published outcomes, and what it means if you're planning a program.
Retiring the legacy warehouse is where the cost comes out
The largest cost reductions in this set come from programs that switched off an on-premises warehouse or a duplicated ETL estate, not from tuning a cloud platform that was already running. If your plan keeps the old warehouse alive "for now", expect results at the bottom of the range.
Speed gains show up in batch windows and time to market
Financial institutions mostly report speed as shorter batch windows (overnight jobs finishing in minutes) and faster delivery of new use cases. Both matter for regulatory reporting deadlines and for getting fraud and credit models into production.
Business outcomes cluster around engagement and fraud
Where firms report revenue or risk results, they come from customer engagement, conversion and fraud detection, all of which depend on fresher data reaching models and apps. Few case studies attribute revenue directly to the platform, which is a useful caution for your own business case.
Cost
Largest reported cost reductions
Median 60% across 19 results. Each bar is one published result.
| Company | Reduction | What was reduced |
|---|---|---|
| PayU GPO | 94% | lower costs for selected data models |
| Korea Credit Data | 90%+ | Reduction in management costs through pipeline optimization strategies |
| NatWest | 80% | reduction in total cost of ownership and carbon footprint |
| AXA Japan | 70% | Lower ETL costs in migrated workloads |
| Coins.ph | 70% | Reduction in operating costs |
| First American Data & Analytics | 70% | Reduction in cost |
| Korea Credit Data | 70%+ | Reduction in additional cluster maintenance costs with Unity Catalog |
| AXA Japan | 60% | Reduction in warehouse costs for certain phases of the migration |
| BC Card | 60% | Cost savings on platform operation and utilization |
| YipitData | 60% | Lower database costs |
| Storebrand | 55% | Platform cost reduction |
| BC Card | 50% | Decrease in data error–related costs |
Speed
Biggest speed-ups
- 20xDiscovery BankFaster data processing times: from 9 hours to < 10 minutes
- 11xExperianFaster in fine-tuning time for Llama 8B model
- 10xABN AMROFaster time to market: use cases deployed in 2 months
- 5xAddeparFaster pipeline development, accelerating client onboarding
- 5xDiscovery BankFaster data product creation and implementation times
Business outcomes
Revenue and risk results
- 40%Discovery BankUplift in the impact of customer engagement activities
- 4.5xHSBCImprovement in engagement on the app
- 5%OakbrookRevenue increase forecast over the next 12 months from new top-up loans
- ~30%Sinyi RealtyIncrease in customer conversions
- 9%StorebrandIncrease in sales
- $38MXP Inc.In incremental revenue from performance media optimization
Method
How we built this benchmark
We collected every public financial services customer story from Databricks and Snowflake that reports at least two numbers, 38 in total, on 27 October 2026. Each headline number was recorded exactly as published and classified as a cost, speed, revenue, risk or scale outcome. Medians use only results that are directly comparable: percentages up to 100% and "N×" multipliers.
Limits: Every number here is self-reported by the customer and published by the platform vendor, so treat it as an upper-quartile view rather than an average outcome. Programs that went badly are not written up as case studies.
These are not Metric Oak engagements. See every source in the library, or read how we approach financial services.
Case studies in this benchmark: ABN AMRO, Addepar, AME Digital, AXA Japan, Banco Bradesco, BC Card, Block, Coastal Community Bank, Coinbase, Coins.ph, Discovery Bank, Experian, First American Data & Analytics, HSBC, IndusInd Bank, Korea Credit Data, Mercury, National Australian Bank (NAB), NatWest, Oakbrook, Parameta, PayU GPO, PicPay, Quantum Capital Group, RavenPack, Sallie Mae, Sinyi Realty, Storebrand, The RAC, The Zebra, Tipalti, TMBThanachart Bank, TS Imagine, Worldline, XP Inc., YipitData, YipitData Fivetran, Zurich
Frequently asked questions
What cost reduction do financial services companies report after modernizing their data platform?
Across 19 cost results published by financial services organizations in this set, the median reported reduction is 60%, ranging from 14% to 94%. 47% of the 38 case studies report a cost outcome at all.
How much faster do data workloads get after modernization in financial services?
Where results are reported as a multiplier, the median is 5x faster across 8 results. Where they're reported as a percentage, the median is 79% faster across 8 results. The largest gains are in batch processing and getting models into production.
Databricks or Snowflake: which gets better results in financial services?
This set includes 27 Databricks and 11 Snowflake case studies, with results across a similar range on both. Published case studies can't settle the question, because each vendor publishes its own best results. The right choice depends on your workloads, your team's skills and your existing contracts. The Data Platform Assessment ends with a written recommendation.
Are these numbers realistic for my organization?
Treat them as an upper-quartile view. Every number here is self-reported by the customer and published by the platform vendor, so treat it as an upper-quartile view rather than an average outcome. Programs that went badly are not written up as case studies. Your own baseline (current cost, batch windows and delivery times) is what a business case should be built on.
Where does this data come from?
From 38 public customer case studies published by Databricks and Snowflake, collected on 27 October 2026. We record each headline number exactly as published and link to every source in our case-study library.
How long does a data platform modernization take?
Published case studies report duration inconsistently, so we don't benchmark it here. As a guide, Metric Oak's Data Platform Assessment takes 3 weeks, and the first migrated domain typically reaches production 4–6 weeks after that.
Have 30 minutes? Get a senior architect’s view of your data platform.
No sales deck. You talk to the engineer who would lead the work.
Book an architecture review