Case study · Identity verification · Analytics
Delivered 20% cost savings and better usage analysis for MetaMap by migrating to AWS.
MetaMap · Customer identity platform
120TB scanned monthly across 30+ dashboards, with costs that spiked unpredictably. We moved the analytics platform onto Redshift.
20%
Reduction in costs
120TB
Scanned monthly, now on Redshift
80+
Analytics models migrated
The opportunity
An analytics platform whose bill nobody could predict.
MetaMap provides identity verification-onboarding, KYC, AI/ML capabilities-to 500+ customers across Latin America, Africa and Asia. Its analytics platform scanned roughly 120TB a month to power more than 30 dashboards.
Costs escalated unpredictably because workloads could not be managed or scaled independently of one another. Migrating was not trivial: 80+ analytics models, each 700-800 lines of SQL with subqueries, aggregations and window functions.
120TB scanned monthly
The volume behind 30+ dashboards drove the cost base.
Unpredictable, spiking costs
Spend moved in ways the team could not forecast or attribute.
Workloads could not scale independently
One platform meant no way to size or manage workloads separately.
80+ models of dense SQL
Each ran 700-800 lines with subqueries, aggregations and window functions.
The impact
The same analysis, on infrastructure that can be sized.
Redshift as the warehouse
Right-sizing was established by proof of concept before anything moved.
Incremental pipeline
Historical data to S3, then incremental loads via AWS Glue and MWAA into Redshift.
Applications on EKS
Dev and production application workloads moved to Amazon EKS alongside the analytics.
Validated, not assumed
Incremental migration was monitored in real-time, with post-migration validation before sign-off.
The stack
What it runs on.
- Warehouse
- Amazon Redshift
- Pipeline
- Amazon S3, AWS Glue, MWAA for Apache Airflow
- Applications
- Amazon EKS across dev and production
- Reporting
- Preset BI, repointed to Redshift
The road ahead
Costs are attributable now. That is what makes them manageable.
With workloads separable and usage visible, the next gains come from tuning the models themselves rather than absorbing whatever the platform happened to charge.
- 01
Model-level query optimisation
- 02
Per-workload cost attribution
- 03
Wider dashboard coverage
Talk to us
Bring one process. Leave with an outcome architecture.
30 minutes on one function, mapped to the architecture behind it. No slides.