The success story of data science at Exact. Empowering the accountant of tomorrow.
summary
Exact builds an automated bank-transaction→GL classifier trained on ~100M historical records, using LSTM with word embeddings and deployed as SageMaker endpoints orchestrated by Step Functions and EMR for ETL, with monitoring in CloudWatch / Splunk / Power BI. EMR memory tuning and instance-sizing reduced costs; DynamoDB-backed rollback and IAM-scoped roles ensure compliant deployments. Outcome: 1.3M booking suggestions/month, per-booking time cut from ~15–20s to under half, ~2,975 hours saved monthly.
AI-written summary · about
chapters (18)
- 0:00Intro
- 0:30About Exact
- 1:30What is Exact solving
- 3:10Exact automation workflow
- 4:15No Hands Accounting
- 6:52Data Platform
- 9:24AWS Services
- 11:05Clusters
- 13:15Approach
- 16:38Process
- 18:56Model overview
- 21:12SageMaker
- 22:29Cost performance optimization
- 24:48Challenges
- 26:19Step Function
- 32:13Recap
- 33:57Summary
- 34:48Takeaways
related talks
29mHow financial services companies derive greater value from data with AWS
29mBecoming a data-driven enterprise
29mData foundation for analytics and AI
52mPreparing data for reporting, ML, and AI on AWS
27mEnterprise AI at Scale
28mDataEngBytes 2023 - PER-T1-07 - Nadia Reyhani
27mEnterprise AI at Scale: Success Stories and Insights
28mTurn Data Into Insights