Case Study
Predictive Spend
87% forecast precision on consumer transactions.
Client
Consumer FinTech
Market
Consumer financial services
Scale
Consumer transaction data at mobile scale
Duration
6 months
Team
3 engineers
Engagement model
Managed Delivery
Delivery hub
Belgrade
Stack
Python, Flutter, custom neural networks, NLP
Capability
AI Engineering
Challenges
Challenges
Opaque spending
Users could not see where money was going until it had gone.
Messy transaction data
Raw transaction descriptions were inconsistent and hard to categorise.
Mobile-first constraint
Forecasts had to run fast enough for a responsive mobile experience.
The solution
Transaction understanding
NLP that categorises messy transaction descriptions reliably.
Spend forecasting
Custom neural networks predicting near-term consumer spending.
Mobile delivery
A Flutter app surfacing forecasts in real time on the device.
The honest section
What we would tell you before starting
Forecast precision depends on transaction history. New users with little history received wider prediction bands. We surfaced that uncertainty in the product rather than presenting a confident number the data could not support.