Client activity prediction for B2B Forex
A financial services company providing foreign exchange services to B2B clients needed to transform years of historical trading data into accurate client activity forecasts for proactive account management and revenue planning. We developed an AI-powered forecasting platform that predicts trading volume, transaction frequency, and profitability, helping account managers identify declining activity early, retain trading relationships, and improve planning.
Industry
FinTech
Client
Under NDA
Region
United Kingdom
Main challenges
Business Challenge: Reactive client management limited revenue retention
The company had years of trading data but no systematic way to predict which clients were likely to reduce activity.
- Client slowdowns became visible only after they happened
- Managers could not prioritize retention outreach effectively
Technical Challenge: Scaling individual forecasts across disparate trading data
The system had to scale forecasting across multiple metrics, currencies, operation types, and timeframes while automatically selecting the optimal model for every prediction.
- Thousands of forecasting jobs demanded parallel processing
- Different client patterns called for specific forecasting models
- Every prediction required automatic model selection
Delivery Challenge: Deploying reliable ML without production system risk
The team had to deliver complex ML functionality quickly while isolating forecasting workloads from the client's live infrastructure.
- Aggressive delivery timelines demanded phased implementation
- Production safety required fully isolated data access
- Business adoption depended on reliable forecast delivery
What we did
We built an AI-powered forecasting platform that transforms years of historical trading data into future activity predictions for B2B clients. The solution forecasts trading volume, transaction frequency, and profitability by combining CRM records, transaction history, market data, and currency exchange rates within a unified forecasting pipeline accessible through an intuitive management interface.
Instead of relying on a single forecasting approach, the platform automatically selects the best-performing model for every client, business metric, and prediction horizon. This adaptive strategy enables reliable forecasting across a diverse client portfolio while giving account managers consistent insights they can use to support proactive engagement and long-term client retention.
B2B clients covered with individual activity forecasts
Node Auto-scaling cluster supporting large-scale parallel forecasting
Forecasting models evaluated automatically for every prediction
Integrated data sources powering a unified forecasting pipeline
Per-client behavior forecasting
Individual forecasts reveal each client's trading behavior instead of relying on aggregate insights.
- Predicts future trading volume for individual clients across multiple forecasting horizons
- Forecasts transaction frequency for every client
- Estimates expected profitability from historical trading patterns
Automated best-model selection
The platform executes the most accurate forecasting approach for each specific prediction task.
- Evaluates Prophet, ARIMA, XGBoost, and LSTM models
- Automatically selects the highest-accuracy model
- Retrains models per client, metric, and time horizon
Self-service forecast configuration
The administration panel gives business users full control over forecasting without relying on engineering teams.
- Enables business users to configure forecasting criteria
- Configures target metrics and forecast periods
- Schedules recurring or on-demand forecast runs
- Generates forecasts automatically without engineering involvement
- Delivers results through the platform interface
Proactive outreach dashboards
Account managers use forecasts to identify declining activity and prioritize outreach before trading volume drops.
- Surfaces at-risk accounts before activity declines
- Highlights clients expected to reduce trading activity
- Supports earlier retention and outreach management
- Provides client-level forecasting with actionable insights
- Displays expected future trade count and revenue
Forecast reconciliation reporting
Reconciliation reporting compares predictions with actual trading results and monitors forecast accuracy over time.
- Reviews predictions against completed trading activity
- Tracks forecast accuracy across clients and time periods
- Identifies recurring prediction errors and model drift
- Supports transparent and auditable forecasting
- Provides detailed client-level forecast reports
Key results and business value
Reactive monitoring replaced by predictive engagement
Fragmented data consolidated into a unified forecasting pipeline
Manual client tracking transformed into automated predictions
Siloed workflows enhanced with AI-driven forecasting
Features Delivered
Key capabilities of the AI-powered forecasting platform:
- Client activity forecasting across key performance metrics
- Self-service forecast setup for non-technical users
- Automated model selection for each forecast
- Nightly, weekly, and on-demand forecast scheduling
- Interactive dashboards for client-level forecasts
- Forecast reconciliation for accuracy tracking
Azure
React
Python
PostgreSQL
Docker
GitHub
Technical Highlights
Forecasting at scale
Auto-scaling cluster runs forecasts across ~3,500 clients in a few hours, monitoring thousands of parallel jobs.
Flexible forecasting
Enables business users to configure forecasting with automated report delivery through the admin panel.
Model selection
Candidate models are evaluated during training, while production applies the best model for each client.
Multi-model engine
Prophet, ARIMA, XGBoost and LSTM work as a unified pool with automated selection based on MAPE scores.
Data reconciliation
Results are continuously compared, providing accuracy validation and supporting model improvement.
Data isolation
Forecasting workloads access data via a dedicated database, keeping the production environment untouched.
CI/CD delivery
Container-based deployments ensure reliable, consistent, and repeatable software releases.
Client Feedback
"One of our account managers came back after the first forecasting cycle and said, 'We would have missed these clients completely before.' That was the moment we realized the platform wasn't just producing predictions — it was helping us keep valuable trading relationships active and protect revenue"
Head of Sales
Client
Ready to integrate AI into your financial operation?
Let's design, build, and scale AI software tailored to your products, workflows, and requirements.
