We collaborated with Viston to optimize routing, dispatch, and driver support with an operations copilot in Brazilian Portuguese.

Business challenge

• Rising last-mile costs, inconsistent on-time delivery, and dispatcher overload.
  
• Urban constraints (traffic, restricted zones) and seasonal spikes.
  

Our approach and solution

• Built a daily and intra-day re-optimization engine with a dispatcher copilot.
  
• Deployed a driver mobile assistant for exceptions, proof-of-delivery, and micro-route tweaks.
  

AI applications and benefits delivered

• OR-based route optimization with live telemetry feedback loops.
  
• ETA prediction and dynamic re-sequencing.
  
• LLM copilot that summarizes exceptions, drafts customer messages, and translates with tone control.
  

Cost of implementation

• USD 90,000 including telematics integrations and mobile app enhancements.
  

Time to implement

• 16 weeks MVP (two cities); 28 weeks nationwide rollout.
  

Tools and technologies used

• Google OR-Tools, PyTorch for ETA models, PostgreSQL/PostGIS, Kafka, FastAPI, Flutter mobile, Mapbox/OSRM, Redis, LangChain, Azure OpenAI or local Llama-3 as fallback, Grafana, Prometheus, Sentry.
  

Quantitative outcomes

• 17% reduction in cost per delivery.
  
• On-time delivery improved by 11 percentage points.
  
• 12% reduction in CO2 per package.
  
• 23% fewer dispatcher interventions.
  

Key performance indicators (KPIs) tracked

• Cost per drop, OTIF, average route duration, first-attempt delivery success, ETA MAE, CO2 per package, dispatcher touches per 100 orders.
  

Pre- and post-implementation metrics

• Cost per delivery: BRL 16.40 → BRL 13.60.
  
• OTIF: 82% → 93%.
  
• ETA MAE: 14.2 minutes → 7.6 minutes.
  
• Dispatcher touches/100 orders: 61 → 47.
  

Stakeholder quotes or testimonials

• “The copilot shaved minutes off every exception and standardized customer comms.” — Head of Operations.
  
• “Drivers trust the app’s re-sequencing when traffic spikes.” — Regional Logistics Manager.
  

Regulatory or compliance considerations

• LGPD (Brazil), telemetry data minimization, opt-in consent for driver analytics, anonymized reporting, compliance with ANTT guidelines for fleet operations.