Repair Dashboard: An AI solution for predicting potholes and managing repairs

The Problem

The City of Edmonton’s extreme temperature fluctuations—marked by aggressive freeze-thaw cycles—allow moisture to infiltrate pavement cracks, expand upon freezing, and create subsurface voids that rapidly collapse under everyday traffic volume. Managing this recurring phenomenon places immense operational strain on the municipality. Historically, the City of Edmonton’s operational logistics depended heavily on manually triaging public 311 telephone notifications and mobile app requests.

This baseline approach introduced three severe constraints:

1. Labour-Intensive Cross-Referencing: Inspectors were forced to sift through unstructured, scripted line data and text addresses. Manually comparing these files with static digital or physical maps to build dispatch itineraries was incredibly time-consuming and inefficient.
2. Triangulation and Duplicate Inefficiencies: Public reports frequently targeted identical hazards but utilised variable address formats or localised descriptions. Without a centralised spatial triage system, crews were routinely cross-dispatched to the same physical location on multiple occasions, inflating administrative burdens and misallocating field capacity.
3. Absence of Predictive Capability: Road crews relied entirely on historical lag data. Lacking the ability to anticipate where pavement failures were actively developing, the city was trapped in a cycle of repairing major, deep-seated defects rather than addressing minor flaws proactively before they escalated.

With an annual Asphalt Road Maintenance Budget tightly fixed at $10.9M, a traditional solution—such as purchasing expensive proprietary enterprise software or deploying extensive hardware networks across municipal fleets—was financially prohibitive and structurally unfeasible.

The Solution

To overcome these structural hurdles, the City of Edmonton’s Data Science and Research (DSR) team developed Repair, a localised, city-wide intelligent dashboard. Positioned as a breakthrough in public sector innovation, the application was built entirely using open-source tools within the city's existing operational budget, led by a Computer Science co-op student working under the guidance of senior data scientists. The software's architecture is built upon three core technical mechanisms engineered to modernise field operations:

Part 1: Unified Spatial Data Aggregation and Visualisation

Repair systematically breaks down systemic data silos by ingesting and consolidating over six years of isolated, historical road damage records, real-time 311 public notification feeds, and internal field logs into a single, centralised database. The platform translates raw text data and disjointed address scripts into a dynamic, user-friendly geographic mapping interface. This visual framework gives inspectors real-time visibility into current road decay, moving the city from text-based lists to coordinated, map-driven routing and automated inspection scheduling.

Part 2: Spatial Clustering and Intelligent Duplicate Detection

To mitigate the duplicate reporting drain, the platform utilises spatial clustering algorithms. When multiple public 311 notifications are received for the same general vicinity—such as distinct residents reporting a single cluster of potholes along a busy arterial wheel path—the system evaluates proximity and groups the disparate notifications into a single, unified ticket. This automated verification eliminates redundant field dispatches and allows administrators to filter out the noise during severe freeze-thaw cycles.

Part 3: Predictive Machine Learning Modelling

The core innovative feature of Repair is its machine learning predictive analytics engine. By leveraging over half a decade of historic road maintenance data, local traffic volumes, pavement attributes, and environmental variables, the AI model maps and predicts where potholes are highly likely to emerge next. This predictive intelligence generates automated hotspots and proactive maintenance recommendations. As a result, municipal teams can shift resources dynamically, conducting targeted inspections and applying durable asphalt patches to minor defects before significant structural failure impacts public safety.

The Impact

The Repair dashboard has transitioned from a localised pilot into a fully scaled, city-wide operational success story. High-level outcomes include.

1. Substantial Operational Efficiency In 2024 alone, the dashboard successfully flagged and consolidated 1,559 duplicate notifications out of 7,544 total public reports. By intercepting these redundant files, the city achieved an immediate 20% savings in administrative and inspection staff time, allowing those vital hours to be reallocated directly to road repair execution.
2. Data-Driven Prioritisation The city successfully managed immense winter surges (such as repairing 11,296 potholes in January 2025 compared to just 1,726 in January 2024) by using the software to accurately separate distinct repair needs from duplicate public reporting noise.
3. Proactive Field Risk Management The city effectively established a structured, safety-first dispatch matrix, automatically elevating the priority of deep defects situated in high-velocity arterial lanes or high-traffic collector routes.
4. Qualitative Civic Gains Shifting to an AI-backed proactive maintenance model has significantly minimised vehicle damage across the community, upgraded municipal customer service, and advanced broader urban equity by replacing a "squeaky wheel gets the grease" model with objective, data-validated risk assessments.

Lessons Learned

Other public sector leaders looking to replicate this deployment can draw three critical, transferable insights from Edmonton's experience.

• Embrace Open-Source Development to Bypass Capital Gridlock. Municipalities do not need multimillion-dollar software procurements to build high-impact AI tools. By utilising open-source data toolkits and leveraging cost-effective talent pipelines—such as academic co-op partnerships steered by internal senior technical mentors—cities can deliver resilient, specialised solutions completely within existing budgets.
• Align Data Volume with Repair Capacity. When designing modern sensor-driven infrastructure systems, teams must balance data collection with field resources. Edmonton originally explored installing IoT sensors on vehicles and capturing aerial photography, but pivoted to designing tools that optimise existing reporting structures to ensure immediate field adoption.
• Prioritise Workflow Integration Over Raw Technology. The true success of the Repair framework lies in its user experience. Moving from raw, tabular data sheets to intuitive, interactive visual maps lowered the barrier to entry for field personnel. Technology projects must be treated primarily as cultural transformations; working directly with end-user focus groups ensures that the final software actively simplifies daily workflows rather than complicating them.

Launch year: 2023