Potholes Cost Cities Millions. This Company Is Using AI and Trucks to Fix Them.

Potholes Cost Cities Millions. This Company Is Using AI and Trucks to Fix Them.

In Brief: San Francisco-based Samsara has launched Ground Intelligence, an AI system that uses cameras already mounted in millions of commercial trucks to detect potholes, assess their severity, and track deterioration over time. Chicago is among the first cities to sign on. The system represents a shift from reactive 311-call-based maintenance to proactive, data-driven infrastructure monitoring, with potential applications extending to graffiti, broken guardrails, and downed power lines.

This is exactly the kind of public sector AI deployment that deserves scrutiny. If the implementation questions interest you, Human x AI Europe on May 19 in Vienna is where Europe's AI ecosystem sits down to work through the operational details.

The Reactive Maintenance Trap

Every public works director knows the drill. A pothole forms. Someone calls 311. The call gets logged. A crew gets dispatched. The pothole gets filled. Repeat, endlessly, across thousands of lane miles.

This reactive model has a fundamental problem: by the time a citizen reports a pothole, it has already damaged vehicles, created liability exposure, and often deteriorated to the point where a simple patch will not hold. According to Samsara's announcement, potholes cost the U.S. more than $750 million annually in material repair costs alone. That figure does not include vehicle damage claims, injury lawsuits, or the compounding cost of deferred maintenance.

The 2025 ASCE Report Card for America's Infrastructure puts the scale of the problem in context: approximately 39% of major U.S. roads remain in poor or mediocre condition. Poor road conditions cost the average driver over $1,400 per year in vehicle damage, operating costs, and lost time. AAA data shows that pothole damage cost U.S. drivers $26.5 billion in repairs in 2021, with one in ten drivers sustaining damage significant enough to require repair.

The question is not whether cities need better infrastructure monitoring. The question is whether AI-powered detection systems can actually deliver on the promise of proactive maintenance.

What Samsara Is Actually Shipping

Samsara's Ground Intelligence, announced this week, takes a different approach than previous pothole detection efforts. The company has spent a decade installing cameras in millions of commercial trucks for driver monitoring, theft prevention, and liability documentation. That existing camera network now becomes an infrastructure sensing platform.

The system works by training AI models on the vast dataset of road footage Samsara has accumulated. The model can detect multiple types of potholes and, critically, assess severity using a g-force threshold filter that measures the actual impact when vehicles hit defects. This goes beyond simple image recognition: the system captures how bad a pothole actually is for vehicles passing over it.

That's the magic here; it takes a process that was reactive and makes it proactive. That means that you don't just go and fix one pothole. You plan it out.

Johan Land, Samsara's senior vice president of product

The dashboard proactively populates warnings on a map of developing potholes. Cities can also pull anonymized footage from vehicle cameras to confirm citizen reports of other infrastructure problems: downed street signs, clogged sewers, broken guardrails.

Chicago is coming on as a customer. Multiple other cities are already under contract.

The Coverage Advantage

Last month, Waymo and Waze announced a pilot program to share pothole data with local governments in five metro areas. The pilot uses Waymo's perception systems to detect potholes and makes the data available through the free Waze for Cities platform.

Samsara's pitch is that commercial trucks offer something Waymo's robotaxi fleet cannot: coverage and repeat observation. Waymo's fleet currently stands at around 3,000 vehicles. Samsara-equipped trucks number in the millions and operate on routes that cover 99% of major U.S. roads, according to the company's blog post.

More importantly, commercial trucks traverse the same routes repeatedly. A delivery truck that runs the same route five days a week generates longitudinal data showing how a pothole changes over time. That deterioration tracking is what enables predictive maintenance: identifying which defects are getting worse fastest and prioritizing repairs before they become expensive failures.

The Waymo-Waze pilot has identified approximately 500 potholes across its five launch cities. Samsara's network, by contrast, claims continuous coverage across most major road networks.

Implementation Questions Cities Should Ask

Before signing a contract, public sector technologists should work through several operational questions.

Data Ownership and Retention

Who owns the pothole data? How long is footage retained? Can cities export their data if they switch vendors? The anonymization claim needs scrutiny: what exactly is anonymized, and what metadata remains attached to footage?

Integration with Existing Systems

Most cities already have asset management systems, GIS (Geographic Information System) platforms, and work order systems. Ground Intelligence works as a dashboard, but the real value comes from integration with existing workflows. What APIs are available? What does integration actually require?

Accuracy and False Positive Rates

Any detection system generates false positives. A crack that looks like a pothole. A shadow that triggers an alert. What is the actual precision and recall of the model? What is the human review burden for a city with 10,000 lane miles?

Coverage Gaps

Commercial trucks do not drive every road. Residential streets, cul-de-sacs, and low-traffic areas may have minimal coverage. Cities need to understand where the system provides good data and where traditional inspection methods remain necessary.

Procurement and Pricing

The announcement does not include pricing details. Cities need to understand the cost model: per-mile, per-alert, subscription-based? How does the cost compare to traditional inspection methods?

Rollback Plan

If the system underperforms or the vendor relationship ends, what happens? Cities should not build critical infrastructure monitoring around a single vendor without understanding the exit path.

The Broader Pattern

Samsara's Ground Intelligence is part of a larger trend: repurposing existing sensor networks for infrastructure monitoring. Vehicles equipped with cameras and accelerometers become mobile sensing platforms. The marginal cost of adding pothole detection to an existing fleet management system is far lower than deploying dedicated inspection vehicles.

This pattern has implications beyond potholes. Samsara's announcement mentions future features including graffiti detection, broken guardrails, and low-hanging power lines. The same approach could extend to traffic sign condition, pavement marking visibility, or vegetation encroachment.

The risk is that cities become dependent on private sensor networks for public infrastructure monitoring. If the business model changes, if the vendor pivots, or if pricing becomes unsustainable, cities may find themselves without the data they have come to rely on.

The opportunity is genuine: proactive maintenance reduces costs, improves safety, and extends infrastructure life. The U.S. Department of Transportation has estimated that every $1 spent on timely maintenance prevents roughly $7 in later repairs. AI-powered detection could help cities capture that seven-to-one payoff.

What Success Looks Like

For cities evaluating Ground Intelligence or similar systems, success should be measured in operational terms:

• Reduction in reactive maintenance. Are crews spending less time responding to 311 calls and more time on planned repairs?
• Decrease in vehicle damage claims. Are liability costs going down as potholes get fixed earlier?
• Improvement in pavement condition scores. Are roads actually getting better, as measured by standard pavement condition indices?
• Cost per lane mile. Is the total cost of maintaining roads decreasing when the monitoring system cost is included?

These metrics require baseline measurement before deployment and consistent tracking afterward. Cities that cannot answer these questions after 12 months of operation have not implemented the system properly.

The technology is interesting. The implementation is what matters.