For more than two decades, Ahmed Darrat has worked at the intersection of transportation policy, operations and technology. As chief product officer at INRIX, he leads the product management team and is responsible for delivering data and SaaS applications in the curbside, location intelligence, safety and traffic operations verticals.

Before joining INRIX, Darrat worked in engineering, policy and operations roles at the city of Seattle for 10 years, culminating in his role as the transportation policy advisor to the mayor.

In this Q&A with Roads & Bridges, which took place over email, Darrat argues that transportation artificial intelligence (AI) should be judged not by engagement or novelty, but by tangible outcomes such as less delay, more reliable travel times and fewer crashes. He also explains which operational decisions are poised for the biggest transformation. He argues that the biggest obstacles to adoption are now institutional rather than technical. 

R&B: What distinguishes infrastructure AI from consumer AI, and why do you believe transportation may become one of the clearest demonstrations of AI’s long-term economic value?

Ahmed Darrat: Consumer AI is judged by engagement. However, infrastructure AI is judged by outcomes. When AI is embedded in the systems that move people and goods, novelty doesn’t matter. What matters is whether the system is reliable, accurate, and delivers measurable operational improvement.

Transportation is one of the clearest tests of that standard because the problem is so quantifiable. Congestion cost American drivers more than $85.8 billion in lost time last year, with the average driver losing 49 hours sitting in traffic. Every improvement AI delivers, whether it’s signal retiming, faster incident response, or more efficient freight routing, shows up directly in time saved, fuel saved, crashes avoided, and dollars recovered. Few applications of AI can draw as straight a line from the technology to economic value.

R&B: Much of the AI conversation focuses on individual productivity gains, but transportation agencies are trying to improve economic efficiency across entire networks. How should the industry measure the success of AI in transportation? Are we talking about reduced congestion, improved safety, lower operating costs, better asset utilization, or something else entirely?

AD: All of those matter, but they’re really components of one larger measurement: the economic efficiency of the network. How much delay, risk, and waste does it take to move a person or a shipment from A to B?

Within that, I’d emphasize two metrics. The first is travel time reliability. For freight and transit, especially, a predictable trip is often more important than a fast one. The second is safety outcomes, because crashes both directly impact the health and safety of drivers and are one of the largest sources of unexpected traffic delay.

Overall, the industry needs to tie AI investments to outcomes, not raw activity. Real results lie in fewer crashes at an intersection, less delay on a corridor, or lower cost per improvement.

R&B: Transportation agencies have collected enormous amounts of traffic, weather, construction, and asset data for years. What has changed in the last few years that allows AI to transform that information into better operational decisions rather than simply better reporting?

AD: Three things have changed in the last few years. First, the data itself changed. Connected vehicles now generate continuous, network-wide observations that previously required physical sensors at every location. This means that agencies can see how every intersection and corridor performs, even if they don’t have traditional hardware.

Second, cloud computing made it possible (and practical) to look at that data in real time and holistically, factoring in other variables like weather, construction, and crashes. Historically, that information was stored in separate systems that didn’t talk to each other, so the best an agency could do was look backward.

Third, and most recently, AI closed the gap between analysis and action. Beyond just reporting what happened, modern systems can also diagnose why it happened, recommend a response, and predict the outcome. This shift now puts insights in the hands of staff who aren’t data scientists, which is most of the people running our transportation networks.

R&B: Looking ahead five years, where do you expect AI to have the greatest impact on reducing congestion and improving freight mobility? Are there particular operational decisions—from signal timing and incident response to work zones and freight routing—that you believe are poised for the biggest transformation?

AD: Signal timing is the biggest near-term opportunity. Signals are the control point of the entire surface network and retiming them is one of the most cost-effective interventions in transportation, with the Institute of Transportation Engineers estimating benefit-cost ratios as high as 40 to 1. And with AI, it’s possible to identify underperforming intersections continuously across a whole region, instead of studying a handful of corridors every few years.

Incident response is close behind. Faster accident detection and clearance can prevent secondary crashes, which make up a significant share of freeway collisions.

For freight, the transformation is at the two ends of the trip: predictive routing that accounts for reliability, and the last mile, where drivers lose time circling for loading zones and curb access. Additionally, throughout the trip, better data can tell agencies when and where to expect lane closures from work zones, lessening the impact of closures on timely deliveries.

 The common thread is a shift from periodic, reactive decisions to continuous, predictive ones.

R&B: The technology is advancing rapidly, but implementation often moves more slowly. What do you see as the biggest barriers preventing transportation agencies from realizing AI’s full potential? Is the challenge primarily data quality, interoperability, organizational change, procurement, public trust, or something else?

AD: The technology itself is rarely the barrier anymore. Rather, the constraints are institutional.

The current procurement approach typically focuses on buying hardware in multi-year cycles, instead of investing in software that improves monthly. As touched on earlier, data is often siloed, both across departments and between neighboring jurisdictions that share the same corridors. And even though agencies are being asked to adopt new tools, they still have to work with flat budgets and thinning technical staff.

When thinking about deploying infrastructure AI, trust is the piece that deserves the most care. These are safety-critical systems, and agencies are right to demand that any AI tools recommended be transparent and verifiable before deploying them. The approach I’ve seen work the best is to start with a specific, defined problem (like a corridor or a set of intersections), demonstrate that the tool can deliver real outcomes, and then scale from there. That’s how agencies can build lasting confidence and trust.