The Descartes Systems Group Webinar Spotlights AI to Close Last-Mile Delivery Gaps, Boost ETAs

Key Points

• Descartes showcased how AI/ML-driven “machine-learned service times” can close the persistent 10–20% planning–execution gap in last-mile delivery, improving ETA accuracy and potentially boosting route density by as much as ~30% without adding trucks or drivers.
• They recommend building a connected operations environment that links planning, execution, and customer communications so ETAs update in real time and proactive notifications reduce missed windows, overtime, and reactive dispatching.
• For adoption, Descartes advises to start capturing actual service time data and focus on a single use case (e.g., ETA accuracy) to prove value quickly—models can begin with limited history and improve over time, with measurable impact possible within 90 days.

The Descartes Systems Group NASDAQ: DSGX used a webinar hosted by SupplyChainBrain Editor-in-Chief Bob Bowman to outline how artificial intelligence and machine learning are being applied to close the persistent gap between planned delivery routes and real-world execution. The session, titled “The AI Exchange: Inside the Last Mile, AI, Delivery Engagement, and the New Standard for On Time and In Full,” was positioned as the first of four Descartes webinars planned for the year.

Rising costs and higher customer expectations

Cyndi Brandt, Vice President of Fleet Solutions at Descartes, said AI is “landing kind of right in the middle of some very, very real operational pressures,” pointing to rising costs and labor challenges alongside shifting customer expectations. Brandt said fleet leaders are facing higher expenses for “fuel, insurance, equipment, wages,” and that inefficiencies that once went unnoticed are now “hitting your margins almost immediately.”

Brandt also said customers have moved away from accepting broad delivery windows and now want “precise, super accurate ETAs,” real-time visibility, and a “seamless experience” that mirrors B2C delivery standards. She characterized last-mile delivery as “mission critical” not only to customer experience but also to business cost structure, forcing fleets to balance efficiency and service quality more tightly than before.

Why planned routes diverge from execution

In the discussion, Bowman cited that many fleets still see a “10%-20% gap between planned routes and actual execution.” Brandt attributed that gap to routing plans that rely on assumptions rather than what happens on the road, citing factors such as traffic variability, inconsistent service times, blocked docks, last-minute order changes, and driver behavior. Without feeding execution data back into planning, she said, “those same inaccuracies are gonna continue day after day,” leading to less accurate ETAs, missed time windows, and dispatch teams forced into “firefighting.”

Sergio Torres, Senior Vice President of Product Management at Descartes, said static service time assumptions persist because they are easy to operationalize, but they fail to reflect modern delivery complexity. “If a delivery to a retail location is fundamentally different from a construction site, why would you use the same service times?” Torres asked, adding that missing these differences introduces “a systematic error into every single route,” which can reduce driver compliance and increase midday adjustments.

Machine-learned service times and connected operations

Torres said Descartes’ approach is rooted in using collected operational data—route planning, route execution, driver behavior, and safety indicators such as harsh braking and speeding—to generate insights through what he called “Fleet Data Intelligence.” He also described a closed-loop concept involving “René,” which he said can identify improvement opportunities, summarize insights, and trigger tasks such as activating machine learning for service-time predictions and applying those predictions during route optimization.

Torres said machine-learned service time predictions improve planning accuracy by grounding routes in actual delivery behavior, factoring in variables including order size, product type, geography, equipment requirements, and customer-specific patterns. He said the result can be more consistent schedules, fewer reactive dispatch interventions, and improved utilization—potentially “improving route density significantly without adding any trucks or drivers,” which Bowman referenced as being as high as roughly 30% in some cases.

Brandt argued that much of the market discussion around “AI-based routing” centers on adjusting routes during execution, rather than improving planning. Descartes’ position, she said, is that measurable gains come from closing the planning-execution gap by using execution data such as actual service times and driver behavior. She highlighted potential operational metrics including more accurate ETAs, fewer missed delivery windows, reduced overtime, reduced idle time, improved route adherence, and better asset and labor utilization.

Torres added that service times can vary widely and said machine learning is suited to capturing multi-factor variability that static models cannot, enabling more predictable performance and reducing downstream disruptions by shifting teams from reactive to proactive decision-making.

On systems integration, Torres said many fleets still operate with disconnected tools for routing, dispatch, and customer communication, which can create “gaps in visibility” and slow decisions due to stale data. A connected environment, he said, links planning, execution, and customer communications so exceptions can be managed in real time with updated ETAs and proactive notifications.

Customer engagement as a competitive advantage

Brandt said delivery has become a continuous “customer journey” rather than “a series of individual stops,” and she emphasized the role of customer engagement platforms that combine planning information with live route tracking and predictive insights. She described using timing and context to improve notifications—for example, waiting to send a “10 minutes away” message until it is more accurate based on updated arrival predictions.

Both speakers said proactive communication can mitigate service issues. Torres compared the connected experience to ride-sharing visibility, arguing that real-time, end-to-end connectivity helps fleets and customers respond to exceptions such as delays, schedule changes, and no-shows while protecting service commitments.

Adoption, first steps, and near-term priorities

Torres said successful adoption depends on aligning technology with operational goals and involving planners, dispatchers, and drivers early so tools fit naturally into workflows. During the audience Q&A, Brandt said a practical first step in moving away from static planning is to “start capturing and using actual service time data that’s coming in from the field,” noting fleets may already have data from telematics and driver apps but aren’t using it systematically.

Torres said fleets do not need a large amount of historical data to begin benefiting from machine learning, adding that models can improve over time as routes are executed and measured. He also cautioned that a common mistake is “trying to do too much at once,” recommending a focused use case, proof of value, and then scaling.

Asked what capabilities will define high-performing fleets over the next 12 to 18 months, Brandt pointed to moving away from “static optimization” toward plans informed by historical execution data, alongside fully integrated customer engagement covering the entire order journey. In closing, Brandt and Torres said leaders should shift from planning based on assumptions to planning based on “actual delivery behavior,” with a near-term action item centered on capturing and using real delivery data, and focusing on a single use case—such as ETA accuracy—to drive impact within 90 days.

About The Descartes Systems Group NASDAQ: DSGX

The Descartes Systems Group Inc NASDAQ: DSGX is a global provider of cloud-based logistics and supply chain management solutions. The company's software-as-a-service platform connects and optimizes the flow of goods, information and payments across the global supply chain, helping businesses coordinate transportation, customs clearance, routing, scheduling and fleet management. Descartes' modular applications serve shippers, carriers, third-party logistics providers and regulatory authorities by enabling real-time visibility, compliance and execution across complex trade networks.

Headquartered in Waterloo, Ontario, Descartes was founded in 1981 and has grown through a combination of organic development and strategic acquisitions.

This instant news alert was generated by narrative science technology and financial data from MarketBeat in order to provide readers with the fastest reporting and unbiased coverage. Please send any questions or comments about this story to contact@marketbeat.com.