Parcel operations have more data than ever: rates, tracking events, service commitments, lane performance, exception codes, package dimensions, customer promises, and carrier scorecards. Yet better visibility does not automatically produce a better shipping decision.
The gap is execution.
If new information reaches a dashboard but the label is still purchased using a static rule, the operation may understand its performance better without materially changing it.
The supplied Luma AI Select case study shows what happens when the intelligence layer is moved to the transaction itself. A global recommerce marketplace processing more than 25,000 labels per day used EasyPost's Luma AI Select to evaluate each shipment at label creation. The reported results: 4-5% lower per-label cost, more than $2 million in annual savings, on-time delivery improving from 80% to 83%, and approximately 273,000 fewer late deliveries per year [1].
The company did not add carriers, renegotiate contracts, change packaging, add headcount, or make significant fulfillment changes during the measured deployment. The intervention was the decision layer.
That distinction matters for every operations leader investing in AI.
Visibility Is Not Decisioning
A dashboard can tell you that Carrier A underperformed on a lane last week. A scorecard can show that Service B is expensive for a particular package profile. A BI system can identify late-delivery clusters.
Those insights are useful. But the operational value depends on what happens next.
If the routing rule remains unchanged until a quarterly review, the organization has created information latency: the evidence exists, but the decision system is not using it when the shipment is committed.
Shipment-level decisioning reduces that gap by bringing relevant context into the label-creation event.
The Luma case study describes a model that evaluates destination zone, package profile, delivery window, cost, and historical carrier performance across more than one billion comparable shipments. It then selects among eligible service levels for the individual shipment.
The result is not simply a more sophisticated report. It is a different operational action.
The Difference Between Knowing and Acting
Consider two shipping organizations with identical carrier data.
Organization A uses the data in weekly dashboards. Routing defaults are updated manually when the team identifies a persistent issue.
Organization B uses the same categories of evidence to influence service selection at label creation, with governance and fallback rules for exceptions.
Both organizations are data-driven. Only one has connected the data to the decision moment.
That connection is where AI can create measurable operational leverage.
This principle extends beyond parcel shipping. In May 2026, HERE Technologies announced Location Reasoning, designed to ground AI agents in live location and road intelligence so that real-world decisions reflect current geographic context. HERE also introduced AI-powered last-meter guidance for delivery drivers. Both announcements reflect the same design principle: operational AI needs context at the moment an action can still change the outcome.
For parcel shipping, the label is one of those moments.
Why Carrier Data Alone Can Mislead
Carrier performance is not one number.
A national on-time rate can conceal large differences by service, destination, zone, lane, package profile, and time period. A carrier that is strong for one shipment type may be weaker for another. A premium service may have excellent aggregate performance but deliver little incremental value for a shipment with a generous delivery window.
Similarly, cost is not one number. Base rates, service level, package characteristics, destination, and surcharges all influence shipment economics.
A useful decision system therefore needs to answer a more granular question:
Given this shipment, these eligible services, this delivery requirement, and the available performance evidence, which option is most appropriate now?
The Luma case study reports that answering this question at scale reduced both cost and late deliveries for the customer.
What the Case Study Proves - and What It Does Not
The source supports several verified statements about the customer deployment:
• More than 25,000 labels were processed per day through EasyPost.
• The primary carrier's on-time delivery rate was 80% before deployment.
• Luma AI Select evaluated shipments at label creation.
• Reported per-label costs fell 4-5%.
• Reported annual savings exceeded $2 million.
• On-time delivery increased from 80% to 83%.
• The customer calculated approximately 273,000 fewer late deliveries annually.
• The company did not add carriers, renegotiate contracts, add headcount, change packaging, or make significant fulfillment changes as part of the measured deployment.
• EasyPost compared before-and-after per-label cost and on-time performance across comparable U.S. shipments.
The source does not prove that every shipper will achieve the same savings or service improvement. The customer is unnamed. The results should therefore be used as customer-specific evidence of potential operational leverage, not as a universal benchmark.
That evidence discipline makes the story stronger, not weaker.
Diversification Creates Choices; Decisioning Allocates Them
2026 parcel strategy is increasingly focused on diversification. PARCEL's January/February 2026 analysis describes parcel sourcing as portfolio design, where shippers can rebalance volume as pricing and network conditions change. TransImpact's May analysis highlights diversification as a tool for cost, leverage, and risk management. DCL Logistics' July 2026 guidance similarly frames carrier choice around shipment fit rather than a single "best" provider.
But a diversified carrier network does not select itself.
More carriers mean more possible combinations of services, rates, performance characteristics, and operating constraints. Without better allocation logic, diversification can simply increase complexity.
This is why the decision layer matters. Procurement establishes the menu. Decisioning chooses the item.
Designing an Executable AI Decision
A production-ready shipment decision needs six components.
1. Defined decision
Be precise: select a carrier/service for this shipment. Avoid vague objectives such as "optimize logistics."
2. Eligible option set
Only evaluate services that are contractually and operationally valid for the shipment.
3. Current context
Provide the variables that materially change the choice: destination, package characteristics, delivery promise, current cost inputs, and relevant performance evidence.
4. Objective function
Define what "better" means. It may be minimum expected cost subject to an on-time threshold, maximum on-time probability under a cost ceiling, or a weighted balance.
5. Governance
Set thresholds for automatic execution, human review, exception handling, and fallback when evidence is incomplete.
6. Read-back
Capture actual transportation cost and delivery performance after execution. Compare prediction to outcome.
Without read-back, the system cannot demonstrate that its recommendations improved the operation.
A Leadership Checklist
Before approving another visibility or AI initiative, ask:
• Which operational decision will change because of this data?
• At what exact moment is that decision made?
• Is the relevant evidence available at that moment?
• Can the system act, or does the insight wait for manual review?
• What happens when data are stale or incomplete?
• How is the recommendation explained and audited?
• What outcome is read back after execution?
• Can we compare the new logic with the prior baseline on comparable shipments?
If the project cannot answer those questions, it may improve reporting without improving execution.
The Intent Amplify View
AI in parcel operations should not be judged by the sophistication of the model alone. It should be judged by the quality of the decision, the control around execution, and the measurable outcome that follows.
The Luma AI Select case study is a useful proof point because the intervention occurred at a high-frequency operational decision: label creation. For the customer, changing that decision was associated with lower per-label cost and better on-time performance without a carrier or fulfillment overhaul.
For IT and Engineering leaders, the practical test is whether trusted shipment, cost, eligibility, and performance data can reach the label-generation workflow with sufficient reliability, observability, fallback behavior, and decision-time speed. Better data matters. But data creates operating value only when it changes what the business does.
What Changes for Transportation Teams
Shipment-level decisioning does not eliminate carrier management; it changes what carrier management can measure. Instead of reviewing only aggregate share and average performance, transportation teams can examine how individual services win or lose shipments under defined operating objectives.
That visibility can improve sourcing conversations. If a service is consistently eligible but rarely selected because its cost-performance combination is weak on important shipment segments, the team has more precise evidence for negotiation or network redesign. If another service repeatedly performs well on a narrow set of lanes, the organization can understand where that option creates differentiated value.
The decision log also provides an early-warning mechanism. A sudden shift away from a service may reflect changing performance evidence, cost, eligibility, or package mix. Rather than discovering the issue only in a monthly aggregate, teams can investigate the decision drivers while the pattern is developing.
This makes allocation intelligence complementary to procurement intelligence. Procurement creates and improves the option set. Shipment-level evidence reveals how those options perform when confronted with real shipment requirements.
What Changes for Finance Teams
Finance should insist on a clean distinction between theoretical savings, predicted savings, and realized savings. A rate comparison can identify theoretical opportunity. A recommendation can produce predicted opportunity. Realized value requires an executed shipment and an agreed comparison method.
The same principle applies to avoided service failures. A model may predict that an alternative will improve on-time probability, but the achieved outcome should be measured after delivery. When financial and operational reporting use the same evidence boundary, the business case becomes easier to defend and easier to improve.
This discipline also prevents successful pilots from being overstated. If concurrent changes in carrier contracts, fulfillment, package mix, or customer promise could affect the result, disclose them. A qualified claim supported by clear evidence is more useful than a larger claim with ambiguous attribution.
A Practical Architecture for Decision Intelligence
The most useful way to operationalize shipment-level intelligence is to separate the workflow into layers that can be measured and governed independently. The first layer is eligibility. Before optimization begins, the system needs to know which carrier and service options are actually available for the shipment. Contractual constraints, destination coverage, package characteristics, pickup schedules, delivery requirements, and operational restrictions belong here.
The second layer is evidence. Eligible options should be evaluated with the most relevant cost and performance context available at decision time. That includes shipment-specific cost where possible and sufficiently granular historical delivery evidence. A national carrier average may be useful for broad management reporting, but it can conceal differences that matter to a specific lane, service, destination, or package profile.
The third layer is objective design. An optimization system needs an explicit definition of better. If the objective is only lowest cost, the system can rationally select an option that undermines the delivery promise. If the objective is only fastest delivery, it can rationally overspend. A production objective should encode the business trade-off: meet the required promise and risk threshold while controlling transportation cost.
The fourth layer is authority. Not every recommendation needs to execute automatically. Common, well-understood shipment profiles with complete data can be candidates for automated selection. Unusual shipments, missing evidence, service disruptions, contractual exceptions, or low-confidence recommendations can remain under human control. This creates a graduated operating model rather than an all-or-nothing automation decision.
The fifth layer is read-back. Once the shipment is delivered, actual cost and service performance should be attached to the original decision. Without this layer, the system can generate increasingly sophisticated recommendations without proving that they improve outcomes. Read-back creates the evidence required to tune rules, retrain models where appropriate, and challenge assumptions that have become stale.
How to Build the Business Case
A credible business case starts with the addressable decision pool, not total parcel volume. Identify shipments where multiple eligible services exist and where the choice can materially affect cost or delivery performance. Remove shipments where operational or contractual constraints leave no meaningful alternative.
Next, establish the baseline. Measure cost and on-time performance under the current allocation logic across comparable shipment segments. Where possible, control for changes in carrier contracts, package mix, fulfillment operations, and customer promises. The purpose is not academic perfection; it is to avoid attributing unrelated operating changes to the decision system.
Then run alternative recommendations in shadow mode. For each shipment, record whether the recommended service differs from the current default, the expected economic effect, the expected delivery effect, and the evidence behind the change. This creates a reviewable decision log and lets transportation teams test whether the system is finding genuine opportunities or merely generating churn.
Only after that evidence is credible should a controlled portion of traffic move to execution. Compare actual outcomes with the baseline and with the predicted result. Track exceptions and overrides as carefully as savings because they reveal where the operating model needs stronger constraints or better data.
The Luma AI Select case study is useful in this context because it demonstrates the type of measurement leaders should demand: a defined decision intervention, comparable shipments, and explicit reporting of major operating factors that did not change during the measurement period [1]. The reported magnitude is customer-specific. The measurement discipline is broadly applicable.
Governance Questions Before Scale
Before expanding automated allocation, leadership should be able to answer several questions clearly. Who owns the objective function? Which data sources are authoritative at label time? How quickly is degraded carrier performance reflected in recommendations? Which shipment classes are excluded from automation? What triggers fallback to static logic? Who reviews overrides? How are realized savings separated from rate-card opportunity? How is service quality protected when a lower-cost option is selected?
These questions keep decision intelligence connected to operating accountability. AI does not remove the need for transportation policy. It makes that policy more explicit because the system needs defined objectives, boundaries, and evidence to act consistently.
The end state is not a black box choosing carriers. It is a controlled decision service that evaluates the options already made available by procurement, applies current evidence to a specific shipment, executes within approved authority, and learns from the actual outcome. That architecture is what turns shipping data into operational intelligence rather than another retrospective dashboard.
Read the full Luma AI Select case study
References
[1] EasyPost. Luma AI Case Study: $2M+ in Savings. 273,000 Fewer Late Deliveries. Without Changing Carriers. Primary evidence. https://intenttechpub.com/POC/supply-chain-now/luma-ai-case-study.html
[2] PARCEL. Rethinking Parcel Diversification. January/February 2026 issue; published April 9, 2026. https://parcelindustry.com/article-6631-Rethinking-Parcel-Diversification.html
[3] TransImpact. Beyond the Big Two: When Parcel Carrier Diversification Makes Sense. May 11, 2026. https://transimpact.com/blog/beyond-the-big-two-when-parcel-carrier-diversification-makes-sense
[4] DCL Logistics. Understanding the Parcel Carrier Landscape and How to Diversify for Ecommerce Growth. July 20, 2026. https://dclcorp.com/blog/shipping/understanding-the-parcel-carrier-landscape-and-how-to-diversify-for-ecommerce-growth/
[5] HERE Technologies. HERE unveils AI-powered last-meter guidance solution to help delivery drivers complete the final handoff. May 14, 2026. https://www.here.com/about/press-releases/here-unveils-ai-powered-last-meter-guidance-solution-to-help-delivery-drivers-complete-the-final-handoff
[6] HERE Technologies. HERE Technologies unveils Location Reasoning, redefining geospatial grounding for real-world AI decisions. May 19, 2026. https://www.here.com/about/press-releases/here-technologies-unveils-location-reasoning-redefining-geospatial-grounding-for-real-world-ai-decisions
