It is 7:15 on a Tuesday morning and the fleet manager of a 38-truck regional carrier is reviewing the monthly numbers she will present to the owner at 9:00. The slide deck looks encouraging: loads dispatched up 12 percent, miles run up 9 percent, customer orders fulfilled on time at 94 percent. The owner will be pleased. What the slide deck does not show, because nobody built the column for it, is that deadhead percentage has climbed from 18 percent to 24 percent over the same three months, that three roadside breakdowns in the last six weeks cost a combined $47,000 in roadside repair, towing, and missed-delivery penalties, and that two Compliance, Safety, Accountability (CSA) violations have accumulated in the hours-of-service (HOS) category that will elevate the fleet's FMCSA (Federal Motor Carrier Safety Administration) score in the next quarterly refresh. The AI dispatch tool the fleet deployed eight months ago is working. The metrics the fleet chose to measure are just not the ones that tell the real story.

Why a Single-Axis Dashboard Fails a Fleet

Freight metrics cluster into two families that most carriers track independently and rarely force into a single conversation. The first family is the volume-and-throughput family: loads moved, miles run, loads per truck, on-time delivery rate, customer satisfaction. These metrics answer the question "are we doing more?" The second family is the margin-and-safety family: deadhead percentage, revenue per truck, breakdown rate, CSA score. These metrics answer the question "are we doing it profitably and safely?"

A fleet that measures only the first family is flying with half its instruments. It can demonstrate activity without demonstrating health. A fleet that deploys an AI dispatch or predictive maintenance tool, and then measures only loads and miles, is in exactly this position: the AI may be delivering genuine gains in matching efficiency and maintenance foresight, but if those gains are not captured in the margin-and-safety metrics, the owner has no way to know whether the tool is paying for itself. Worse, a volume-focused dashboard can actively conceal a problem. A fleet running more loads because its AI is proposing higher-volume routing may simultaneously be running more deadhead if the routing decisions are not optimized for loaded-mile efficiency. More loads, worse margins. More activity, more breakdowns if the maintenance signal is being ignored. The dashboard says progress. The P&L says otherwise.

The dual-axis scorecard is the answer. It places the margin metrics and the safety metrics in a single reporting structure so that leadership sees both families together, identifies trade-offs when they emerge, and refuses to declare AI success on volume alone. This is not an abstract framework. It is the measurement architecture that makes the difference between an owner who knows the AI is working and an owner who believes it is.

This lesson defines each metric on the dual-axis scorecard, shows how to calculate it, and explains how the two axes interact. The next lesson covers presenting the full picture to the owner in a defensible narrative. The lesson after that examines the specific patterns that cause metrics to hide problems rather than surface them.

The Margin Axis: Four Metrics That Reveal the Real Story

The margin axis of the dual-axis scorecard has four metrics that capture, between them, most of what matters about whether the AI is delivering the revenue recovery and cost reduction that justified deploying it. Each is a direct indicator of the empty-mile problem or its inverse.

Deadhead Percentage

Definition: Deadhead percentage is the share of total miles driven by a truck in a period that are driven without a paying load. It is calculated as follows: deadhead miles divided by total miles multiplied by 100. If a truck drives 1,000 miles in a week and 220 of those miles are empty, deadhead percentage is 22 percent. Some carriers refine this to loaded-mile percentage (the inverse: loaded miles divided by total miles), which is mathematically equivalent and perhaps more intuitive to present to an owner. Either convention is acceptable; the critical requirement is that the same convention is used consistently before and after the AI deployment so that the comparison is valid.

Why this is the primary margin metric: Every deadhead mile is fuel burned, driver-hours consumed, and truck wear accumulated that produces zero revenue. In the middle of an 80,000-driver shortage where driver-hours are the scarcest and most expensive input in a fleet, burning those hours on empty miles is the most costly waste in the operation. An AI dispatch system that reduces deadhead is directly converting the fleet's most wasted input into revenue. A fleet running at 25 percent deadhead that reduces to 18 percent has recovered 7 percent of total miles as paying miles. On a 38-truck fleet averaging 2,500 miles per truck per week, that is 6,650 miles per week converted from empty to loaded. At a hypothetical blended loaded-mile rate of $2.20 per mile, the weekly revenue recovery is $14,630, or roughly $760,000 per year. This is the math the owner needs to see, and it is only visible if deadhead percentage is tracked before and after deployment.

Baseline requirement: Before an AI dispatch tool is deployed, the fleet must document its current deadhead percentage, measured consistently over at least 30 to 60 days and broken down by lane, driver, and day-of-week if data quality allows. Without the baseline, the post-deployment change is unverifiable. A vendor that claims "our tool reduces deadhead by 8 percent" can only be evaluated against a baseline the fleet has already measured. An owner who asks "is the AI working?" deserves an answer grounded in pre-deployment data, not a comparison to an industry average that does not reflect the fleet's specific lanes and freight mix.

Decomposition: Deadhead percentage should be decomposed by driver, by lane, and by day of week. An aggregate improvement that conceals a specific lane or driver group with worsening deadhead is not a clean win. The decomposed view is what allows the fleet manager to target the AI's routing suggestions more precisely and to identify whether the improvement is driven by the busiest high-volume lanes (where it is easiest to find backhauls) or is occurring across the fleet's harder lanes as well.

Revenue Per Truck

Definition: Revenue per truck is the gross revenue generated by a truck (or by the fleet divided by truck count) in a defined period, typically per week or per month. It is distinct from revenue per mile (which measures rate quality rather than asset utilization) and distinct from revenue per driver (which conflates driver utilization with asset utilization). Revenue per truck is the asset-utilization metric: it measures whether the fleet's capital investment in tractors is generating returns proportional to the investment.

How AI affects it: An AI dispatch system that reduces deadhead and improves load-to-driver matching directly increases revenue per truck by increasing the share of miles that are paid. A fleet where each truck averages $8,500 per week in revenue at 25 percent deadhead could, if deadhead is reduced to 17 percent, see revenue per truck rise to approximately $9,300 per week, holding load rates constant, because more of the truck's available driving time is spent under load. The gain is the empty mile converted to a paying mile, valued at the fleet's average loaded-mile rate.

Why it beats revenue per mile as a strategic metric: Revenue per mile rewards high-rate loads. Revenue per truck rewards high asset utilization combined with reasonable rates. For a fleet manager evaluating an AI dispatch tool, the question is not just "did the AI find higher-rate loads?" but "did the AI use the trucks more productively?" Those are different questions and they are answered by different metrics. A fleet that uses AI to cherry-pick premium loads may improve revenue per mile while running the same deadhead and keeping trucks idle between loads. Revenue per truck exposes that outcome; revenue per mile does not.

Baseline requirement: As with deadhead percentage, the pre-deployment revenue per truck baseline must be documented at the truck level, over a representative period that covers seasonal variation if the fleet's freight mix varies by season. A comparison of summer post-deployment revenue per truck to winter pre-deployment revenue per truck is not a valid measure of the AI's impact; it is a measure of seasonality. The baseline period should match the post-deployment measurement period in time-of-year characteristics.

Breakdown Rate

Definition: Breakdown rate is the number of unscheduled roadside breakdown events per truck per defined period, typically per month or per quarter. A roadside breakdown is any mechanical failure that takes a truck out of service while it is in revenue service (as opposed to a failure detected during a scheduled shop visit). Some fleets refine this to breakdown rate per 100,000 miles driven, which normalizes for differences in mileage across trucks and periods. Either calculation is acceptable; the critical requirement is that the definition of "roadside breakdown" is specific and consistent: does it include tire-only events? Fuel-related stops? The fleet must define the scope and hold it constant.

Why this is the primary safety-and-cost metric: A roadside breakdown is simultaneously a safety event, a cost event, and a service event. The safety dimension is obvious: a truck disabled on the shoulder of I-80 is a hazard to the driver and to other road users. The cost dimension is severe: industry data consistently puts the cost of a roadside breakdown event in the range of $1,000 to $3,000 for a minor event (roadside repair and delays) and $5,000 to $20,000 for a major event (towing, component replacement, missed delivery penalty, driver detention, and substitute load costs). The predictive maintenance economics of approximately 34 percent cost savings on a roughly 44-day payback are achievable precisely because in-shop repair costs a fraction of roadside repair costs: catching the failing wheel-end bearing in the bay costs a few hundred dollars in parts and labor; the same bearing failing on I-80 at mile marker 214 costs $7,000 to $15,000 all in. Breakdown rate is the metric that captures whether the AI's predictive maintenance integration is converting roadside events into scheduled bay events.

The service dimension: Beyond cost, a roadside breakdown triggers a missed or late delivery, which can result in a penalty from the shipper, damage to the customer relationship, and in some cases disqualification from a shipper's carrier list. For a fleet that depends on contract freight or dedicated service agreements, late deliveries from breakdowns are not just operational inconveniences but contract liabilities. The breakdown rate metric should be paired with an on-time delivery impact metric: for each breakdown event in the period, was the associated load delivered late, and if so, was a penalty incurred?

Baseline requirement: The fleet should document its pre-deployment breakdown rate over at least three to six months to capture seasonal variation (winter weather produces more tire and brake events; summer heat produces more cooling-system events). Without a multi-season baseline, the post-deployment comparison may attribute seasonal improvement or worsening to the AI when the causal factor is weather.

CSA Score

Definition: CSA (Compliance, Safety, Accountability) is the FMCSA's safety measurement and enforcement system. It uses data from roadside inspections, crash records, and investigation results to calculate Behavior Analysis and Safety Improvement Category (BASIC) scores for each carrier. There are seven BASIC categories: Unsafe Driving, Hours of Service (HOS) Compliance, Driver Fitness, Controlled Substances/Alcohol, Vehicle Maintenance, Hazardous Materials Compliance (where applicable), and Crash Indicator. Each category accumulates violation points from roadside inspection events; points are weighted by severity and time (recent violations carry more weight than older ones). A carrier whose BASIC score in any category exceeds the FMCSA's threshold for that category is placed in an "alert" status and may be subject to an intervention or investigation.

Why CSA is on the margin-and-safety scorecard: CSA score is the regulatory-facing summary of a fleet's safety posture. Elevated CSA scores in HOS Compliance, Unsafe Driving, or Vehicle Maintenance do not just trigger FMCSA attention. They are visible to shippers, brokers, and 3PLs (third-party logistics providers) who check carrier safety ratings through SMS (Safety Measurement System, the FMCSA's online tool for viewing carrier CSA data) before awarding freight. A carrier with a Vehicle Maintenance alert status loses freight bids before the conversation begins. An AI-integrated predictive maintenance program that keeps the Vehicle Maintenance BASIC score below alert status is not just a safety win; it is a freight-sourcing win. The connection between CSA compliance and revenue is often underappreciated in dispatch-focused AI discussions.

The HOS connection: An AI dispatch system that proposes routes that respect hours-of-service limits (the federal rules, enforced through ELD (electronic logging device) data, that limit commercial drivers to 11 hours of driving in a 14-hour on-duty window and 70 hours on duty in 8 days, among other restrictions) does not just reduce HOS violations for their own sake. It protects the fleet's HOS Compliance BASIC score, which protects the fleet's freight relationships and its operating authority. An AI system that produces efficient routes but does not verify HOS compliance is generating plans that put the carrier's CSA score at risk every time a driver pushes past the legal limit to make the schedule work.

How to track it: CSA BASIC scores are updated monthly by FMCSA and are publicly available through the SMS portal. The fleet's safety manager should pull the fleet's BASIC scores monthly, document them in the dual-axis scorecard, and flag any upward trend in HOS Compliance or Vehicle Maintenance scores for root-cause review before the trend reaches the alert threshold. The goal is not to react to an alert. The goal is to prevent one.

Building the Dual-Axis Scorecard

The dual-axis scorecard is not a software product and it is not a vendor dashboard. It is a reporting structure that pairs the margin metrics and the safety metrics in a single document reviewed on a consistent schedule by the appropriate decision-makers. The format matters less than the discipline of reviewing both axes together every time.

A functional dual-axis scorecard for a fleet AI program has the following structure:

Period covered: Define a consistent period, monthly or quarterly, and stick to it. Changing the measurement window to hide a bad month is the most common way a scorecard loses credibility with an owner.

Margin axis:

• Deadhead percentage (total empty miles divided by total miles, expressed as a percentage) with prior-period comparison and pre-deployment baseline
• Revenue per truck (gross revenue divided by active truck count in the period) with prior-period comparison and pre-deployment baseline
• Loaded-mile rate (revenue divided by loaded miles) with prior-period comparison, to separate rate changes from utilization changes in the revenue-per-truck number
• Cost per mile (total operating cost divided by total miles) with prior-period comparison, to track whether the AI's routing changes are affecting fuel and operational efficiency

Safety axis:

• Breakdown rate (roadside breakdown events per truck or per 100,000 miles) with prior-period comparison and pre-deployment baseline
• CSA BASIC scores (all seven categories) with prior-period comparison and alert-threshold flag
• HOS violation count (logged ELD violations in the period) with prior-period comparison and trend flag
• DVIR (driver vehicle inspection report) defect rate (open defects found at pre-trip inspection divided by inspections conducted) as a leading indicator of vehicle maintenance issues before they reach roadside severity

Narrative summary: A two-to-three sentence summary written by the fleet manager or operations director that characterizes the current period's performance on both axes and identifies any trade-offs. The narrative is where the fleet manager says "deadhead improved by 4 points but breakdown rate rose, suggesting the predictive maintenance integration needs attention." Without the narrative, the scorecard is just numbers. With it, the scorecard is a management document.

Baselining Before Deployment

The most common measurement failure in fleet AI deployments is the failure to establish pre-deployment baselines before the AI goes live. Fleet managers who deploy an AI dispatch tool and then try to reconstruct their pre-AI deadhead percentage from memory or from TMS (transportation management system) exports that were not designed for that query are in a weak position when the owner asks "how much better are we doing?"

The baseline discipline is straightforward but it must happen before deployment, not after. The fleet should pull 60 to 90 days of clean data from the TMS on each of the four margin metrics and each of the safety metrics listed above. It should document that data, freeze it as the baseline, and store it in a form that is retrievable when the comparison is needed. Ideally, the baseline period and the first post-deployment measurement period cover the same months of the year to eliminate seasonal artifacts from the comparison.

For the CSA BASIC scores, the baseline is already available from FMCSA's SMS portal. Pull the current BASIC scores for all seven categories before deployment and save them with the date. This takes ten minutes and provides the safety baseline that makes every subsequent BASIC score comparison defensible.

Who Owns the Metrics

The margin metrics (deadhead percentage, revenue per truck, loaded-mile rate, cost per mile) are typically owned by the dispatch or operations team, which generates the underlying data through the TMS. The safety metrics (breakdown rate, CSA scores, HOS violations, DVIR defect rate) are typically owned by the fleet manager or safety director, who pulls them from the telematics system, the ELD logs, and the FMCSA SMS portal. The dual-axis scorecard should be assembled by one person who owns both axes and presents them together, typically the fleet manager or, in a larger carrier, the director of operations. The key discipline is that the person who owns the margin axis should not be presenting the margin metrics without the safety metrics in the same document, because the two axes tell a single story and separating them creates the conditions for the opening scenario of this lesson.

How the Axes Interact: Trade-offs and Signals

The dual-axis scorecard is valuable precisely because it forces the fleet to look at margin and safety simultaneously, where each axis provides context for the other. The most important interactions are the trade-off signals: the patterns that appear in the scorecard when the margin axis and the safety axis are moving in opposite directions.

The deadhead-improves but breakdown-rate-rises pattern: This pattern typically indicates that the AI is improving load matching and reducing empty miles but that the maintenance workflow is not keeping up with the higher utilization. When trucks are running more loaded miles with fewer gaps, the interval between preventive maintenance events compresses, and if the maintenance schedule is not adjusted to compensate, failure rates rise. The signal the scorecard sends is: adjust the maintenance interval, not the dispatch algorithm.

The revenue-per-truck-improves but CSA-score-rises pattern: This pattern typically indicates that the higher utilization driven by improved dispatch is creating pressure on drivers to push their HOS limits to make the improved schedule work. The AI is proposing tighter plans that legal, rested drivers can execute, but the field execution is adding driver-initiated HOS exceptions. The signal is: the driver coaching program and the HOS compliance check in the dispatch workflow need to be tightened simultaneously.

The everything-improves pattern: When deadhead falls, revenue per truck rises, breakdown rate falls, and CSA scores hold or improve, the AI deployment is working on both axes. This is the result the dual-axis scorecard is designed to make visible and defensible. It is the deck the fleet manager should bring to the owner at 9:00 on Tuesday, and it is credible precisely because it shows both the gains and the safety evidence that the gains are not being purchased at hidden cost.

Leading and lagging indicators: Not all metrics on the scorecard respond at the same speed. Deadhead percentage responds within the first month of an AI dispatch deployment because it reflects current routing decisions. Revenue per truck responds within the first one to two months. CSA BASIC scores respond with a lag of one to three months, because they are calculated from inspection data that accumulates over rolling 24-month periods. Breakdown rate can improve quickly if the AI's predictive maintenance integration is catching near-term failure risks, but more deeply embedded wear-pattern improvements take two to four months of consistent maintenance practice to show up in the data. The fleet manager who reviews only the first-month metrics and concludes "the AI is not working" may be looking at a lagging indicator that has not had time to move. The scorecard review cadence must account for indicator lag.

The Math Behind the Metrics

Concrete numbers make the dual-axis scorecard credible to an owner. Here are the calculations that appear on the scorecard, worked through a representative fleet scenario.

Scenario fleet: 38 active trucks. Pre-deployment period: 90 days. Post-deployment period: first 90 days after AI dispatch deployment.

Deadhead percentage change:

• Pre-deployment: 38 trucks x 2,400 average weekly miles = 91,200 total weekly miles. Deadhead miles documented at 22,800 per week. Deadhead percentage = 22,800 / 91,200 = 25.0 percent.
• Post-deployment: total weekly miles 94,500 (slight increase from better utilization). Deadhead miles 17,010. Deadhead percentage = 17,010 / 94,500 = 18.0 percent.
• Change: 7 percentage points of improvement.

Revenue per truck change:

• Pre-deployment loaded miles: 91,200 x 0.75 = 68,400 loaded miles per week. At $2.20 average loaded-mile rate: $150,480 per week fleet revenue. Revenue per truck = $150,480 / 38 = $3,960 per truck per week.
• Post-deployment loaded miles: 94,500 x 0.82 = 77,490 loaded miles per week. At $2.20: $170,478 per week. Revenue per truck = $170,478 / 38 = $4,486 per truck per week.
• Change: $526 per truck per week, or $19,988 per week fleet-wide. Over 52 weeks: approximately $1,039,000 per year in recovered revenue from better utilization alone, before any improvement in loaded-mile rates.

Breakdown cost savings:

• Pre-deployment: 3.2 roadside breakdown events per month across the fleet. Average cost per event: $4,800 (blended across tire-only, mechanical, and major events). Monthly breakdown cost: $15,360.
• Post-deployment (with predictive maintenance integration active): 1.1 events per month. Monthly breakdown cost: $5,280.
• Savings: $10,080 per month, or approximately $121,000 per year.

CSA Vehicle Maintenance BASIC score:

• Pre-deployment: 68 (approaching the alert threshold of 80 for the Vehicle Maintenance BASIC).
• Post-deployment after 90 days: 52, due to a reduction in out-of-service defect violations at roadside inspection events. The predictive maintenance program is catching and correcting defects before they appear at inspection.
• Business impact: the fleet is no longer approaching the Vehicle Maintenance alert threshold that was creating friction in shipper bids.

These are the numbers the owner needs to see on Tuesday morning. They are only available if the baseline was documented before deployment and the dual-axis scorecard was built to capture both families of metrics.

Key Takeaways

• A fleet AI deployment measured only by volume and throughput metrics is flying with half its instruments; the dual-axis scorecard pairs margin metrics (deadhead percentage, revenue per truck) with safety metrics (breakdown rate, CSA score) in a single governance document reviewed together every period.
• Deadhead percentage is the primary margin metric because it directly measures the conversion of the fleet's scarcest resource (driver-hours) from wasted empty miles to paying miles; a 7-point reduction in deadhead on a 38-truck fleet can recover over $1 million per year in revenue at typical loaded-mile rates.
• Revenue per truck is the asset-utilization metric that tells the real story of AI dispatch performance; it rises when better load matching puts more paying miles on each truck, regardless of whether loaded-mile rates have changed.
• Breakdown rate measured in events per truck per month captures both the safety cost and the financial cost of roadside failures; at $1,000 to $20,000 per event and approximately 34 percent cost savings available through predictive maintenance, this metric is often the fastest ROI item on the scorecard.
• CSA (Compliance, Safety, Accountability) BASIC scores are the regulatory-facing safety summary that connects fleet compliance to freight revenue; carriers with elevated Vehicle Maintenance or HOS Compliance BASIC scores lose shipper bids before the conversation starts, making CSA score management a margin issue, not just a safety issue.
• Pre-deployment baselines for all metrics must be documented before the AI goes live; reconstructing the baseline after deployment from memory or inconsistent TMS queries produces unverifiable comparisons that an owner or auditor will not accept.
• The trade-off patterns (deadhead improves but breakdown rate rises; revenue per truck rises but CSA HOS score worsens) are the signals the dual-axis scorecard is designed to surface; they indicate that one workflow is outrunning another and allow the fleet to intervene before the problem compounds.
• CSA BASIC scores lag by one to three months; breakdown rates may lag by two to four months as the predictive maintenance program matures; only deadhead percentage and revenue per truck respond within the first 30 days, which means the scorecard review cadence must be patient enough for the safety indicators to reflect the operational changes.