One Service Time, Two Workforces: Why Route Plans Overload New Drivers

Driver management covers recruiting, scheduling, allocating and retaining the people who execute deliveries, and tenure is usually filed under retention. It belongs in planning. A driver who has completed two thousand stops in a territory is measurably faster at the same addresses than one who started last month, and route plans are almost universally built on a single service-time assumption that describes neither of them. The result is a plan that fits the average of a workforce that is not distributed around its average.

On a 35-stop route the gap between a tenured driver and a new one is 87 minutes, which is more than a sixth of a shift. Locus, the world’s first Decision-Intelligent, Agentic TMS, plans against more than 250 real-world operating constraints, and service time by driver and territory is one of the few inputs that changes plan feasibility more than routing quality does.

Why one average describes nobody

The learning effect here is well established outside logistics. Theodore Wright set it out in 1936 in Factors Affecting the Costs of Airplanes, observing that labor requirements fell 10 to 15% for every doubling of cumulative production, a pattern since confirmed across semiconductors, solar panels and batteries. The part that matters for driver management is Wright’s distinction: cumulative experience drives the curve, not elapsed time. A driver is not fast because they have been employed for six months. They are fast because they have completed a large number of stops in that specific territory, which is why rotating a tenured driver to a new patch resets much of the advantage.

Build a route on the average and the arithmetic is unforgiving. Take an eight-hour shift with 200 minutes of driving, leaving 280 minutes for stops. At an assumed eight minutes a stop, that is a 35-stop route which fits the shift precisely.

The spread on the identical route is 87.5 minutes. The new driver is 52 minutes over, which is five or six stops undelivered or that much overtime. The tenured driver finishes 35 minutes early, which is capacity the operation paid for and did not use.

How much of the workforce sits in the slow group depends on churn, and the relationship is simple. At a 90-day ramp, the steady-state share below competence is annual churn multiplied by the ramp as a fraction of the year.

Last-mile churn sits at the upper end of that range in many operations, so one driver in five is being handed a plan calibrated for somebody else. Scale it to a hundred routes a day at 80% churn and the two errors appear simultaneously: twenty new drivers running 52 minutes over is 17.5 hours of overtime or failed stops, while eighty tenured drivers finishing 35 minutes early is 46.7 hours of capacity bought and unused. The same assumption overloads one group and underloads the other, on the same day, in the same depot.

Much of this workforce is not yours to manage, which changes what you can do about it. Armstrong & Associates put US 3PL gross revenues at $323.4 billion in 2025 against net revenues of $138.2 billion, and the American Trucking Associations reports almost 580,000 active US motor carriers as of June 2025, of which 91.5% operate 10 or fewer trucks. Where a provider supplies the driver, their tenure and rotation policy are theirs, so you inherit whatever experience distribution they happen to have, without visibility into it and without the ability to change it. The plan still has to fit those drivers, which makes measuring service time by provider a procurement input as much as a planning one.

Cost sharpens both halves. ATRI’s 2026 report puts the industry-average cost of operating a truck at $2.336 per mile in 2025, a record for the series and 3.4% above the prior year, so unused vehicle hours are expensive and overtime is more expensive still.

The loop that makes it worse

The consequence is not merely inefficiency. It is a feedback loop that manufactures the churn the plan assumed.

A new driver receives a route calibrated for a tenured one. They run late, because the plan was never achievable at their current service time. Late running produces missed windows, which are measured as on-time performance, which is attributed to the driver rather than to the plan. The driver is coached on a problem they did not create, works unpaid or unrecognized overtime to close the gap, and concludes within a few weeks that the job is harder than it was described. They leave earlier than they otherwise would. Churn rises, which raises the share of drivers in ramp, which makes the average assumption less representative, which overloads the next cohort slightly more.

Nothing in that sequence requires anybody to behave badly. It runs on a single reasonable-looking parameter and a performance metric pointed at the wrong party.

The loop is also self-accelerating rather than self-correcting. Every driver who leaves early is replaced by somebody starting at the bottom of the curve, so the share of the workforce in ramp rises, the network mean service time drifts slower, and the gap between the plan’s assumption and the slower cohort widens. An operation can therefore watch its on-time performance decline year on year while its routing engine, its plans and its people are all working exactly as designed.

How to make tenure a planning input

1. Measure service time by driver and by territory, not as a single mean

Most operations hold one service-time figure per stop type. Hold a distribution instead, broken down by driver and territory, and the bimodality usually appears immediately. This requires no new data collection in an operation already capturing arrival and departure timestamps, only a different query.

Two cautions on the measurement. Strip out stops where the delay was external, such as a recipient who could not be found or a site with a queue, because those inflate a driver’s apparent service time without saying anything about their competence. And segment by stop type before comparing drivers, since a residential parcel and a multi-case retail drop are different tasks and a driver weighted toward the harder mix will look slow for reasons that have nothing to do with experience.

2. Count cumulative stops on the territory rather than days employed

Following Wright, the variable that predicts speed is experience rather than time. A driver six months into the job who has rotated across five territories is a new driver in each of them. Track stops completed per driver per territory and use that as the competence measure, because it is the one that actually forecasts service time.

3. Plan to the driver where the assignment is known in advance

Where the route is allocated before it is built, or where allocation is stable week to week, build the route to that driver’s service time. A 30-stop route for a new driver and a 38-stop route for a tenured one both fill a shift, and both are achievable, which is the point. A single 35-stop route is achievable for one of them.

4. Where the assignment is not known, plan to the achievable rather than the average

If any driver might take any route, the average is the wrong parameter because it fails for the slower half. Planning nearer the slower end costs the tenured driver some idle time and removes the systematic overload, which is the cheaper of the two errors: unused capacity is visible and recoverable, while a failed stop is neither.

5. Separate plan failure from driver failure in the metric

On-time performance currently attributes an infeasible plan to the person executing it. Report plan achievability alongside it, computed as route duration at that driver’s measured service time against the shift. A route that was never achievable should not appear in a coaching conversation, and knowing which routes those are is a five-minute calculation once step one exists.

6. Protect the ramp deliberately

If a fifth of the workforce is in ramp at any moment, ramp is a permanent operating condition rather than an onboarding phase. Give new drivers shorter routes on stable territories for a defined period, measure their curve, and expand as the data supports it. The alternative is discovering the ramp through attrition.

Planning to the average and planning to the driver

The right-hand column is not more sophisticated routing. It is the same routing with a parameter that reflects who is driving. No new engine, no new data capture, and the change is a query before it is a project.

Five questions about tenure in your planning

Do you hold one service time or a distribution? If it is one number, the plan fits the average of a workforce that is not distributed around its average.

Is competence measured in days employed or stops completed on the territory? Only the second predicts service time, and only the second survives a driver being moved.

What share of your drivers are inside their ramp right now? Churn multiplied by ramp length as a fraction of a year. At 80% and 90 days it is one in five.

Can you tell an infeasible plan from an underperforming driver? If on-time is the only metric, the answer is no, and coaching conversations are being held about the wrong thing.

What happens to a tenured driver’s spare 35 minutes? It is either unused or absorbed informally. Both are choices, and neither is usually a deliberate one.

What this looks like in enterprise deployments

A Fortune 50 parcel operation running centralized dispatch across a 120-country network moves more than a million freight shipments a year across 51 sites with a 4,500-strong driver pool split between captive and third-party. At that pool size, a single service-time assumption is describing several thousand people with very different territory experience, and the deployment lifted weekly execution adherence from 75% to 92%. Adherence is the right metric to watch here, because a plan calibrated for the wrong driver shows up as a plan that was not followed rather than as a plan that was wrong.

A global FMCG network running logistics automation across 10 Asian countries works through 1,000-plus distributors and 5,000-plus riders, saving 12,000-plus trips a month across 1.8 million retail outlets. Rider-based last-mile networks at that scale typically run the highest churn in the industry, which means the share of the workforce inside its ramp is structurally high and permanent, not a transitional problem waiting to resolve.

Four mistakes operations make about driver tenure

Filing tenure under HR. Retention programs address why drivers leave. They do not change the plan those drivers are being given, and the plan is part of why they leave.

Measuring experience in months. Wright’s finding is that cumulative output drives the curve, not elapsed time. Months on payroll overstates the competence of anybody who has been rotated.

Coaching on on-time without checking achievability. If the route needed 532 minutes at that driver’s measured service time and the shift is 480, the conversation is about arithmetic rather than effort.

Treating ramp as temporary. At realistic churn, a fifth of the workforce is permanently in ramp. Designing for it once is cheaper than absorbing it continuously.

How Locus handles service time as a real variable

Locus, the world’s first Decision-Intelligent, Agentic TMS, plans against more than 250 real-world operating constraints, and service time is one of them rather than a fixed global parameter. Because the route planning system produces dispatch-ready plans in roughly two minutes and re-optimizes continuously, a route can be built to the driver who will actually run it rather than to a network mean, and rebuilt when the allocation changes.

The Driver Companion App carries sequenced tasks and captures arrival, departure and exception events, which is the source data a per-driver service-time distribution needs. Explainability and Traceability record the trigger, context, reasoning, action and outcome of each decision, so plan achievability can be evaluated after the fact against what the driver’s measured service time actually was, rather than against the assumption the plan used. That distinction is what separates a plan failure from a driver failure in a performance review.

Two boundaries belong here. Locus does not set your service-time parameters, and a platform given one global figure will plan to one global figure, so the value depends on an operation choosing to measure and maintain the distribution. And tenure is not the only driver of service time: stop type, access difficulty, customer behavior and vehicle all matter, and attributing everything to experience will produce a model that fits badly. The honest position is that experience is one significant and currently ignored term in a function with several.

Locus supports more than 360 enterprise customers across 30-plus countries, with over 1.5 billion deliveries optimized, more than $320 million in documented client logistics savings and 99.99% uptime. It has been recognized by Gartner for seven consecutive years, featured in the 2026 Hype Cycle for Supply Chain Execution and Logistics Technologies, named a Leader in TMS by QKS Group (SPARK Matrix), and ranked #1 in Route Planning on G2’s 2026 Best Software Awards.

In October 2025, Ingka Investments, the investment arm of Ingka Group, the world’s largest IKEA retailer, acquired Locus. Locus continues to operate independently.

So why do route plans systematically fail for new drivers? Because they are built on a single service-time average that describes neither end of the workforce. A 35-stop route at an assumed eight minutes a stop fits an eight-hour shift exactly, leaves a tenured driver 35 minutes spare at seven minutes a stop, and puts a new driver 52 minutes over at nine and a half, an 87-minute spread on the identical route. At 80% annual churn with a 90-day ramp, one driver in five is in that slower group permanently, so a hundred-route day carries 17.5 hours of overtime alongside 46.7 hours of unused capacity from the same assumption. Worse, it is self-reinforcing: the overloaded driver runs late, is measured on on-time, is coached for a plan failure, and leaves sooner. Following Wright’s 1936 finding that cumulative experience rather than elapsed time drives the learning curve, the right competence measure is stops completed on that territory. Locus supports this by treating service time as one of more than 250 planning constraints, capturing arrival and departure events through the Driver Companion App, and recording outcomes per decision so plan achievability can be separated from driver performance. Request a Locus assessment to measure your own service-time distribution.

Frequently Asked Questions

How much slower is a new driver than a tenured one? Enough to break a plan. On a 35-stop route, a tenured driver at seven minutes a stop finishes 35 minutes inside an eight-hour shift, while a new driver at nine and a half minutes runs 52 minutes over. That is an 87-minute spread on the identical route with the identical plan.

Why does planning on the average not work? Because the workforce is not distributed around its average. The average plan is achievable for exactly the group at the average and fails progressively for everyone slower, which at realistic churn is one driver in five.

How many of our drivers are below competence right now? Annual churn multiplied by ramp length as a fraction of a year. At 40% churn and a 90-day ramp it is 10%; at 80% churn it is 20%. Last-mile operations commonly sit at the upper end.

Should competence be measured in months or in stops? In stops completed on that territory. Wright’s 1936 work established that cumulative experience rather than elapsed time drives the learning curve, so a driver six months in who has rotated across five territories is a new driver in each of them.

How does this create churn rather than just cost? The overloaded driver runs late, misses windows, is measured on on-time performance, and is coached for a plan failure they did not cause. They work unrecognized overtime, conclude the job is harder than described, and leave sooner, which raises the share of drivers in ramp and worsens the assumption for the next cohort.

What should we do if routes are allocated dynamically? Plan nearer the achievable end rather than the average. It costs the faster drivers some idle time and removes the systematic overload, and unused capacity is visible and recoverable while a failed stop is neither.

How do we separate a bad plan from a bad driver? Report plan achievability alongside on-time performance, calculated as route duration at that driver’s measured service time against the shift length. A route requiring 532 minutes in a 480-minute shift was not a performance problem, and it should not appear in a coaching conversation.