What a Driver’s First Thirty Days Actually Cost, and Which Part is Dispatch’s Fault

The plan was built for a driver who does not exist yet

Day four. A driver two weeks out of induction is given a route of 82 stops. The route was sized from throughput data, the service times came from the planning model, and the sequence is optimal against both.

By end of the shift they had completed 68. Fourteen stops roll to tomorrow. The scorecard the following morning shows 83% completion against a network benchmark of 96%, which places them in the bottom decile, which triggers a conversation with a supervisor about whether they are picking things up.

Nothing about that driver was measured. What was measured was the gap between a plan calibrated on people with two years of experience and a person with two weeks of it.

This is the part of new-driver cost that belongs to dispatch rather than to recruitment or training, and it is worth separating out, because the recruitment and training parts get attention and this one is usually invisible. The driver’s inexperience is real and it is not a defect. The plan treating inexperience as absent is the defect.

Also Read: Driver Onboarding and Scheduling Software Guide 2026

Four parameters calibrated on tenured performance

Each of these is set from historical actuals, and historical actuals are dominated by experienced drivers because experienced drivers do most of the work.

Service time per stop. The largest single input. If the model says a residential drop takes six minutes, that figure came from a population that knows where to park, which bell to ring, and which building has the awkward gate. A first-month driver takes longer on the same stop for reasons that have nothing to do with effort, and the difference compounds across eighty stops.

Route size. Stops per route is derived from what a route can absorb in a shift, which means from tenured throughput. Handing a full-size route to a week-one driver is not an aggressive target. It is an infeasible one, and infeasible plans produce the failure modes infeasible plans always produce.

ETA models. Predicted arrival times are trained on how routes actually run, which again means how tenured drivers run them. A new driver’s actual progress diverges from prediction early in the day and the divergence grows, so every downstream ETA on that route is wrong and getting wronger.

Performance benchmarks. The network median completion rate, stops per hour, or on-time percentage is a tenured median. Comparing a two-week driver against it is comparing them against a cohort they are not in.

The pattern across all four is the same. Each parameter is sensible, derived from real data, and correct on average. Averaged across a workforce, the new-driver population is small enough not to move the number, which is exactly why the number does not describe them.

Also Read: Best Driver Management Software for Delivery Fleets 2026

What the miscalibration actually costs

Five costs, and only the first is the one operation usually discussed.

Rolled stops and next-day recovery. Fourteen stops that did not happen have to be absorbed somewhere, usually into tomorrow’s plan, which was already full. The cost is the recovery, not the individual stops.

ETA error on new-driver routes. This is the sharpest one and it is almost never reported this way. If ETA accuracy is materially worse on routes driven by first-month drivers, then your customer experience varies by driver tenure, and the customer has no idea which kind of day they were assigned. Notification quality, WISMO contact rate, and promise adherence all move with a variable nobody is measuring.

Overtime. A new driver given an 82-stop route does not complete it faster. They complete more of it by working longer. The route was infeasible in the shift, so the plan converts inexperience into paid hours rather than into rolled stops, or into both.

Supervisor time. Coaching conversations about a performance problem that is a calibration problem. The supervisor is not wrong to have the conversation given what the scorecard says, and the conversation cannot fix what is causing the number.

Attrition risk. A driver who is late every day, told they are underperforming, and given no reason to believe it will change is a driver considering other work. The earnings and predictability dimension of that is a substantial argument in its own right and is not restated here.

Also Read: Driver Routing and Scheduling: What Enterprises Get Wrong

Why the feedback loop makes it worse

There is a second-order problem that keeps the ramp invisible even in operations with good analytics.

When a new driver’s actuals arrive, one of two things happens.

They are treated as outliers and excluded, because they sit far from the distribution and including them would degrade the model. Reasonable, and the consequence is that the model never learns what a week-two driver actually does, so the ramp curve is never estimated and route sizing for new drivers has no basis to improve.

Or they are blended in with everything else, which drags the model slightly toward inexperience without representing either population well. Slightly worse for everyone and still no ramp curve.

The missing ingredient in both cases is a tenure dimension in the service-time model. Almost no operation holds one. The consequence is not just that new drivers get bad plans. It is that the operation cannot answer a basic question about itself: how long does it take a driver to reach normal productivity, and what does that curve look like.

Without that curve, route sizing for new drivers is guesswork, onboarding length is guesswork, and the cost of turnover cannot be calculated properly because the ramp is the largest part of it.

Two fixes that do not work

Once the ramp problem is recognized, two responses come up first and neither addresses it.

Extending classroom onboarding. More induction days do not close this gap, because the gap is not knowledge that can be taught in a room. It is route-specific familiarity accumulated by repetition, plus a plan calibrated on someone else. A driver can leave a longer induction knowing more and still be given an 82-stop route on day four.

Lowering the benchmark globally. Dropping the network completion target so new drivers are not flagged does remove the false signal, and it also removes the true signal for everyone else. Tenured underperformance becomes invisible at the same time. The correction belongs in the comparison group, not in the threshold.

There is a structural reason this matters more in US last-mile than the effort involved would suggest. Where turnover runs high, the newest cohort is not a seasonal intake that passes through. It is a permanent standing share of the roster, replenished continuously. An operation with meaningful churn is therefore always planning for a workforce mix that its model does not represent, every week, rather than during an onboarding season.

Uncalibrated and tenure-aware planning compared

Dimension	Uncalibrated planning	Tenure-aware planning
Service time source	One model, tenured-dominated	Separate model per tenure cohort
Route size for a new driver	Full, from day one	Ramped over defined weeks
ETA confidence	Uniform across drivers	Adjusted by driver tenure
Notification width	Same for all routes	Wider where confidence is lower
Performance comparison	Network median	Cohort benchmark
New-driver actuals	Excluded or blended	Used to fit the ramp curve
Visible failure	Driver underperformance	Ramp progress against expectation
Cost absorbed as	Rolled stops, overtime, coaching, churn	A planned, shorter ramp

The row that changes the others is the last but one. Under uncalibrated planning the failure presents as a person, which routes the response to coaching and eventually to replacement. Under tenure-aware planning the same data presents as progress against an expected curve, which routes the response to route sizing.

What tenure-aware planning looks like

Five changes, in order of how much they return for the effort.

Ramp route size deliberately. Set stops per route by tenure cohort for the first weeks, then step up. The lost capacity is real and it is smaller than the rolled stops, overtime, and recovery a full route generates in week one. This is the single highest-return change and it requires no new data.

Hold service-time models per cohort. Weeks one to two, three to six, seven to twelve, and tenured. Four models rather than one, fitted from actuals you already collect once you stop discarding them.

Adjust ETA confidence by tenure, and widen the promise accordingly. If a route’s arrival predictions are less reliable, the customer-facing window should reflect that rather than asserting the same precision. A slightly wider window that holds is better than a narrow one that breaks.

Benchmark against the cohort. A week-two driver should be compared against week-two drivers, and the useful question is whether they are progressing along the curve rather than whether they have reached the end of it.

Consider composition, not just size. The hardest geography and the most access-difficult stops are the least suitable work for someone who has never seen them. This interacts with route attractiveness and with the value of local knowledge, both of which are arguments in their own right, and the practical version here is simply not to hand week-one drivers the routes tenured drivers avoid.

Also Read: Rider Management at Scale: Running Large Fleets

What to measure

Completion rate by tenure cohort, week by week. This is the ramp curve, and producing it for the first time usually surprises people. Most operations have the data and have never grouped it this way.

Service time actual against planned, by cohort. The input that makes route sizing improvable rather than intuitive.

ETA accuracy by driver tenure. The customer-facing metric, and the one that turns this from a driver issue into a delivery experience issue with a named cause.

Rolled stops attributable to cohort. How much of your daily recovery burden originates in plans that were infeasible for the person assigned.

Time to parity. Weeks until a cohort reaches 95% of tenured completion rate. This is the outcome number. It makes onboarding investment, route sizing, and turnover cost comparable on one axis, and almost nobody has it.

Also Read: The Real Cost of Manual Dispatch in North American 3PLs

How Locus supports tenure-aware dispatch

Locus, the world’s first Decision-Intelligent, Agentic TMS, treats driver state as a first-class input to allocation rather than as a payroll attribute, which is the prerequisite for planning differently for different cohorts. Its DiSCO framework, the Digital Supply Chain Officer, runs specialized agents across the lifecycle on a continuous Sense-Decide-Execute-Learn cycle against a model of more than 250 real-world constraints.

Three properties map onto the changes above.

The Capacity agent holds the roster with live per-driver state, including availability, remaining hours, skills, and certifications. Because driver attributes participate in allocation rather than sitting alongside it, a tenure cohort can act as a constraint on route size and composition rather than as a note a planner is expected to remember.

Route size and composition are expressible as constraints. A cap on stops for a defined cohort, or a limit on access-difficult stops per route, is configuration rather than a manual override, which is what makes a ramp policy survive contact with a busy morning.

The Learn stage consumes actuals. Because the cycle closes, completion and service-time data feed back into planning rather than into reporting, which is the mechanism that turns a year of new-driver actuals into a ramp curve instead of a year of discarded outliers.

Locus’s rider and driver management capabilities cover the roster, allocation, and driver application together, which matters here because the ramp is visible only when planning data and execution data sit in the same system.

Locus has processed more than 1.5 billion deliveries for 360-plus enterprise customers across 30-plus countries at 99.99% uptime. It is 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. Further analyst recognition is published in full.

Two deployments show cohort-scale driver operations.

A Fortune 50 parcel and logistics provider centralized dispatch across 51 sites in a 120-country network against a driver pool of 4,500 split between captive and third-party capacity. Weekly execution rate rose from 75% to 92%. At that pool size, a continuous intake of new drivers is a permanent condition rather than an event, so execution rate is partly a measure of how well plans fit the actual mix of experience on the road on any given week rather than the mix the model assumes.

A global FMCG operation across ten countries coordinates more than 1,000 distributors and a rider pool exceeding 5,000, reaching over 1.8 million retail outlets, saving more than 12,000 trips a month at 3X ROI. A rider pool of that size with normal churn means a substantial share of the roster is always inside its first months, which makes cohort-aware allocation an operating requirement rather than a refinement.

Request a Locus driver ramp assessment to produce your completion curve by tenure cohort, compare planned against actual service time for your newest drivers, and size a ramp policy against what rolled stops and overtime currently cost you.

Produce the curve before you change anything

One query, from data you already hold, will tell you whether this is worth acting on.

Pull twelve months of completed routes. Tag each by the driver’s tenure in weeks at the time. Then plot median completion rate, median stops per hour, and median finish time against tenure.

You will get a curve. Three things are worth reading off it: where it flattens, which is your true ramp length; how far below tenured performance week one sits, which tells you how badly a full route mis-sizes; and whether the curve has changed year over year, which tells you whether anything you have done to onboarding has worked.

Then compare that curve against the route sizes you are currently issuing in week one. In most operations the gap is large, unexamined, and being paid for in rolled stops and overtime rather than in a slightly shorter route.