Why New Drivers Quit Before They Learn the Route: Earnings Variance is a Dispatch Decision

The pay review that changes nothing

The standard response to driver churn is a compensation review. Benchmark the market, adjust the rate, improve the benefits, and expect retention to follow. Operations run this exercise, spend real money on it, and frequently lose the same drivers on the same timeline.

The reason is that the rate and the problem are different variables. A rate determines what a driver earns per unit of work. It says nothing about how much work arrives, how consistently, or how that compares to what the driver next to them received. A driver who tells you they earned well one week and badly the next is not describing a pay problem. They are describing an allocation problem, and allocation is a dispatch output.

This is not a soft argument. The National Academies consensus study on long-distance trucking states it directly: more variability in driver earnings on a weekly basis is associated with higher turnover. The same work names uncertain load availability affecting weekly earnings, and limited predictability in dispatching, among the primary causes of departure. It also finds that effective frontline dispatchers correlate with lower turnover, which is the same finding from the other direction.

The stakes are not marginal. ATRI’s cost data puts driver compensation at roughly 44% of operating cost, the largest single line, and the American Trucking Associations project a shortage of around 60,000 unfilled driver seats in 2026, growing toward 160,000 to 175,000 by 2028 on current trends. Replacing a driver is expensive in a market where the replacement may not exist.

The natural experiment inside your own industry

The most useful number in the National Academies work is not the turnover rate. It is the spread between segments.

Those figures cover roughly 1996 to 2023 and describe the same national labour market. The drivers are drawn from the same pool, hold the same licences, and face the same fuel prices and the same regulations. Truckload turnover runs almost eight times LTL linehaul.

The differences that plausibly explain that gap are not primarily about pay level. LTL linehaul work is scheduled, repeated, and predictable: the same lane, the same hours, a knowable week. Truckload work is assigned load by load, so the week is constructed dynamically and the driver finds out what they earned by living it. Private fleets, at 15%, share LTL’s predictability characteristics rather than its pay scale.

That is as close to a controlled comparison as this industry offers, and it points at predictability rather than rate as the dominant variable. Which locates the lever in dispatch.

Also Read: The Three-Workforce Fleet Reality: How Owned, 3PL, and Gig Drivers Actually Operate at Most Enterprises

Where the variance actually comes from

Six dispatch design choices generate most week-to-week earnings variance. None of them is a pay decision, and all of them are usually made without reference to earnings at all.

The fourth row deserves attention because it is the least visible. ATRI found drivers were detained at 39.3% of all stops in 2023, losing between 117 and 209 hours per year depending on sector. Where that time is unpaid or paid below the driver’s effective rate, detention is a pay cut distributed unevenly by route, and the routes with the worst dwell tend to go to the drivers with the least choice.

Why the first 90 days absorb it worst

Identical variance produces different attrition depending on tenure, for three reasons that have nothing to do with commitment or work ethic.

No buffer. An experienced driver has banked good weeks and can absorb a bad one. A three-week driver has no reserve, so a low week is not an inconvenience but a bill that does not get paid.

No baseline. A two-year driver experiencing a poor week knows it is a poor week. A new driver has no distribution to compare against, so a bad week is evidence about the job rather than evidence about the week. This is the mechanism that turns normal variance into a resignation.

No workarounds. Tenured drivers have learned which dispatcher to ask, which routes to request, when to make themselves available, and how the allocation actually works in practice. That accumulated knowledge is a variance-reduction strategy the new driver does not yet have. The National Academies work notes that new-to-industry drivers quit at substantially higher rates than experienced drivers, and this is a large part of why.

The practical consequence is that variance and tenure interact rather than adding. An operation with acceptable average variance can still have unacceptable variance for its newest cohort, and the aggregate number will not show it.

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

The equity and efficiency tension nobody names

Here is the uncomfortable part, and it is a genuine trade-off rather than an oversight.

Allocating the best routes to the highest performers is rational. It maximises drops per hour, lowers cost per drop, and rewards the drivers who earn it. Most operations do some version of this, and the better their performance data, the more precisely they do it.

Learn here how a rider/driver management software can improve driver experience.

It also compounds. A driver who receives dense, well-sequenced routes posts better numbers, which qualifies them for more of the same. A driver who receives the sparse routes and the difficult addresses posts worse numbers, which confirms the original assessment. After a few months the allocation is defensible on the data and the data is partly a product of the allocation. The bottom half of the roster is being managed out by a system nobody designed to do that.

The resolution is not equal allocation, which would waste real skill differences. It is a floor rather than a split: a minimum standard of assignment quality below which no driver falls, applied as a constraint alongside cost. That costs something in raw efficiency and buys retention in the cohort where replacement is most expensive. Stating the trade explicitly is better than making it accidentally.

What to measure

Five measures, and the first is the one almost nobody reports.

Earnings variance per driver, not average earnings. Track the week-to-week spread for each driver, expressed as a coefficient of variation. An operation with a healthy average and a wide distribution has a retention problem its reporting is hiding.

Variance by tenure band. Split at 0 to 30, 31 to 60, 61 to 90, and 90-plus days. If variance is higher for the newest band, allocation is loading risk onto the drivers least able to absorb it.

Route density spread per driver. The operational cause behind the earnings pattern, and the thing dispatch can actually change.

Attrition by tenure band, plotted against variance. This is the correlation that makes the argument internally. If your 0 to 90 day attrition tracks your 0 to 90 day earnings variance, the case for changing allocation stops being a theory.

Share of drivers whose worst week fell below a viability threshold. Set the threshold from local cost of living rather than from your pay scale. One unviable week is what triggers a job search, and averages never show it.

Also Read: How AI Dispatch and Allocation Works for Freight Carriers: A Practical Guide (2026)

The California complication

In most of North America, earnings variance is a retention problem. In California it is also a compliance calculation.

Under Proposition 22, app-based delivery and transportation companies owe qualifying drivers an earnings guarantee tied to a multiple of minimum wage for engaged time, plus a per-mile allowance. AB5’s classification framework sits underneath it. The practical effect is that a week where allocation produced little work does not simply produce a disappointed driver; it produces a shortfall the operator may be obliged to make up.

That changes the economics of variance. Where a guarantee exists, unpredictable allocation transfers cost to the operator directly rather than to the driver, so smoothing allocation stops being a retention investment and becomes a cost-control measure. Operations running mixed classifications across states need allocation logic that is aware of which rules apply to which driver, which is a dispatch constraint rather than a payroll adjustment.

How Locus treats allocation as a constraint rather than an outcome

Locus, the world’s first Decision-Intelligent, Agentic TMS, makes allocation an explicit decision with modelled constraints rather than a by-product of route optimisation.

Within DiSCO, the Capacity agent maintains the full roster with live state on availability, remaining hours, and qualifications, and matches capacity to demand across owned, contracted, and on-demand pools. Because the roster is held as state rather than recalculated per route, cumulative allocation across a period is visible to the system, which is the prerequisite for treating equity as a constraint rather than an afterthought. The Dispatch agent plans and re-sequences against more than 250 real-world constraints per computation, so a minimum assignment-quality floor, a cap on consecutive low-density routes, or a jurisdiction-specific earnings rule can be expressed as constraints alongside cost rather than applied by a dispatcher’s judgement. Six governance mechanisms bound autonomous action, including explainability, which matters here specifically: a driver who asks why they received a given route should be answerable with a reason rather than a shrug.

Locus 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.

Two North American deployments show allocation governed as policy rather than improvised.

A Fortune 50 parcel and logistics leader ran a 4,500-strong driver pool split across captive fleets on zone-based routing and third-party carriers requiring tendering and on-demand assignment, with no single tool unifying them. That structure is exactly where allocation inequity hides, because different pools operated under different logic and nobody could see the whole. Locus deployed Capacity and Carrier agents to govern the entire pool under one policy, running zone-based, tendering, dynamic, and on-demand assignment inside a single decision engine. Weekly execution across 51 service-centre locations moved from 75% to 92%, and a single-site analysis surfaced $565,000 in unused capacity, including premium-tier service given away on cheaper classes. Unallocated capacity and inequitable allocation are two readings of the same data.

A global lottery operator running field service across more than 25 US states shows allocation under differing labour rules. Contracts, labour laws, and revenue terms differed by state, six job types each demanded different skills, and rosters shifted through the day, so every assignment was a three-way match of case, skill, and location. Locus models each state’s contracts, labour laws, SLA windows, zones, and skills as live constraints rather than post-plan checks, with the Capacity agent maintaining the roster and the Dispatch agent assigning against it. The operator reported SLA penalty risk down 20%, fuel spend down 18%, and drive distance and time down 15%.

Also Read: Best AI Dispatch and Allocation Platforms for Logistics Carriers in 2026

Measure the spread before you move the rate

The next time driver churn appears on an agenda, ask for a different number. Not average earnings, and not the market benchmark, but the week-to-week earnings spread for drivers in their first ninety days.

If that spread is wide, a pay increase will raise the average and leave the mechanism intact. The drivers will still be unable to plan a month, and the ones with the least buffer will still leave first. If the spread is narrow and churn is still high, the problem genuinely may be the rate, and the compensation review is the right response.

That single diagnostic decides which lever to pull, and it costs nothing but a query. The broader point is that driver retention has an operations owner as well as an HR owner, and the National Academies finding that dispatcher effectiveness correlates with lower turnover is a reasonable place to start the conversation.

Book a Locus demo to review how allocation is currently distributed across your roster, and where a fairness floor could be applied without losing route efficiency.