Driver Management Under AB5 and Prop 22: A Workforce Architecture Playbook for California Last-Mile Operators

A note before the architecture

Worker classification in California is fact-specific, actively litigated, and changes. Nothing here is legal advice, and no operating model is inherently compliant. The tiers below describe how operators commonly structure capacity and where the classification exposure sits in each, so that you can have a more informed conversation with employment counsel, not so that you can skip one.

Two specifics worth raising with counsel early: AB5 contains a business-to-business exemption with a multi-condition test that may or may not apply to your contractor relationships, and app-based network protections apply to qualifying platforms under defined conditions rather than functioning as a status an operator can adopt.

Why California driver management is structurally harder

AB5’s ABC test presumes a worker is an employee unless the hiring entity establishes all three prongs. Prong B is the one that binds last-mile operators specifically: the work performed must be outside the usual course of the hiring entity’s business.

For a courier network, a regional parcel carrier, or an enterprise D2C fleet, delivery is the usual course of business. That makes engaging individual 1099 drivers to perform deliveries structurally difficult to defend, regardless of how the relationship is documented, because the test asks what the work is rather than what the contract says.

This lands on operators already under workforce pressure. US Bureau of Labor Statistics data shows annual separation rates in transportation and warehousing regularly exceeding 40 percent, with last-mile and package delivery segments running higher, and the ATA estimates a US driver shortage of roughly 60,000, projected above 170,000 by 2030. Flexibility is not a preference in this market; it is how operations cover the day.

The result is the dilemma the architecture below addresses: how to hold flexible, scalable capacity without building it on a classification foundation that will not hold.

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

The three-tier driver management architecture

Rather than a single driver pool, operators partition capacity into three tiers with different characteristics and different exposure profiles.

Two corrections to how this model is usually described.

The W2 tier is not merely the compliant option, it is the operationally correct one for the volume it covers. Baseline urban density is predictable, and predictable volume is where employed capacity is cheapest per stop and where control over customer experience matters most.

And the 3PL tier is not a firewall. Outsourcing to a commercial operator materially changes the analysis, but “completely insulated” overstates it, and the arrangement’s substance matters more than its label.

The economics support the split independently of compliance. The US Postal Regulatory Commission has found average cost per delivery in rural areas runs approximately twice that of urban areas, which is a strong argument for pushing low-density work to partners with existing density on those lanes rather than building your own.

Also Read: The End of the “Captive Fleet Only” Era: Orchestrating Hybrid Last-Mile Capacity in 2026

What the driver management platform has to do

A tiered workforce without a dispatch layer that understands the tiers produces the worst of both: fragmented execution and a structure that does not hold up under examination. Three platform requirements.

Tier-aware allocation. Predictable, high-density, brand-sensitive volume should route to W2 capacity by default, with surge and volatile volume flowing to flexible tiers. This has to be allocation logic rather than a dispatcher’s habit, because habits are inconsistent and inconsistency is what an examiner looks for.

The alternative, sizing employed capacity for peak, is expensive in both directions. McKinsey has found that static planning models can leave as much as 60 percent of operating hours either understaffed or overstaffed.

Differentiated control by tier. How work is offered is part of the classification analysis. Non-employee capacity should receive work as discrete proposals that can be declined, without mandated schedules or prescribed methods, while employed drivers can be dispatched directly with sequence and method specified. A platform that applies identical control logic across all three tiers erodes the distinction the architecture depends on.

This is the requirement most dispatch tools cannot express, because they were built on the assumption that a driver is a driver.

Unified visibility without unified control. Customer service needs one view of every delivery regardless of who is carrying it: live status, ETA, and proof of delivery across W2, app-based, and 3PL work. Visibility is not control, and consolidating the view does not require consolidating the direction.

Also Read: AI-Powered Rider and Driver Management Software

Five ways operators undermine their own structure

Treating the tiers as interchangeable in dispatch. If surge volume routinely goes to the same non-employee drivers on the same routes at the same times, the arrangement starts to resemble the thing it was structured to avoid.

Applying employee-style performance management to non-employee capacity. Acceptance-rate requirements, mandated shift blocks, and prescriptive route adherence rules all cut against the independence the structure assumes.

Documenting the structure without operating it. The contract describes a business-to-business relationship while dispatch practice looks like supervision. Practice is what gets examined.

No audit trail of allocation decisions. When asked why a specific batch went to a specific driver, “the system decided” is not an answer. Decision records matter here for the same reason they matter in any regulated automated decision.

Letting the architecture drift. Tier boundaries set in January erode through a peak season of exceptions. Review allocation patterns by tier quarterly against what was designed.

Also Read: Best Last-Mile Delivery Company for Driver Management in 2026: A Software-First Guide

Where the platform layer fits

Locus, the world’s first Decision-Intelligent, Agentic TMS, models workforce type as a constraint alongside vehicle capability, geography, and service commitment, so allocation respects tier boundaries by design rather than by dispatcher discipline.

Within DiSCO, the Capacity agent maintains the roster across employed and contracted pools, the Carrier agent handles allocation to third-party and network capacity, and the Dispatch agent plans and re-sequences. Six governance mechanisms bound autonomous action, including explainability, traceability, autonomy levels, and human-in-the-loop override. Traceability is the one that matters most in this context: every allocation decision leaves a record of why that work went to that pool, which is exactly what an operator needs when the question is asked months later.

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 US deployments show the mechanisms. A Fortune 50 parcel and logistics provider governs 4,500+ drivers, 1,500+ captive and 3,000+ third-party, under one policy across 51 service-center locations, with zone-based routing for captive shifts and tendering plus on-demand assignment for third-party carriers running inside the same decision engine. Weekly execution moved from 75 percent to 92 percent, and every autonomous decision is logged for explainability, traceability, and human override. That is a mixed workforce allocated by different logic per pool without splitting the operation across systems.

A global lottery operator running a US field-service network across 25+ states models each state’s contracts, labor laws, SLA windows, zones, and skills as live constraints drawn from the 250+ the platform holds per computation. That is the same requirement California imposes, applied across jurisdictions: labor rules as planning constraints rather than as a compliance review after the plan is built.

Execution speed is part of the compliance picture too. Gartner found that while 95 percent of supply chains must react quickly to change, only 7 percent can execute decisions in real time. Operators who cannot reallocate quickly tend to solve surges by pressing whichever capacity is nearest, which is how tier boundaries erode.

Also Read: Agentic Driver Management for Enterprise-Scale Last-Mile Delivery

Questions to take to counsel

Bring the operating reality, not the org chart.

1. Which of our contractor relationships, if any, could satisfy the business-to-business exemption conditions, and what would we need to change for that to hold?
2. Does the platform capacity we use qualify under app-based network provisions, and what is our exposure if a platform’s status changes?
3. Where does our dispatch practice exert control over non-employee capacity that our contracts say we do not?
4. What records do we hold showing how work was allocated across tiers, and would they support our characterization of these relationships?
5. What in our current peak-season practice differs from our designed architecture?

The fifth question surfaces more risk than the other four combined, because peak is when structures get bypassed and the bypass becomes the precedent.