---
title: "Your Dispatch Plan is an Offer, Not an Assignment"
id: "26556"
type: "post"
slug: "driver-allocation-offer-not-assignment-2026"
published_at: "2026-09-11T15:30:00+00:00"
modified_at: "2026-09-13T18:56:31+00:00"
url: "https://locus.sh/blogs/driver-allocation-offer-not-assignment-2026/"
markdown_url: "https://locus.sh/blogs/driver-allocation-offer-not-assignment-2026.md"
excerpt: "When drivers can decline, rejection is not random and the residual routes are the worst ones. The offer cascade, and why the plan stabilizes 40 minutes late."
taxonomy_category:
  - "General"
---

#### [General](https://locus.sh/blogs/category/general/)

# Your Dispatch Plan is an Offer, Not an Assignment

[Anas T](/author/anas_locus/)

Sep 11, 2026

15 mins read

Driver management covers how an operation recruits, schedules, allocates and retains the people who execute deliveries, and allocation is the part most often modeled wrongly. A route optimizer treats allocation as an assignment problem: given routes and drivers, produce the best pairing. That model is correct when drivers are employees on fixed shifts executing what they are given. It is wrong in gig, flexible, owner-operator and much contracted work, where the output of the optimizer is a proposal somebody can decline.

Once refusal is possible, the problem changes category. It becomes a two-sided matching market in which both parties have preferences, and the plan is not final when the engine finishes. It is final when the workforce agrees. Locus, the world’s first Decision-Intelligent, Agentic TMS, produces dispatch-ready plans in roughly two minutes against more than 250 real-world operating constraints, and in an operation where drivers can refuse, the two minutes is the start of the allocation rather than the end of it.

## Key Takeaways

- Allocation with refusal is a two-sided matching market, not an assignment problem. The two have different solutions.
- Rejection is not random. The routes left unaccepted are the least attractive by construction, so the residual pool degrades every round.
- Filling the last fifth of a plan costs 2.00 offers per route against 1.25 for the first 80%.
- The optimizer finishes at T plus two minutes and the plan stabilizes around T plus 40, because agreement takes rounds.
- The fix is in the offer, not the route: bundling, sequencing, and giving the engine an acceptance probability per driver and route.
- What drivers weigh is knowable and stable: earnings, distance from home, stop density, time of day, vehicle, and whether that route ran long last time.
- Locus re-plans in roughly two minutes, so a refusal can be answered with a re-optimized plan rather than a manual reshuffle.

## Why refusal changes the problem

An assignment problem has one set of preferences, the planner’s. A matching market has two, and the mathematics has been settled since 1962, when Gale and Shapley introduced [deferred acceptance and the concept of a stable matching](https://en.wikipedia.org/wiki/Stable_marriage_problem)
, proving a stable outcome always exists in a finite two-sided market. The 2012 Nobel Memorial Prize went to Shapley and Roth for the theory of stable allocations and the practice of market design, and Roth’s applied work covered real labor markets, including medical residency matching, where participants on both sides rank each other and a naive assignment produces outcomes people walk away from.

Last-mile allocation has the same structure and is rarely treated that way. The planner ranks drivers by fit. Drivers rank routes by earnings, distance from home, difficulty, vehicle, and whether the last one ran long. Both rankings exist. Only one is in the software.

What drivers weigh is knowable and fairly stable. Total earnings for the shift comes first, but rarely alone. Distance from home to the start point and back is unpaid time and weighs heavily. Stop count and drop density determine how hard the day feels rather than how long it is. Time of day interacts with everything else a person has scheduled. Vehicle type matters where the driver supplies the vehicle. And reputation follows a route: a run that went 90 minutes over last month is remembered, and one bad experience suppresses acceptance on that route long after the cause was fixed. None of those is mysterious, and almost none of them appears in a routing constraint set.

| Also Read: Rider Management at Scale: Running Large Fleets |
| --- |

The structural consequence is adverse selection in the residual pool. Offers go out, the attractive routes are taken, and what remains is not a random sample of the plan. It is the least attractive fraction, so the next round faces a lower acceptance rate on a worse set.

| Round | Routes offered | Acceptance | Filled | Still open |
| --- | --- | --- | --- | --- |
| 1 | 100 | 80% | 80.0 | 20.0 |
| 2 | 20 | 50% | 10.0 | 10.0 |
| 3 | 10 | 50% | 5.0 | 5.0 |
| 4 | 5 | 50% | 2.5 | 2.5 |
| 5 | 2.5 | 50% | 1.2 | 1.2 |

Two features of that table are worth separating. The first is the acceptance drop between round one and round two, from 80% to 50%, which is not a modeling choice but a consequence of what remains. The second is that the tail never fully closes: each round halves the gap without reaching zero, so an operation relying on re-offering alone is running an open-ended process and calling it a plan.

Across those rounds the operation makes 139 offers to place 99 routes. The first 80 routes cost 1.25 offers each. The remaining nineteen cost 2.00 each, because every one of them has already been declined at least once and nothing about it has improved. At eight minutes a round, the plan is not stable until roughly forty minutes after the optimizer produced it.

That gap is the operationally important number. The engine reports a plan at T plus two minutes. Dispatch is still working the tail at T plus forty, and the routes settled last are covered by whoever remained rather than by the best match, so service quality correlates with fill order in a way no plan-quality metric captures.

Cost makes the tail expensive rather than merely slow. ATRI’s 2026 report puts the industry-average cost of operating a truck at [$2.336 per mile in 2025](https://truckingresearch.org/2026/07/new-atri-report-details-accelerating-costs-and-low-profitability-despite-cuts/)
, a record for the series and 3.4% above the prior year, and a route covered by a poor match runs longer than the plan assumed.

The workforce this applies to is larger than the gig framing suggests. The American Trucking Associations reports [almost 580,000 active US motor carriers](https://www.trucking.org/economics-and-industry-data)
 as of June 2025, of which 91.5% operate 10 or fewer trucks, and an owner-operator deciding whether to take your load is making exactly this choice. Armstrong & Associates put [US 3PL gross revenues at $323.4 billion in 2025](https://www.3plogistics.com/3pl-market-info-resources/3pl-market-information/us-3pl-market-size-estimates/)
, most of which moves through parties that accept or decline rather than comply.

## How to allocate when the answer can be no

### 1. Measure acceptance rate by route attribute, not in aggregate

A single acceptance percentage tells you nothing actionable. Break it down by the attributes drivers actually weigh: total earnings, distance from the driver’s start point, stop count, time of day, vehicle type, and whether the route has a history of running over. The output is a model of what gets declined, which is the input everything below needs.

Two measurement notes. Record declines rather than only acceptances, because an operation that logs the final assignment has no record that three people said no first. And separate a decline from a non-response, since silence usually means the offer was never seen and has a completely different fix from a route that was read and rejected.

### 2. Give the engine an acceptance probability per driver and route

Once acceptance is predictable it becomes a constraint like any other. An allocation that maximizes plan quality while ignoring a 30% refusal probability is optimizing something that will not happen. Weighting by expected acceptance produces a slightly worse plan on paper that is materially better after the workforce responds, which is the only comparison that matters. The comparison to run is plan quality at T plus two minutes against plan quality at T plus forty, for both approaches. The acceptance-weighted version usually loses the first and wins the second, and only the second is executed.

### 3. Design the offer, not just the route

The most effective intervention is usually bundling. Where geography permits, pairing an unattractive route with an attractive one converts two separate decisions into one, and the combined offer clears where the unattractive half alone would not. Sequencing matters too: releasing every route simultaneously lets drivers cherry-pick, while staged release with short response windows gives the operation control over the order in which the pool depletes.

| Also Read: Driver Retention: Why the Operational Layer Beats Bonuses |
| --- |

### 4. Price the residual instead of re-offering it

After two declines, a route has revealed something about itself. Continuing to re-offer it at the same terms is the definition of expecting a different result, and each round costs eight minutes of dispatch time. A standing uplift that triggers automatically on the third offer converts an open-ended cascade into a bounded cost, and knowing that cost per route is more useful than not knowing how long the tail will take.

### 5. Run deferred acceptance rather than first-come allocation

First-come allocation gives routes to whoever responds fastest, which is uncorrelated with fit. Deferred acceptance, as Gale and Shapley set out, holds offers rather than confirming them immediately, so a better match arriving later can still be made. In operational terms that means a short collection window in which responses accumulate before anything is confirmed, rather than confirming on first reply. Ninety seconds is usually enough to change the outcome materially, because the fastest responder and the best match are different people more often than not, and the cost of waiting is far smaller than the cost of a poor pairing that runs long all day.

### 6. Track fill order as a service risk

Record which routes settled last and check whether they underperform. They usually do, because they were covered by the residual pool rather than by the best available match. That correlation is invisible in plan-quality reporting and it identifies precisely which routes need redesigning rather than re-offering. Keep the list short and act on it quarterly: a route that settles last three months running is not unlucky, it is badly shaped, and redesigning ten of those removes more cascade than any change to the offer process will.

## Assignment and matching are different problems

|  | Assignment model | Matching model |
| --- | --- | --- |
| Assumes | Drivers execute what is allocated | Drivers accept or decline |
| Preferences represented | Planner’s only | Both sides |
| Plan final at | Optimizer completion | Agreement, several rounds later |
| Failure mode | Refusals treated as exceptions | Handled as expected behavior |
| Residual routes | Assumed comparable | Known to be the least attractive |
| Right lever | Better routing | Better offer design |
| Fits | Employed drivers, fixed shifts | Gig, flexible, owner-operator, contracted |

Most enterprise operations run a mixed workforce and a single allocation model, which means the assignment model is being applied to the half of the fleet that can refuse. The result is not a broken plan. It is a plan that quietly takes forty minutes longer to become real.

## Five questions about allocation in a workforce that can decline

**What is your first-round acceptance rate?** If nobody measures it, the offer cascade is invisible and its cost is absorbed as normal dispatch work.

**How long after the optimizer finishes is the plan stable?** This is the honest planning cycle time, and it is usually many times the number the vendor quotes.

**Which route attributes predict refusal?** Without this you can re-offer but you cannot redesign, and redesigning is the only durable fix.

| Also Read: Best Driver Management Software for Delivery Fleets |
| --- |

**What happens on the third decline?** If the answer is a fourth offer, the cascade has no bound. A triggered uplift or a redesign rule gives it one.

**Do last-filled routes underperform?** Test the correlation. If it holds, fill order is a leading indicator of service failure that nobody is currently reading.

## What this looks like in enterprise deployments

A Fortune 50 parcel operation running [centralized dispatch](https://locus.sh/case-studies/fortune-50-parcel-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. That split is exactly the mixed-model condition described above: the captive half executes what it is allocated and the third-party half decides. Running one allocation model across both means one of them is modeled wrongly, and the deployment lifted weekly execution adherence from 75% to 92%, which is the metric that moves when allocation starts surviving contact with the people performing it.

A global FMCG network running [logistics automation](https://locus.sh/case-studies/global-fmcg-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. At that rider count, small changes in acceptance behavior dominate small changes in route quality, because a route nobody takes has no quality.

## Four mistakes operations make when drivers can refuse

**Treating refusal as an exception.** It is expected behavior in a market with two sets of preferences. Logging it as an exception puts it in a queue rather than in the model that produced it.

**Re-offering unchanged.** A declined route carries information. Offering it again at the same terms discards that information and spends eight minutes doing so.

**Allocating first-come.** Speed of response is uncorrelated with fit, so first-come systematically trades match quality for match speed on every route it settles.

**Reporting optimizer completion as planning time.** It is the moment a proposal exists. Quoting it as cycle time understates the real planning window by the length of the cascade, which is where most of the elapsed time lives, and it sets an expectation the operation then fails against every single morning.

## How Locus handles allocation across a workforce that chooses

Locus, the world’s first Decision-Intelligent, Agentic TMS, plans against more than 250 real-world operating constraints and allocates across owned fleet, contracted transporters and a network of more than 1,000 carriers, decided per shipment against live cost and serviceability. That breadth matters here because a refusal at one tier can be answered by sourcing at another rather than by re-offering into a depleted pool.

The mechanism that does most of the work is re-planning speed. Because the [route planning system](https://locus.sh/route-planning-system/)
 re-optimizes in roughly two minutes and runs continuously, a decline is a trigger for a new plan rather than a hole in the existing one, which is structurally different from patching a fixed plan by hand. The Driver Companion App carries sequenced tasks and captures response, so acceptance and refusal become recorded events rather than phone calls. Explainability and Traceability record the trigger, context, reasoning, action and outcome of each allocation decision, which is what allows acceptance rate to be analyzed by route attribute rather than estimated.

Two boundaries belong here. Locus does not set your rates, and where refusal is a pricing signal rather than a fit signal, the answer is commercial and belongs to you. And the platform models acceptance only from the data you capture, so an operation that handles declines by telephone and records the final state has no acceptance history to learn from, which is a process change before it is a configuration one.

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](https://locus.sh/analyst-recognition/)
, 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.

| Also Read: Crowdsourced Delivery: When It Works and When It Fails |
| --- |

So how should an operation allocate work to drivers who can say no? As a matching market rather than an assignment problem, because both sides have preferences and only one of them is usually in the software. Refusal is not random, so the routes left after the first round are the least attractive by construction, and acceptance falls on each subsequent pass: 139 offers to place 99 routes, with the first 80 costing 1.25 offers each and the last nineteen costing 2.00. The plan the optimizer finished at T plus two minutes is not stable until around T plus forty. The levers sit in the offer rather than the route, through bundling, staged release, a triggered uplift after the second decline, and deferred acceptance instead of first-come confirmation. Locus supports this by re-planning in roughly two minutes so a decline produces a new plan rather than a gap, allocating across owned, contracted and 1,000-plus carrier capacity so a refusal can be sourced elsewhere, and recording each allocation decision so acceptance can be analyzed by route attribute. [Request a Locus assessment](https://locus.sh/schedule-demo/)
 to measure your own offer cascade.

## Frequently Asked Questions

**Why is driver allocation a matching problem rather than an assignment problem?** Because both parties have preferences. The planner ranks drivers by fit and drivers rank routes by earnings, distance, difficulty and history. An assignment model represents only the planner’s ranking, which works for employed drivers on fixed shifts and fails wherever refusal is possible.

**Why does acceptance fall in later rounds?** Because the unaccepted routes are the least attractive by construction. Nothing about them improved between offers, and the pool of available drivers has also shrunk, so the second round faces a worse set of routes and a smaller audience simultaneously.

**How much does the cascade actually cost?** On a 100-route plan at 80% first-round acceptance, it takes around 139 offers to place 99 routes. The first 80 cost 1.25 offers each and the remaining nineteen cost 2.00 each. At eight minutes a round the plan stabilizes roughly forty minutes after the optimizer finished.

**What is the most effective single fix?** Bundling unattractive routes with attractive ones where geography permits, so the two become one decision. It converts a route that clears at 40% into part of an offer that clears far higher, without changing the rate.

**Should we just pay more for the routes nobody takes?** Eventually, and deliberately rather than by attrition. A standing uplift that triggers automatically on the third offer bounds the cascade and makes the cost visible per route, which is more useful than an unbounded sequence of re-offers absorbed as dispatch time.

**What is deferred acceptance in an operational setting?** A short window in which responses accumulate before anything is confirmed, so a better match arriving thirty seconds later can still be made. It is the alternative to first-come allocation, which trades match quality for response speed on every route it settles.

**Does this apply to employed drivers?** Less directly, though most enterprise operations run a mixed workforce and a single allocation model. The risk is applying the assignment model to the portion of the fleet that can decline, which is where the unmeasured forty minutes comes from.

MEET THE AUTHOR

Anas T

Senior Content Writer - Product Marketing

Anas is a product marketer at Locus who enjoys turning complex logistics problems into simple, clear stories. Outside of work, he’s usually unwinding with a book or catching a good movie or series.

### Related Tags:

[https://locus.sh/blogs/logistics-automation-slow-revert-2026/](https://locus.sh/blogs/logistics-automation-slow-revert-2026/)
#### [Logistics Automation & Orchestration](https://locus.sh/blogs/category/logistics-automation-orchestration/)

## [The Slow Revert: How Logistics Automation Goes Back to Manual Without Anyone Deciding](https://locus.sh/blogs/logistics-automation-slow-revert-2026/)

[Aseem Sinha](https://locus.sh/blogs/author/aseem_locus/)

Sep 11, 2026

Relaxations accumulate because disabling is easy and re-enabling is not. After two years, a third of the ruleset is off while the dashboard still reads 85%.

[Read more](https://locus.sh/blogs/logistics-automation-slow-revert-2026/)

[https://locus.sh/blogs/driver-tenure-routing-input-2026/](https://locus.sh/blogs/driver-tenure-routing-input-2026/)
#### [General](https://locus.sh/blogs/category/general/)

## [One Service Time, Two Workforces: Why Route Plans Overload New Drivers](https://locus.sh/blogs/driver-tenure-routing-input-2026/)

[Ishan Bhattacharya](https://locus.sh/blogs/author/ishan_locus/)

Sep 11, 2026

The same 35-stop route runs 87 minutes apart depending on who takes it. Why planning on an average overloads new drivers and creates the churn it assumes.

[Read more](https://locus.sh/blogs/driver-tenure-routing-input-2026/)

## Your Dispatch Plan is an Offer, Not an Assignment

- Share
- [Print](javascript:window.print())
- [Download](#)
- [Schedule a Demo](https://locus.sh/schedule-demo/)

### Is your team spending more time on fixing logistics plan than running the operation?

- Agentic transportation management from order intake to freight settlement
- Route optimization built on 250+ real-world constraints
- AI-driven dispatch with automatic execution handling

20%Cost Reduction

66%Faster Planning Cycles

[Schedule a demo](/schedule-demo/)

Insights Worth Your Time

#### [General](https://locus.sh/blogs/category/general/)

## [Locus 2026 UK Consumer Survey: Why Returns Visibility is Now the Conversion Engine for AI-Driven Shopping in UK Retail](https://locus.sh/blogs/returns-visibility-conversion-engine-ai-shopping-uk-retail-locus-q2-2026-consumer-survey/)

[Aseem Sinha](https://locus.sh/blogs/author/aseem_locus/)

May 29, 2026

#### [General](https://locus.sh/blogs/category/general/)

## [Locus 2026 US Consumer Survey: Generative AI isn’t Just Changing How Consumers Shop, it’s Breaking the Demand Patterns US Retail Was Built On](https://locus.sh/blogs/generative-ai-shopping-effect-retail-fulfillment-operations-locus-q2-2026-consumer-survey/)

[Ishan Bhattacharya](https://locus.sh/blogs/author/ishan_locus/)

May 29, 2026

#### [General](https://locus.sh/blogs/category/general/)

## [Embedded vs Bolted-On AI: The Architecture Question European Logistics Buyers Are Asking](https://locus.sh/blogs/embedded-vs-bolted-on-ai-european-logistics-platform-architecture-business-benefits/)

[Aseem Sinha](https://locus.sh/blogs/author/aseem_locus/)

May 21, 2026

#### [General](https://locus.sh/blogs/category/general/)

## [Hybrid Fleet Management: How Owned, 3PL, Gig, ICE, and EV Capacity Actually Operate at Most Enterprises](https://locus.sh/blogs/three-workforce-fleet-reality-owned-3pl-gig-drivers/)

[Aseem Sinha](https://locus.sh/blogs/author/aseem_locus/)

May 7, 2026

#### [General](https://locus.sh/blogs/category/general/)

## [US Returns Hit $850 Billion in 2025: Why US Retailers Are Restructuring Reverse Logistics in 2026](https://locus.sh/blogs/850-billion-us-returns-ai-routing-reverse-logistics-2026/)

[Ishan Bhattacharya](https://locus.sh/blogs/author/ishan_locus/)

May 7, 2026