Managing Drivers You Do Not Employ in 2026: What the Delivery Service Partner (DSP) Average Hides

Contracted driver management is the problem of holding service accountability for drivers employed by somebody else. In North America this is the dominant last-mile structure: delivery service partners, contracted courier fleets and franchise operators carry the volume while the brand carries the promise and the customer relationship. The instinct is to manage these partners on their aggregate numbers, and the aggregate turns out to be both easy to measure and close to useless for deciding what to do. Two partners reporting identical performance can require completely different interventions worth thirteen times different amounts. Locus, the world’s first Decision-Intelligent, Agentic TMS, holds driver-level qualification, history and allocation in one constraint set whether the driver is employed, contracted or gig.

Why the Contracted Model Changes the Problem

Almost every technique that improves driver performance operates at the level of the individual. Tenure-aware routing needs to know who is new. Coaching needs to know who to coach. App adoption work needs to know who is working around the app. Ramp support needs to know who is on a ramp.

The contracted structure removes exactly that layer. The partner employs, trains, schedules and pays the driver. The brand receives outcomes, usually aggregated to the partner and often to the week. So the operation holding the customer promise has the least granular view of the thing that determines whether the promise is kept.

The economics make this the expensive place to be blind. McKinsey’s out-of-home delivery work puts the last mile at 60% to 70% of total parcel delivery cost, and ATRI’s 2026 operational cost report put marginal operating cost at $2.336 per mile in 2025. A failed first attempt is a second visit over that cost base, and it is produced by a driver whose performance you are managing through a contract rather than through a system.

Conditions compound it. INRIX’s 2025 Global Traffic Scorecard found congestion increased in 254 of the 290 US cities it analyzed, which widens the gap between a driver who knows a territory and one who does not, precisely the difference the aggregate conceals.

Also Read: Best Driver Management Software for Delivery Fleets

The Aggregate is Easy to Measure and Hard to Use

We started by testing whether partner-level measurement is noisy, expecting that small partners would be hard to judge fairly. That turned out to be wrong, and the result is worth stating because it removes a common excuse.

A 15-driver partner running 95 stops per driver-day across 22 working days produces 31,350 deliveries a month. At a 3.0% baseline failure rate, that volume reports between 2.81% and 3.19% in 95% of months. A partner genuinely running 20% worse than baseline is detectable in well under a month. Identical partners essentially never produce a large apparent gap by chance.

So partner scorecards are statistically sound. The problem is not that the number is unreliable. It is that the number does not describe anything you can act on.

Consider two partners, both 30 drivers, both reporting almost exactly 3.0%.

	Uniform partner	Concentrated partner
Aggregate failure rate	3.03%	3.11%
Best driver	2.3%	1.6%
Median driver	3.0%	2.0%
Worst driver	4.0%	11.3%
Share of all failures from the worst 3 drivers	12%	34%

The second partner has a better median driver than the first and a worse headline number, because four people are carrying a third of its failures. On a scorecard the two look the same. Operationally they have almost nothing in common, and the action that would help each one is different in kind rather than in degree.

The Same Intervention, Worth Thirteen Times More

The reason this matters is that it determines what to do, and the answer differs completely.

Distribution	Aggregate	Reduction in failures if the worst 10% are brought to median	Resulting aggregate
Uniform	3.01%	2%	2.94%
Concentrated	3.04%	27%	2.22%

In the uniform partner, targeting the worst drivers is close to pointless. Everyone is performing similarly, so the improvement has to come from something systemic: the routes, the vehicles, the territory, the plan. In the concentrated partner, the same targeted intervention removes more than a quarter of all failures.

Thirteen times the return, from an identical action, at an identical headline number. A program that treats every partner the same will waste effort on the first and leave the second’s largest available gain untouched.

The honest implication for how these relationships are run is uncomfortable. Quarterly business reviews conducted on aggregate performance can be entirely accurate and still direct attention at the wrong thing, because the aggregate answers whether a partner is underperforming and never answers why.

Also Read: Driver Tenure as a Routing Input

Which Drivers You Can Actually Identify

Driver-level data does not solve everything either, and the limits are specific rather than general.

Driver’s true failure rate	Relative to baseline	Deliveries needed to establish it	Working days
3.6%	1.2x	13,898	146
4.5%	1.5x	2,514	26
6.0%	2.0x	748	8
9.8%	3.3x	202	2

A driver performing at twice the baseline is identifiable within about eight working days. A driver at three times baseline shows up inside two weeks. Those are the drivers carrying the concentrated partner’s failures, and they are findable quickly.

A driver 20% worse than baseline needs most of a year of deliveries before the difference is distinguishable from noise. That driver is effectively unmanageable at the individual level, and pretending otherwise produces scorecards that rank people on randomness.

That asymmetry is the practical design rule for this whole problem. Driver-level data is worth having because it finds severe outliers fast, and those outliers carry a disproportionate share of failures in exactly the partners where targeted intervention pays. It is not worth building a system that ranks the middle of the distribution, because the middle cannot be ranked honestly at delivery volumes.

How to Manage Contracted Driver Performance

1 Negotiate for driver-level outcome data, not driver identity

Most partners will resist sharing employee records and are right to. What is negotiable is anonymized driver-level outcomes: a stable pseudonymous identifier with stop outcomes attached. That is enough to find the outliers and does not require the partner to hand over personnel data.

2 Ask for the distribution before asking for the average

The single most useful change to a partner review is requesting the spread of driver performance rather than the mean. A partner that cannot produce it is telling you something about its own visibility into its people.

3 Classify each partner as uniform or concentrated

Then choose the intervention accordingly. Concentrated partners need targeted support for a handful of drivers. Uniform partners need the route, the territory or the plan examined, because the drivers are not the variable.

4 Look for outliers only, and act quickly

Set the detection threshold where the statistics support it, at roughly double baseline, and act within weeks. Do not attempt to rank drivers who differ by 20%, because that ranking is noise dressed as performance.

5 Check whether the outlier is the driver or the route

A driver consistently assigned the hardest territory will look like a poor performer. Before escalating to the partner, confirm the pattern survives when route difficulty is held constant, or the intervention lands on the wrong person.

6 Give partners the planning inputs that lower the rate at source

Tenure-aware routing, accurate service times and access instructions reduce failures regardless of who employs the driver. A brand that shares those with its partners improves outcomes it cannot directly manage.

Put the Distribution in the Contract

Most of what this article recommends is a data question, and data questions in contracted relationships are settled at renewal rather than in a review meeting. That makes the contract the practical instrument, and it is usually written without any of this in it.

Three clauses do most of the work. The first is a reporting obligation for the distribution rather than the mean: failure rate by driver, anonymized, at an agreed cadence. Partners rarely object to this once it is clear that personnel data is not being requested, and the ones who do object are frequently the ones who cannot produce it.

The second is an agreed outlier threshold and a response window. Naming the level at which a driver is treated as an outlier, and what happens next, converts an awkward conversation into a process both sides agreed to in advance. Setting it at roughly double baseline keeps it defensible, because that is where the statistics support a conclusion within two weeks.

The third runs the other way and matters more than it looks. A commitment from the brand to supply planning inputs, accurate service times, access instructions and tenure-aware route construction, gives the partner something in return and addresses the largest share of failures that are produced by the plan rather than by people. A partner asked for more data and given nothing will comply slowly. One given inputs that make their drivers’ days easier has a reason to engage.

None of the three require a change in who employs anyone, which is the point. They are the version of driver management that is available across a commercial boundary.

Employed Fleet and Contracted Fleet Driver Management Compared

Dimension	Employed drivers	Contracted or DSP drivers
Who holds the employment relationship	The brand	The partner
Granularity of available data	Individual, continuous	Usually partner-level, often weekly
Tenure visibility	Known	Rarely shared
Coaching path	Direct	Through the partner, with their agreement
What a scorecard can support	Individual performance management	Partner selection and volume allocation
Realistic intervention	Targeted per driver	Outlier escalation plus better planning inputs
Biggest blind spot	Few	The distribution beneath the average

The fifth row is the one to design around. A partner scorecard is a commercial instrument for deciding who gets volume, and it is good at that. It was never an operational instrument for improving performance, and using it as one is why partner reviews so often end with agreement that something must improve and no specific action.

Also Read: Fleet Utilization for 3PLs Across a Multi-Client Operation

What to Look for in a Platform That Spans Employed and Contracted Capacity

One driver record regardless of employer. Employed, contracted and gig drivers should resolve into the same data model, so that allocation, qualification and history are read the same way. Where contracted drivers are a separate integration, they will be a separate process.

Anonymized driver-level outcome capture. The platform should be able to hold a stable pseudonymous driver identifier from a partner feed and attach stop outcomes to it. This is the minimum needed to find outliers without requesting personnel data.

Distribution reporting, not just averages. Ask whether the platform can show the spread of driver performance within a partner. If it reports only partner means, it cannot support the distinction this article is built on.

Route difficulty normalization. Outlier detection is only fair when territory difficulty is held constant, otherwise the system will reliably flag the drivers given the hardest work.

Planning inputs that reach the partner. Service times, access instructions and tenure-aware route construction should be shareable with contracted operators, because lowering the failure rate at source is the one lever that works without any visibility into individuals at all.

Also Read: Peak Season Driver Management: The Real Constraint

Mixed Employment in Action

A Fortune 50 parcel enterprise ran more than a million freight shipments a year across a 120-country network, with a 4,500-strong driver pool split between captive and third-party capacity across 51 sites. That split is the structure this article describes, at the largest scale it occurs. Through centralized dispatch, weekly execution rose from 75% to 92% and more than $14M in unused annualized capacity was identified, including $565K at a single site, at 99.99% uptime.

The execution figure is the relevant one. Moving from 75% to 92% across a pool that is partly employed and partly contracted is not something achieved by managing employees harder, because most of the pool is not employed. It came from planning and allocating across both populations on the same basis.

A Canadian grocery brand delivering fresh and perishable orders across more than 30 cities ran its last mile entirely through contracted third-party fleets, so the promise belonged to the retailer and the drivers did not. After carrier orchestration, deliveries ran 33% faster, fulfillment cost fell 15% and customer support resolution ran 10 to 20 times faster.

Both cases point at the same mechanism rather than at a management technique. Neither operation gained control over drivers it did not employ. Both improved the quality of the plan those drivers executed, which is the lever that works through a contract.

Common Mistakes in Contracted Driver Management

Managing partners on the average. The aggregate is accurate and tells you nothing about what to fix, because two identical averages can conceal distributions with thirteen times different improvement potential.

Ranking drivers who differ by 20%. At delivery volumes that difference takes most of a year to establish. A scorecard that ranks the middle of the distribution is ranking noise and will be resented accordingly.

Escalating an outlier without checking the route. A driver given the hardest territory every day will look like the problem. Normalizing for route difficulty before escalating avoids acting on the wrong person.

Treating visibility as the only lever. Better planning inputs lower the failure rate regardless of who employs the driver, and they are available immediately without any contractual change.

How Locus Approaches Drivers You Do Not Employ

Locus, the world’s first Decision-Intelligent, Agentic TMS, holds employed, contracted and gig drivers in one allocatable pool with one data model, which is what makes rider and driver management work across an employment boundary rather than stopping at it. Qualifications, availability and service history sit in the same constraint set the route planning engine solves against, evaluated across more than 250 real-world operating constraints, so a contracted driver is planned for on the same basis as an employed one.

That structure is what makes the planning lever usable. Tenure-aware route construction, learned service times and access instructions can be applied to work executed by a partner, which lowers the failure rate at source without requiring visibility into that partner’s employment decisions. The Dispatch agent holds allocation across the combined pool, and DiSCO governance mechanisms including Explainability and Traceability mean an allocation or an escalation can be traced back to the observations behind it, which matters when the conversation is with a commercial partner rather than an employee.

Locus has been recognized by Gartner for seven consecutive years across multiple research categories, including Representative Vendor status in the 2026 Gartner Hype Cycle for Supply Chain Execution and Logistics Technologies and the 2026 Gartner Market Guide for Multicarrier Parcel Management Solutions, where ShipFlex is featured as a Representative Vendor. QKS Group positions Locus as the Leader in its SPARK Matrix for Transportation Management Systems 2025, and G2 ranked Locus number one in Route Planning in its 2026 Best Software Awards. The platform has run more than 1.5 billion deliveries for 360+ enterprise customers across 30+ countries at 99.99% uptime.

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

The question worth putting to your largest delivery partner at the next review is one sentence long: what is the spread of failure rates across your drivers. Not the average, which you already have and which is probably accurate. The spread tells you whether the improvement available to you is worth 2% or 27%, and a partner who cannot answer it has just told you something useful about how well they can see their own operation. Locus plans employed, contracted and gig drivers against 250+ constraints in one pool. Talk to a Locus specialist about managing drivers across employment boundaries.

Also Read: What a Driver’s First Thirty Days Actually Cost

Frequently Asked Questions

How do you manage driver performance when the drivers work for a delivery partner? Through outlier detection and better planning inputs rather than through direct performance management. Ask for anonymized driver-level outcomes so severe outliers can be found, and share tenure-aware routing, service times and access instructions, which lower the failure rate regardless of who employs the driver.

Is partner-level performance data reliable? Statistically, yes. Even a 15-driver partner produces over 31,000 deliveries a month, so a partner running 20% worse than baseline is detectable in under a month. The limitation is not reliability, it is that the aggregate does not indicate what to fix.

Why is the average failure rate not enough to manage a delivery partner? Because identical averages conceal different distributions. In our illustrative model, two partners at roughly 3.0% differed in where their failures came from, with the worst three drivers producing 12% of failures in one and 34% in the other, and the same intervention was worth 2% in one and 27% in the other.

How quickly can you identify an underperforming driver? It depends entirely on how far from baseline they are. A driver at twice the baseline failure rate is identifiable in about eight working days, and one at three times baseline in about two. A driver only 20% worse needs roughly 146 working days, which makes them effectively unmanageable at the individual level.

What driver data should a brand ask a contracted partner for? Anonymized driver-level outcomes with a stable pseudonymous identifier rather than personnel records. That is sufficient to find outliers and to see the distribution, and it avoids asking a partner for employee data they should not share.

Can you improve contracted driver performance without any driver-level visibility? Yes, through the plan. Tenure-aware route construction, accurate service times and recorded access instructions reduce failures at source and can be applied to work executed by a partner, which is why planning quality is the first lever to use rather than the last.