Why Can an 85% Satisfaction Rating Decide Whether a Delivery Worker Keeps Working?

When Numbers Start Making Decisions · Season Two, “When the Measure Becomes the Target” · Article 8

1. A thumbs-down is more than an opinion

Uber Eats explains that in Australia and New Zealand, a delivery person’s Satisfaction Rating is formed from thumbs-up and thumbs-down feedback by restaurants and customers. The minimum is 85 per cent, and delivery people below the minimum for their city may lose access to the Driver app. Uber: “Your Guide to Satisfaction Ratings”

A customer who taps thumbs-down may want only to express that food arrived cool, the delivery was late or communication was difficult. The customer may not know whether delay came from restaurant preparation, platform routing, traffic, building access or the courier. Nor may they realise that feedback enters a decision about access to work. For the platform, the rating is low-cost, immediate quality information available at enormous scale. For the delivery worker, it can affect continued access to income.

Consumer feedback has changed identity. It is no longer only an answer to “how was this experience?” It helps decide whether this person can continue working. The number does not merely describe service. It manages labour.

Eighty-five per cent is not a natural safety line in customer satisfaction. A courier at 84 per cent does not become categorically less capable than one at 86. The threshold comes from the platform’s need to maintain standards, identify accounts with persistently poor feedback and manage a dispersed workforce. The question is not whether feedback may be used. It is whether this feedback can bear the consequence of stopping work.

2. What does a satisfaction rate measure?

The rate aggregates reactions from customers and restaurants who choose to evaluate completed orders. It can carry real information about courtesy, communication, food handling, finding an address and following delivery instructions.

Yet an order is a joint product. The platform assigns it and estimates time; the restaurant prepares the meal; traffic and weather affect the trip; the customer supplies an address and entry instructions; and the courier collects and delivers. The final thumb is commonly attached to the account closest to the customer.

This is attribution by compression. The system delegates judgement of a complex service chain to users, then aggregates their decisions into a personal score. No customer sees the courier’s long-term performance and no platform employee understands every order in its context. The total nevertheless acquires managerial authority.

The sample is also selective. People with very good or very bad experiences may respond more often. A new courier has fewer ratings, making one response more influential. Language, disability, racial or ethnic bias and differences among service areas can affect the result. A platform can filter obvious abuse or feedback tied to factors outside the courier’s control, but those filtering rules themselves need transparency and audit.

Uber’s separate explanation for passenger-driver star ratings states that the result uses the latest 500 ratings and that low scores due to matters outside a driver’s control, such as traffic, will not count. Uber: “How star ratings work” The example shows that a rating is not simply raw public opinion. The platform already decides which user judgements qualify as knowledge and which count as noise.

3. Why do platforms depend so heavily on ratings?

A conventional employer can use a workplace supervisor to observe, train and record employees. A digital platform coordinates many contractors moving around a city and logging in and out without a shared workplace or a manager who follows each person. Ratings offer scalable delegated knowledge: distribute observation among users and allow a system to aggregate it.

The mechanism can detect persistent service problems. If a courier repeatedly ignores instructions, treats people abusively or damages food, the platform cannot abandon quality control simply because it lacks a traditional supervisor. Warnings and, where necessary, loss of access can protect users and other delivery workers who take care.

Efficiency at scale changes the burden of proof. The platform can produce the conclusion “your satisfaction rate is below the standard” with ease. The worker may be unable to see which orders, reasons, exclusions or denominator produced it. Anonymity can protect customers, but it also makes error and malicious feedback harder to contest.

Workers adapt once ratings affect opportunities. Better communication and more careful handling are genuine improvements. Avoiding high-risk orders, pleasing evaluators and absorbing delay created by a restaurant or platform are forms of target displacement. If workers avoid complex buildings, distant addresses or customers with communication difficulties because of rating risk, a quality measure can produce service exclusion.

4. Who decides what happens below 85 per cent?

Saying “the algorithm deactivated the worker” is incomplete. The platform sets the threshold, defines feedback, chooses the filtering rules, sends warnings and controls account access. Software can execute steps, but institutional responsibility remains with the platform.

Uber’s Australian deactivation policy states that, where possible, workers receive warning, can provide additional information and have access to human review. It also describes notices near a minimum quality line and acknowledges the need to identify users who misuse ratings and customer-support systems through fraudulent reports. Uber: “Deactivations—Losing Account Access”

Those procedures are better than silent automatic removal, but an internally written policy is not the same as an enforceable external standard. The company sets the threshold, explains the reason and reviews itself, while the worker’s income depends on that company. An internal appeal without sufficient facts, independence or power to change the result can become a second calculation rather than a remedy.

Australia has begun transforming the relationship from a platform policy into labour procedure. Since February 2025, the Digital Labour Platform Deactivation Code has generally required a covered platform to provide written warning, identify a conduct- or capacity-related reason and supply enough information for a reasonable worker to understand it, followed by notice and a chance to respond before formal deactivation. Eligible employee-like workers can apply to the Fair Work Commission for an unfair-deactivation remedy, usually within 21 days. The Commission can order reactivation and compensation for lost income. Fair Work Ombudsman: “Employee-like workers”

5. The first cases show that “there was a reason” is not enough

In January 2026, the Fair Work Commission found in Zeeshan Aslam Khan v Uber Eats that a deactivation did not comply with the Code and was unfair. It ordered restoration of access to the platform. The case involved complaints about conduct rather than the satisfaction threshold alone, and for that reason reveals the broader procedural issue.

The Commission found that Uber’s early warning referred generally to inappropriate behaviour and unwelcome sexual remarks without providing the concrete allegations needed for a response. The platform possessed the complaint content, while the abstract labels given to the worker did not allow him to understand the case. The final notice’s statement that the decision was final did not preclude external review. [Fair Work Commission: [2026] FWC 48](https://www.fwc.gov.au/documents/decisionssigned/pdf/2026fwc48.pdf)

The judgement distinguishes a result label from an appealable reason. “Below 85 per cent” and “breached community guidelines” are classifications. A fair procedure needs to identify the relevant orders or conduct, the evidence considered, how the worker’s account was treated, why the information demonstrates an ongoing capacity problem, and why a less severe response is inadequate.

There are limits. Platforms should not disclose information that places a customer at risk. Serious safety, fraud or licence cases can justify immediate suspension. The Code itself preserves exceptions. Procedural fairness does not require every worker to remain active during every investigation. It requires exceptions to have clear reasons, records and timely review.

6. Making the consequence proportionate to the score

Satisfaction is suitable as a warning and a way to discover repeated patterns. A single boundary crossing should not automatically produce permanent deactivation. A more reasonable system would:

1. publish the threshold, sample window, minimum rating count and important filtering rules;
2. distinguish restaurant, platform, traffic and customer causes from conduct the courier controls;
3. provide specific, actionable feedback and a reasonable improvement period near the threshold;
4. audit rating disparities across locations, times, languages and groups;
5. require a person with authority to change the outcome to review context before deactivation;
6. allow workers to submit location, message, photograph and restaurant-wait evidence;
7. rapidly review any immediate suspension and compensate income lost through error; and
8. feed overturned cases back into filtering rules and warning templates instead of only restoring one account.

A second metric should not be introduced thoughtlessly to repair the first. Adding acceptance or cancellation rates to a composite may punish a worker for declining an unsafe or economically irrational order. Each measure must correspond to something the worker genuinely controls.

Conclusion: a rating can trigger investigation, but cannot replace proof

Customer and restaurant feedback is valuable quality information, and an 85 per cent line can identify sustained low ratings. But the result mixes multiple causes, anonymous judgement and a selective sample. It cannot independently prove that a person is no longer fit to work.

My judgement is that a rating may trigger a warning, training and review. To support deactivation, the platform must provide specific reasons, human judgement and an external remedy proportionate to the consequence. Australia’s Deactivation Code and Fair Work Commission jurisdiction make those minimum procedures enforceable, which is an important advance.

When a thumbs-down contributes to whether another person can earn income, the evaluator, platform and law have jointly entered labour management. The number did not dismiss anyone by itself. The central questions remain: who authorised it to carry this function, who can explain the decision, and who is responsible for restoring the opportunity to work when the aggregate judgement was wrong?

Principal sources

• Uber: Your Guide to Satisfaction Ratings
• Uber: How star ratings work
• Uber: Deactivations—Losing Account Access
• Federal Register of Legislation: Digital Labour Platform Deactivation Code, compilation 12 June 2026
• Fair Work Ombudsman: Employee-like workers
• [Fair Work Commission: [2026] FWC 48](https://www.fwc.gov.au/documents/decisionssigned/pdf/2026fwc48.pdf)