Does using scheduling software make my contractors employees?

Software alone does not change status — a booking calendar contractors opt into is not control. What matters is whether the system directs the work: assigning specific jobs rather than offering them, penalizing declines, prescribing the sequence and method, monitoring performance in real time, and removing access as a sanction. Those are the same facts an agency or court would weigh if a human supervisor did them, and encoding them in software does not make them lighter.

Which test applies to us — ABC or the common-law factors?

Several at once, which is the practical difficulty. A number of states apply an ABC test for wage and unemployment purposes, where the burden is on the hiring entity and Prong B often decides the case. Federal wage-and-hour analysis uses a multifactor economic-reality approach, and tax classification uses common-law control factors. A worker can be a contractor under one and an employee under another, so the useful question is never 'are they a contractor' but 'under which statute, in which state, for which purpose'.

Why is Prong B the hardest part for a platform?

Prong B asks whether the work performed is outside the usual course of the hiring entity's business. A software company that connects customers to cleaners can argue that cleaning is not its business; the counterargument — that its business is delivering cleaning, and the software is the means — has succeeded often enough that no one should treat the framing as settled. The heavier your algorithmic control, the weaker the 'we are only a marketplace' story reads, because marketplaces do not usually assign, score and deactivate their sellers.

Does letting contractors set their own rates fix it?

It helps, and only if it is real. Rate-setting that operates within a band the platform sets, or where the algorithm steers volume toward workers who price where the platform prefers, is a nominal freedom with a control mechanism behind it. The question decision-makers ask is whether a worker who prices high simply earns less per job or effectively stops receiving work — the second pattern is the platform setting the price through the ranking function.

What does misclassification actually cost?

It is rarely a single bill. Exposure typically compounds across unpaid overtime and minimum wage with liquidated damages, unreimbursed business expenses in states that require reimbursement, unpaid payroll taxes with penalties and interest, unemployment and workers' compensation contributions, benefit-plan claims, and in some states penalties for willful misclassification. Because the algorithm treats every worker in a class identically, these claims aggregate naturally, and the aggregation is usually what makes them worth bringing.

When the Algorithm Becomes the Boss: AI Scheduling and Worker Misclassification in 2026

The Substitution Nobody Priced

The pitch for algorithmic management is that it removes overhead: no dispatcher, no shift manager, no quality supervisor. The work still gets assigned, sequenced, measured and sanctioned — the software does it. That substitution is a real efficiency, and it is also the reason the classification analysis gets harder rather than easier.

Every classification test in circulation is, at bottom, asking who controls the manner and means of the work and whether the worker runs an independent business. Those tests were written with human supervisors in mind, but nothing in them turns on the supervisor being human. A rule that deactivates anyone below an acceptance-rate threshold is a rule about whether the worker may decline work, and that is one of the oldest control questions there is.

Marketplace Story vs. What the System Does

Most platforms describe themselves in the left column. The classification analysis reads the right one, because it looks at the product rather than the positioning.

The Marketplace Story

•We connect independent businesses with customers
•Workers choose when and whether to work
•Ratings are just customer feedback
•Pricing is set by supply and demand
•Removal is an ordinary account decision

What the System Actually Does

•Assigns a specific job to a specific worker
•Penalizes declines through ranking or thresholds
•Converts ratings into eligibility and discipline
•Sets the rate and the worker accepts or waits
•Terminates the relationship for policy violations

Five Features That Read as Control

Assignment Instead of Offer

The distinction that carries the most weight and costs the least to change. A board of available jobs a worker browses and claims looks like a marketplace. A system that pushes one job to one worker with a countdown and a penalty for declining looks like a dispatcher. If your product does the second, the classification story has to survive that fact rather than talk around it.

Acceptance and Cancellation Thresholds

A minimum acceptance rate is a direct restriction on the freedom to refuse work, which is close to the definitional core of contractor status. Cancellation penalties, priority loss and temporary suspensions are the same mechanism in softer language. These thresholds are usually documented in a policy page, which means the evidence is already written and served.

Prescribed Method and Route

Required app-based navigation, mandated task sequences, scripted customer interactions, uniform or equipment requirements, in-app checklists that must be completed in order. Independent businesses decide how the work gets done; if your system decides, you have taken the manner-and-means question off the table. Quality standards for the deliverable are a materially different thing from step-by-step direction, and the drafting distinction is worth making explicit.

Ratings as Discipline

Customer feedback is normal in any business. What converts it into supervision is the automated consequence: fall below a threshold and lose ranking, lose access to premium work, or lose the account. At that point the rating is a performance-management system, and the fact that customers supply the inputs does not change who administers the sanction.

Automated Deactivation

The most consequential and least examined feature. Deactivation is termination, and doing it by model creates two problems at once: it is strong evidence of control, and it is the point where algorithmic-transparency and appeal obligations are appearing in newer state and non-US rules. A deactivation path with no human review and no stated reason is difficult to defend on either axis.

Automation Is the Plaintiff's Best Exhibit

A misclassification claim against a traditional business is fought worker by worker, because the degree of control varied with the supervisor. That variability is the standard defence to class treatment, and algorithmic management destroys it.

The rule is uniform. Every worker in the class was governed by the same threshold, which is precisely the commonality a class action needs.
The rule is written down. Configuration, policy pages and release notes describe the control in your own words.
The outcomes are logged. Assignments, declines, scores and deactivations are all in a database with timestamps.
The history is preserved. Version control shows when control tightened, and by how much.
Nobody exercised discretion. There is no supervisor to testify that this worker was treated differently.

The uncomfortable summary: the same properties that make algorithmic management efficient — uniform, documented, logged, discretion-free — are the properties that make the resulting claim easy to certify and hard to individualize.

Frequently Asked Questions

Can our contractor agreement solve this?

It is necessary and it is not sufficient. Every classification test looks at the actual relationship, and an agreement reciting that the worker controls the manner of work while the product assigns tasks and penalizes declines is evidence against you, not for you. Where agreements do real work is in defining the deliverable rather than the method, preserving genuine substitution and multi-platform rights, and not contradicting what the software does.

We are a B2B SaaS selling the scheduling tool, not the employer. Are we exposed?

Your direct classification exposure is your customers', but two indirect exposures are real. Enterprise buyers increasingly diligence whether a workforce tool creates classification risk, so the feature set becomes a sales objection. And joint-employer theories can reach a vendor whose system, rather than the customer, is setting the terms of work. The product decision worth making is to keep control settings configurable and default them to the looser setting.

Which single change reduces risk the most?

Convert assignment into offer and remove the penalty for declining. That change touches the control factor with the most weight in the most tests, it is visible in the product rather than buried in a policy, and it is generally the least damaging to unit economics of the available options. Removing automated deactivation without human review is a close second.

Does an AI-driven staffing marketplace change the analysis?

Only in the direction of more scrutiny. Matching supply to demand is a marketplace function; scoring workers, setting the rate, sequencing the work and removing people from the platform are employer functions. Marketplaces that added those features to improve quality took on the classification question that came with them, and the sequencing is usually invisible internally because each feature was shipped on its own.

Should we reclassify preemptively?

It is a decision to take with counsel and with a clear view of both directions of risk — reclassifying can be read as an admission about the prior period, and doing it badly creates back-liability while removing the flexibility you were paying for. What is unambiguously worth doing first is the inventory: list every automated decision your system makes about a worker, and grade each against the tests that apply in your states. Most teams discover control they never intentionally chose.

Read Your Own Product as an Exhibit

Nobody at your company decided to supervise contractors. The assignment engine was a fill-rate fix, the acceptance threshold was a reliability fix, the deactivation rule was a quality fix. Together they describe a manager.

Print the list of automated decisions your platform makes about a worker and read it as an opposing party would. If it reads like a job description, that is the finding.

This article is general information about worker classification and algorithmic management, not legal advice. Classification tests differ by state, by statute and by purpose, and outcomes are intensely fact-specific — consult employment counsel in your jurisdictions before changing worker classification or platform controls.