Before an Hourly Shift Starts, Workers Price Gas and Childcare
On this page 7 sections
On September 17, Workday and DailyPay published a number that belongs beside every open-shift notification. Seventy-eight percent of the hourly workers they surveyed said they calculate costs such as gas, commuting, childcare, and other expenses before accepting a shift.
A notification may show a start time, an end time, and an hourly rate. The worker still has to finish the price.
A six-hour shift at $18 an hour appears to offer $108 before tax. A parent may need an extra block of care. Another worker must buy fuel before payday. Someone else may lose the last bus home before closing. If the site sends people home when demand falls, even the six hours may be an estimate. The accept button arrives before those variables have settled.
Harris Poll conducted the online survey from April 23 through May 2, 2026. It covered 2,208 US adults ages 18 to 45 who were employed and paid hourly. Responses were weighted, and the reported Bayesian credible interval was plus or minus 3.1 percentage points at a 95% confidence level. Workday and DailyPay sponsored the study and sell products tied to workforce management and employee pay.
The age limit matters. The sample does not describe hourly workers older than 45 or salaried employees. It also excludes contractors paid by a platform and people who wanted work but were not employed when Harris Poll recruited respondents.
Subgroup estimates carry wider uncertainty than the overall interval. A retailer with an older workforce should test the finding against its own shifts before treating 78% as a local forecast. So should a hospital with union schedules or a restaurant franchise.
That sponsorship does not make 2,208 answers disappear. It does change the work required of a reader. A stated preference for flexible schedules is not a measured retention gain. A worker saying that on-demand pay would influence a job choice is not evidence that a particular pay product caused a person to stay. A mobile tool can speed a shift swap without increasing the hours available or reducing the cost of getting to work.
DailyPay chief operating officer Andrew Brandman presented the expectation gap as a recruiting and retention opportunity. Josh Secrest, a Workday vice president for Paradox, argued that scheduling, pay clarity, and frontline support should sit together rather than arrive as piecemeal tools. Both propositions are commercially coherent. They also come from executives whose companies benefit when employers buy the integration. A buyer still needs a baseline, an alternative, and worker-side measures before calling the gap closed.
One release joins several decisions that employers often keep in separate systems. Operations forecasts demand. Managers post and approve shifts. HR selects workforce software. Payroll determines when earned wages become available. Finance reviews labor cost. A worker sees the combined result as one question: after the costs and risks, is this shift worth taking?
That is the unit a frontline technology purchase should preserve. Employers can still measure fill rate, time to fill, attendance, overtime, and manager effort. They also need enough data to calculate whether a filled shift was viable for the person who worked it. Without that second view, a faster scheduler can make an employer’s empty slot disappear while leaving the worker’s economic problem unchanged.
A shift offer arrives before the household budget closes
Workday and DailyPay describe a workforce making small financial decisions at high frequency. Forty-six percent of respondents said they borrow between pay periods. Sixty-nine percent said they check their pay after every shift. High wages were important to 88%, while clarity about how pay is calculated was important to 84%.
Those answers belong to different parts of the household ledger. The wage rate sets gross earnings. Hours determine how much of that rate reaches the pay statement. Pay clarity determines whether the worker can predict the result. Pay timing determines whether the money arrives before rent, fuel, food, or care is due. Borrowing fills a gap when those dates do not line up.
National household data is broader than hourly work but helps explain why timing matters. The Federal Reserve’s 2025 household survey, fielded in October 2025 among nearly 13,000 adults, found that 63% could cover a $400 emergency expense using cash or its equivalent. Twelve percent said they could not cover it by any method. Sixteen percent had failed to pay all their bills in full during the previous month.
Federal Reserve researchers did not ask the same question of the same sample. They did not measure whether a shift app improved anyone’s finances. Its results show the limited buffer surrounding many household decisions. A transport or care payment that looks small in a staffing model can decide whether work is possible before the wage arrives.
Childcare makes that mismatch visible. Among parents who used paid care, the Federal Reserve reported that the median monthly cost was at least half of their housing payment. A worker considering an extra four-hour shift may be buying a minimum care block that does not shrink with the shift. The employer pays for four hours. The parent may pay for six, including travel and the provider’s booking rules.
Commute costs behave differently but create the same comparison. Fuel is a cash expense before work. A bus fare is paid whether a shift lasts three hours or eight. A ride-hailing trip can increase sharply at the end of a late shift. Walking or cycling may avoid a fare while introducing weather, safety, disability, and time constraints. None of these costs is fully represented by distance alone.
An employer rarely knows the worker’s full household budget, and it should not collect intimate details merely to make a schedule. It does need to understand the categories that shape acceptance. A shift offer can show guaranteed hours, location, notice, expected end time, cancellation rules, pay date, and available transport or care support. The worker can then compare the job’s terms without disclosing the name of a child or the balance in a bank account.
That distinction matters for analytics. If a worker declines an open shift, a system may classify the event as low engagement. The actual reason might be that two hours of care cost more than the after-tax value of the final two hours of work. A model trained only on accept and decline events will learn the pattern without understanding the price.
Employers can respond by sending more notifications, raising the rank of workers who often accept, or asking an agent to search a wider pool. Each response may improve the employer’s fill probability. None changes the economics of the offer unless the terms change too.
Schedule control carries a twenty-one-point gap
Seventy-two percent of survey respondents described control over their schedule as important. Only 51% were satisfied with the control they had. The 21-point difference is not a product adoption rate or a turnover forecast. It is a gap between what people said mattered and what they said they experienced.
Pay clarity showed an even larger separation: 84% considered it important, while 57% were satisfied. Mobile tools for schedules, pay, and communication were important to 70%, with 53% satisfied. The pattern suggests that access to an app and control over work cannot be counted as the same thing.
A mobile schedule can make a fixed decision easier to see. It can also make a negotiable decision easier to change. The difference lies in the rights and operating rules around the interface.
Consider an open-shift marketplace. One employer may let workers claim a posted shift immediately, display guaranteed hours, and preserve the shift if demand later falls. Another may accept a worker’s request but leave it pending until a manager reviews staffing.
A third employer may use the same interface to call people in and send them home as volume changes. All three can report mobile scheduling. Only the first gives the worker a firm basis for arranging transport and care.
Managers face constraints of their own. A restaurant manager may be filling a dinner gap after two callouts while watching sales, service time, food safety, and overtime. A warehouse supervisor may not know the next day’s volume until a late forecast. A hospital unit can face patient demand that refuses to follow a retail schedule. Rules that freeze every shift far in advance can transfer forecasting error into overstaffing or exhausted teams.
That is the strongest case for flexible workforce technology. A faster swap can let one employee handle a family event without forcing a manager through a call tree. An open shift can give another employee access to hours they want. A forecast can reduce the number of last-minute requests if it is used to plan rather than merely to optimize labor at the final moment.
Real flexibility depends on who can initiate, refuse, and reverse an action. A worker should be able to decline an extra shift without an opaque penalty. A manager needs a clear escalation path when coverage becomes unsafe. Payroll needs a trace from the approved schedule to recorded time and calculated pay.
If an AI recommendation changes who gets offered hours, its inputs need to be visible. Availability, seniority, skill, overtime, location, and acceptance history can each change the ranking.
Job choice provides another clue. Forty-five percent ranked schedule among the top three factors in accepting their current job, behind only pay. That does not mean nearly half would leave because of one late schedule. It does mean schedule quality belongs in a job’s value proposition rather than in a back-office usability score.
Employers often split those records. Recruiting advertises flexibility. Workforce management records the shifts. Payroll records the earnings. Employee listening asks whether people plan to stay. A candidate can accept the promise without seeing the actual distribution of notice, cancellations, swaps, or weekly hours at the site where the job sits.
A credible recruiting claim would use operating evidence. What share of shifts is posted at least two weeks ahead? How often do start or end times change after posting? How many requested swaps receive an answer before the worker must make a care or transport commitment? What share of scheduled hours is actually worked and paid? These measures turn flexibility from a label into a condition someone can compare.
Early pay moves the wait and can add a fee
On-demand pay addresses a genuine timing problem. A worker may have completed the labor while the regular payday is still days away. Moving part of those accrued wages earlier can cover fuel for the next shift, prevent an overdraft, or reduce the need for another form of short-term borrowing.
In the Workday and DailyPay survey, 77% said access to on-demand pay would influence their choice between two otherwise similar job offers. That answer measures stated influence. It does not show how many respondents had used the service, what they paid, whether they avoided other borrowing, or whether the feature changed retention after hire.
One product can be free under one path and carry a fee under another. DailyPay’s current fee page lists a $0 account fee and a $0 transfer that arrives in one to three business days. It lists a $3.49 flat fee for an instant transfer, while the company’s program terms describe an instant-transfer range from $1.99 to $3.99 depending on the employer. The exact experience therefore depends on employer configuration, transfer method, and timing.
Calling on-demand pay either free or expensive without those fields would hide the decision. A no-fee next-day option may solve a worker’s problem. A $3.49 instant fee may also be rational if it prevents a larger overdraft or gets a car to the next shift. Repeated fees can still absorb part of an already narrow margin.
Historical regulator data illustrates the repetition. In a 2024 Data Spotlight, the Consumer Financial Protection Bureau analyzed information from eight employer-partnered earned wage providers. The sample covered more than 7 million workers and roughly $22 billion accessed during 2022, which the agency estimated was less than half of the market.
An average transaction in the sample was $106. The average user made 27 transactions and accessed about $3,000 during the year. Employers subsidized fewer than 5% of total fees. Where employers did not cover fees, about 90% of workers paid at least one, averaging $68.88 over the year. The bureau calculated an illustrative annual percentage rate of 109.5% for a common transaction using its assumptions.
Those figures should not be pasted onto a current DailyPay customer. They describe 2021 and 2022 data from a provider sample, aggregate different fee models, and predate current terms. An annualized rate can make a short fixed fee easier to compare with credit. It cannot tell an employer whether a particular worker avoided a more costly alternative or whether the employer could have paid the transfer fee.
Legal guidance also changed after the spotlight. A December 2025 CFPB advisory opinion addressed a narrow class of covered earned wage access products. Such a product is not credit under Regulation Z when it relies on employer payroll data and does not exceed accrued wages. The definition also requires no worker recourse or liability, among other conditions.
That opinion does not say that every product outside the definition is credit. It does not decide whether a fee is good value, whether faster pay improves retention, or how another law applies. A compliance team may need the classification. A worker still needs the dollar amount and arrival time.
Employers can make the feature easier to evaluate by retaining the choice set. For each transfer, record the wages available, free option and delivery time, paid option and fee, employer subsidy, selected path, and any failed transfer. Aggregate the results without turning individual cash-flow behavior into a manager’s performance signal. Then compare usage with payroll corrections, overdraft complaints if voluntarily reported, missed shifts, and retention cautiously.
That record creates a better renewal question than adoption alone. If employees use the instant route repeatedly, is that evidence of product value, a regular payday mismatch, insufficient wages, volatile hours, or a fee that the employer could absorb? More taps cannot answer.
AI can fill a shift without making it worth taking
Workday’s current frontline product pages describe a chain that begins before a shift. Candidates can search and apply by text, schedule interviews, and receive offers. Employees can view schedules, claim open shifts, request time off, check in, and access pay information on mobile devices. Managers can use demand forecasts and labor information while handling schedule changes.
Workday’s Workforce Management Agent page brings AI into that chain. It says the agent can handle schedule changes, time, and shift swaps, with manager approval when needed. Workday reports a 90% reduction in time spent managing shift changes, 65% average process automation, and a 75% reduction in manual time-entry errors.
Those are vendor-reported measures on a product page. The page does not establish that the 2,208 survey respondents worked for customers using the agent. It does not publish the sites, baselines, sample sizes, worker outcomes, or distribution behind the percentages. They are reasons to ask for customer-specific evidence, not independent proof of a general result.
Useful capability remains beneath the marketing numbers. Shift administration consumes time that managers could spend on customers, coaching, safety, or exceptions. A worker who can complete a swap without waiting for a manager’s free moment may save hours of uncertainty. An automated check can prevent a swap that violates skill, rest, age, union, or overtime rules.
Yet the same system can optimize the wrong unit. If its target is open-shift fill rate, it may rank people most likely to accept and send them repeated offers. If its target is labor-cost adherence, it may steer hours away from workers close to overtime. If its target is speed, it may approve the first eligible response. Each rule distributes income and opportunity, even when the interface presents the action as scheduling convenience.
An acceptance prediction trained on past behavior can reproduce constraints. A worker who previously declined late shifts because childcare was unavailable may receive fewer offers. A worker who accepts every notification because rent is due may be treated as the best match even when the schedule creates fatigue. A worker with limited mobile data or an inaccessible interface may appear unresponsive.
Automation has a credible counterargument. A rules-based marketplace can offer every qualified worker the same window, document why overtime or rest rules excluded someone, and reduce the favoritism of a manager’s private call list. It can remember preferences more reliably than a paper roster.
That is a reason to test distribution, not a reason to assume it. Compare the automated offer pool with the prior manual one. Then ask whether access to hours widened across sites, shifts, languages, ages, disabilities, and availability patterns.
UKG’s 2026 international frontline survey provides useful context for that tension. The vendor-sponsored study covered more than 8,000 frontline employees in ten countries. It reported that 38% used AI at work, while 64% worried AI might replace their job and 51% said their employer had not explained its plans.
Those percentages cannot be transferred to the US hourly sample. Self-reported use cannot prove productivity either. They show why a faster workflow can arrive with uncertainty about who benefits.
An employer therefore needs two AI records. The first is operational: recommendations made, changes approved, time saved, errors prevented, exceptions created, and manager work remaining. The second is distributive: who received offers, who was omitted, hours offered and worked, notice, undesirable-shift concentration, overtime, cancellations, declines, and accessibility failures.
Those records may point in different directions. Manager processing time can fall while notice becomes shorter. Fill rate can rise while the same people absorb the least predictable hours. Payroll errors can fall while workers pay more to access earnings quickly. A product can be effective at its configured objective and still produce a poor job.
Procurement should require an export at the shift level, not only an executive dashboard. The employer needs enough history to reconstruct why a recommendation appeared, who approved it, what changed, what was paid, and whether a worker could correct the record. Without that export, the organization cannot separate a forecast error, policy choice, manager override, model ranking, or worker constraint after the outcome.
Stable schedules have evidence behind them
Software vendors do not own the full scheduling evidence base. Cities and states have spent years testing rules about advance notice, predictability pay, access to hours, rest between shifts, and compensation for late changes.
In July 2026, the Shift Project’s fair-workweek research review counted implemented laws in eight municipalities and one state since San Francisco’s 2015 ordinance. Its review concluded that these policies reduce precarious scheduling. It found no conclusive evidence that they reduce employment or work hours. Evidence for sleep, material hardship, and broader well-being was less comprehensive.
That conclusion is narrower than a victory claim. Jurisdictions use different rules. Employers and sectors respond differently. Some studies identify associations while others exploit policy changes. A law can reduce a particular scheduling practice without making every worker’s weekly income stable or every manager’s forecast accurate.
Los Angeles fast-food data shows the operating problem in a specific market. The Shift Project reported that 59% of surveyed workers received less than two weeks of notice, 40% experienced on-call work, and 70% faced last-minute changes to timing or length. The typical worker had a 13-hour difference between the busiest and slowest weeks. Only about one third worked at least 40 hours, while another third wanted full-time work but could not obtain it.
This sample should not be treated as a national estimate for hourly employees. It describes fast-food work in Los Angeles and reflects that project’s recruitment and measurement choices. It still identifies variables that a national technology buyer can measure at its own sites: notice, variation, on-call incidence, change timing, desired hours, and actual hours.
Another Shift Project study followed 1,827 US service-sector workers across two survey waves. Its turnover analysis found that schedule instability predicted turnover among the third of workers with the most unstable schedules. The panel design establishes sequence more clearly than a one-time opinion poll, but it still does not randomly assign schedules. Local management, wages, life changes, and workplace conditions can affect instability and departure together.
Policy evidence also reframes flexibility. Employers sometimes describe flexibility as the ability to adjust labor to demand. Workers may mean the ability to plan and to change a shift when life intervenes. Both can exist, but they are not interchangeable.
A useful scheduling policy defines the allocation of uncertainty. If the employer changes a posted shift, who absorbs the transport or care cost? If a worker requests a change, how quickly must a manager answer? If demand falls, are hours guaranteed, offered elsewhere, or removed without pay? If demand rises, are additional hours offered fairly or sent to a narrow pool selected by an acceptance model?
Technology can administer those choices consistently. It cannot choose what fairness means without an employer policy. An AI agent may enforce advance notice, predictability pay, rest rules, qualifications, and worker preferences. It can also be configured to search for exceptions until a staffing target is met. The policy determines whether automation protects a boundary or treats the boundary as friction.
Managers should not be left as an invisible buffer. Stable schedules can require better forecasts, cross-training, relief capacity, higher base staffing, and authority to spend when demand moves. If a company buys an AI scheduler while holding staffing and manager labor constant, the implementation may simply compress more decisions into the same hour.
Manager incentives belong in the implementation file. A site leader measured only on labor percentage and coverage will have little room to preserve hours after demand falls, even if the corporate policy praises predictability. Give that manager a variance budget, a clear exception rule, and credit for notice and pay accuracy. Otherwise the software will enforce a promise that the operating scorecard quietly punishes.
Workers should not be the only source of flexibility either. A scheduling system can expose where demand volatility comes from, which sites rely on repeated call-ins, and which managers carry high exception loads. That information can support a budget for float staff, transport support, care partnerships, minimum shift lengths, or paid cancellation.
A fixed schedule is not the only acceptable answer. Buyers still need to compare the product against the operating change the employer could make. A better forecast, a larger relief pool, a minimum-hours rule, and a faster swap tool solve different parts of the problem. Buying one does not make the others unnecessary.
Build a shift acceptance economics file
A shift acceptance economics file connects the employer’s coverage decision to the worker’s observable terms without turning a household into an HR database. It uses one shift identifier across forecast, offer, acceptance, time, pay, and outcome records.
Begin the file before the notification is sent. Record the site, role, required skill, posted start and end, guaranteed and expected hours, hourly rate, premium, overtime status, notice, and reason the shift opened. Keep the forecast version and staffing target that produced it. If an agent selected recipients, preserve the eligibility rules, ranking inputs, exclusions, and model or rules version.
Then preserve the offer workers actually saw. Did it show the full location, expected end, guaranteed hours, pay rate, premium, pay date, cancellation policy, and response deadline? Was the message available in the worker’s language and an accessible format? Did a person without a smartphone have an equivalent route?
Do not require workers to reveal private household details. Offer a small, voluntary decline taxonomy that supports operating decisions. It can include schedule conflict, transport, care, notice, hours too short, pay, health or fatigue, accessibility, qualification mismatch, another reason, and prefer not to say. Prevent managers from using a decline category as a performance score.
After the response, join only the facts needed to evaluate the workflow. Record when the offer was opened, accepted, declined, expired, approved, changed, or cancelled. Preserve manager response time, worker correction, swap attempts, and any policy exception. Link scheduled hours to clocked and paid hours, including early send-home, premium, overtime, and payroll correction.
A worker’s economic estimate can remain local or voluntary. An employer can supply a calculator that lets a person enter commute, care, meal, uniform, tool, parking, toll, and optional pay-access costs without storing the entries. What matters for the employer is whether the displayed terms were sufficient and which categories repeatedly block acceptance across a site.
A smaller employer does not need a data warehouse to start. A four-week sample can use a protected spreadsheet with one row per offered shift, a short decline list, and weekly payroll reconciliation. The minimum viable record is the promise and the result: notice, rate, guaranteed hours, response, changes, clocked hours, paid hours, and fee path. More fields are useful only when someone can protect them, explain them, and change a decision because of them.
Use the following record for a pilot:
| Field | Minimum record | Decision it supports | Boundary |
|---|---|---|---|
| Shift value | Rate, premium, guaranteed hours, expected gross pay | Whether the offer is financially legible | Gross pay is not take-home pay |
| Notice and change | Posted time, response deadline, edits, cancellation | Whether workers can plan transport and care | A viewed notification is not control |
| Worker cost categories | Optional local estimate or voluntary reason code | Which terms block acceptance | Do not collect names of dependents or bank balances |
| Pay access | Regular pay date, free path, paid path, fee, subsidy | Whether faster pay moves or adds cost | Use is not proof of financial health |
| AI action | Rule or model version, recipients, ranking, explanation | Whether opportunity was distributed as intended | Prediction is not worker preference |
| Human action | Manager approval, override, response time, reason | Whether accountability survives automation | Approval alone does not show review quality |
| Completion | Clocked hours, paid hours, correction, early release | Whether the promised work occurred | Acceptance is not a completed shift |
| Outcome | Fill, attendance, overtime, repeat acceptance, retention window | Whether operations improved over time | Retention is not automatically caused by the tool |
Run the pilot at a level where managers and workers can see the change. Compare similar sites or phase the rollout so the organization can distinguish a product effect from seasonality, wage changes, local leadership, weather, transit, or demand. Set a measurement window before launch. A two-week improvement in fill rate cannot establish six-month retention.
Pair every employer measure with the worker-side term it can affect. Put time to fill beside notice. Put manager minutes beside worker response time. Put scheduled hours beside hours worked and paid. Put instant-pay use beside fee incidence and free-path use. Put acceptance beside decline reasons. Put overall fill beside the distribution of desirable and undesirable hours.
A pilot also needs a stop condition. Pause an automated ranking if qualified workers are systematically omitted, if notice deteriorates, if cancellation rises, if explanations cannot be reconstructed, or if payroll cannot reconcile the approved shift. A manager must be able to offer a shift manually, and a worker must be able to correct availability or challenge a missing payment without navigating the model.
Finance should calculate loaded cost, not just license price. Include implementation, integration, devices, mobile data support, payroll reconciliation, manager training, worker training, accessibility, fee subsidy, support, audits, and exception labor. Compare that cost with overtime, agency labor, missed coverage, manager time, payroll corrections, and turnover, while keeping causal claims proportional to the design.
Sometimes the file will expose a better answer than software renewal. Repeated transport declines may support a shuttle or a different shift boundary. Care conflicts may justify minimum shift lengths or more notice. High paid-transfer use may support employer-funded instant access or a different payroll cadence. Slow manager approval may require authority and coverage, not another employee notification.
The answer comes before the clock-in
This survey’s most useful result arrives before work begins. A worker does not experience a shift as an isolated hourly rate. The decision includes getting there, arranging care, trusting the hours, understanding the pay, waiting for the money, and judging whether another late change will erase the plan.
Workday and DailyPay can help employers move parts of that decision faster. Scheduling tools can expose open hours and process swaps. Payroll connections can show earned wages. An agent can check rules and route an exception. Fast access to pay can be valuable, particularly when the free path arrives in time or the employer covers the fee.
A strong case for those products does not require pretending that technology created the underlying job quality. A faster swap is useful. A clear pay calculation is useful. A no-fee transfer is useful. A manager who receives fewer administrative tasks may have more time to solve a real coverage problem.
Limits need equal visibility. A notification does not guarantee hours. An accepted shift does not guarantee the worker can keep it after a care provider cancels. An instant transfer can move money forward while charging for speed. An AI ranking can improve fill while concentrating volatile work. A vendor survey can reveal a decision without proving that the vendor’s product improved it.
Use a concrete operating test. Before deployment, choose several sites and reconstruct the last month of open shifts. For each one, identify what the worker saw, how much notice existed, and what later changed. Record who approved it, what hours were worked, what was paid, and why the offer was declined when a worker chose to say. Then run the same record through the new workflow.
Success should appear on both sides of the shift. Managers spend less time chasing coverage. Workers receive earlier and more reliable offers. More scheduled hours are worked and paid. Payroll corrections fall. Fees are visible and avoidable. Desirable hours do not narrow around a favored group, and volatile hours do not collect around the people least able to refuse.
If fill rate improves while notice falls, the result is incomplete. If manager time falls while corrections rise, the work moved. If more workers use instant pay while average fees grow, the feature deserves a cost review. If employees say schedule control matters and the system merely sends more offers, the implementation answered a different question.
Each open-shift alert should carry enough information for two decisions. Can the site operate? Will the shift still work after gas, childcare, time, and pay are counted? The clock-in is too late to discover that only one side did the math.