Workday time off requests: mobile 4.2% error rate vs 1.1% desktop — rebuild form

The 4.2% mobile versus 1.1% desktop error gap on Workday time-off requests is a submission-error claim with an undefined denominator, so it does not by itself justify rebuilding the mobile form. A rebuild is warranted only if a four-week re-measurement on one absence type and one time off plan reproduces the mobile-minus-desktop gap with a lower 95% bound of at least 2.0 points.

How a request error gets counted in Workday

The 4.2% mobile against 1.1% desktop figure means nothing until two things are written down: what counts as one error, and what sits in the denominator beneath it. Fix the numerator to one error per submitted time-off request — not per field, not per screen, not per validation message. A request rejected for balance eligibility and resubmitted the same day is one request carrying one error, even though the worker may have seen two error states across two screens. Put that rule in the measurement plan before pulling a single record, because the alternative — counting each flagged field — inflates whichever arm surfaces validations sequentially rather than all at once.

Then verify how your own tenant actually behaves. Confirm whether Workday records the same-day resubmission as a new request or as an edit against the original, because that single answer changes both the numerator and the denominator. The doc.workday.com "Setup Considerations: Absence Management" page describes worker-facing behavior — what a requester sees and can do — not the audit fields your report will read. Take a sample of rejected-then-resubmitted requests from your own absence records and trace the keying before trusting either count.

The denominator is every submitted time-off request in the same window that shares one absence type, one time off plan, and a known device class. A draft that never reaches submission has no submission to be in error; count pre-submission abandonment separately or not at all, but keep it out of this fraction.

Attach device class at the request level, not the worker level. Someone who submits from a phone one week and a laptop the next belongs in both arms, and assigning that person a single device would silently move requests between arms. If device class is not a field on the request, join it from access logs and report how many requests failed to match — an unmatched share large enough to matter tells you the arms are not cleanly separated.

Hold one absence type and one time off plan constant across both arms. Annual leave, sick leave, and FMLA-style plans sit behind different balance rules, eligibility checks, and plan associations, so the number of places a request can fail differs by construction. Mix the three and the mobile-versus-desktop rate can move on composition alone, with no change to the form. After the four-week re-measurement on that single type and single plan, the rebuild is licensed only if the lower 95% bound on the mobile-minus-desktop gap reaches 2.0 points; any narrower result sends you back to the error codes, not to the form.

What three independent sources actually support

The first question about a mobile-versus-desktop time-off error gap is not whether the gap is large but whether the metric has an independent warrant. Kualitatem’s enterprise usability list puts error rate among five gold-standard metrics, alongside task success and time-on-task, so error rate is a defensible family of measure for a form-submission question. The check is whether the source reports a rate value for the product in question. Kualitatem names the metric, not a value. None of the three sources examined here supplies a Workday error rate; they support the choice of metric, not the size of the gap.

Microsoft’s ESUS research, presented at CSCW 2023, makes the instrument argument. The paper contends that enterprise products need their own reliability-tested usability instrument because generic usability scales do not transfer cleanly to enterprise contexts. For a Workday time-off form, that is an argument for measuring inside your tenant with a stable task and a documented counting rule, not for borrowing a benchmark from another product or a generic scale. The check is whether the instrument has been reliability-tested where you will use it; ESUS supplies a method and rationale, not a tenant result.

Nielsen Norman Group’s enterprise usability guidance adds a variability warning: enterprise usability is less researched and more variable than better-defined usability levels, which is why it should not be ignored. The practical implication is that an external enterprise benchmark is a weak comparator unless you can show it is stable for your population and task. The check is to ask whether the benchmark was gathered on the same task, same user group, and same instrument. NN/g gives you the reason to measure locally, not a benchmark value.

Read together, the three sources answer three different questions. Kualitatem answers which metric family is legitimate; Microsoft answers whose instrument should produce the number; NN/g answers how much weight an external benchmark can bear. None answers the tenant-level question. The check is a source-to-question map: if a citation cannot name the task, the user population, and the instrument, it cannot settle the gap.

The rule is to cite these sources for metric choice, not for effect size. Use Kualitatem to justify error rate as an enterprise usability measure, Microsoft ESUS to justify tenant-specific reliability testing, and NN/g to justify skepticism toward borrowed benchmarks. Then treat any external mobile-versus-desktop rate as an untested hypothesis until your own re-measurement on a fixed task reproduces a gap that clears the pre-agreed decision threshold. That is the only path from a source-backed metric family to a rebuild decision.

Four options, one conditional winner

Four moves sit on the table, and only one should get funded this quarter — which one depends on the audit trail, not on how bad 4.2% against 1.1% feels. The gap is 3.1 points; half of it is 1.55 points. That half-gap line is the pivot for the entire comparison.

A, a full mobile form rebuild, addresses layout, defaults, and input load. It touches nothing in the time off plan configuration — not absence type defaults, not the eligibility display, not the approval path. Its cost is the quoted build plus regression risk to the desktop parity behind the 1.1% figure, and it is unaimed until instrumentation says which step fails. If rejections originate in a plan-level default the form renders faithfully, the rebuild reproduces the same error on a cleaner screen.

B, error-code triage then a targeted fix, reads the top three validation and rejection codes from the absence audit trail over the same four weeks, same absence type, same time off plan. Rank those codes by their share of the 3.1-point gap. If one code carries 50% or more — at least 1.55 points — and traces to a default absence type, an eligibility display, or an approval step, fix that one thing and re-measure before touching anything else.

C, a time off plan configuration fix, changes those defaults and eligibility rules without touching the form — the cheaper branch when the trace lands in configuration. D, defer and instrument, keeps desktop as the sanctioned channel for that absence type while code capture improves.

Error-code triage beats a full rebuild unless the leading code accounts for under half the gap. Sequence it: re-measure first, and if the lower 95% bound on the mobile-minus-desktop gap falls below 2.0 points, D wins by default and no build is warranted. If it clears 2.0 points, triage. A single code at 1.55 points or more concentrates the problem enough to fund B — or C, when the trace is configuration — and re-measure. Only when the top three codes spread thin, none reaching half the gap, does A become the candidate, and then only with the quoted build scoped against a desktop parity regression suite.

Costs: 15 minutes per error sets the ceiling

The 3.1-point gap becomes a budget only after three inputs are substituted: monthly mobile submissions, monthly desktop submissions, and a loaded hourly rate. Write the formula down before plugging in numbers, because it shows which input does the work — excess annual errors equals mobile submissions per month times the gap times twelve. Everything downstream, hours and dollars and payback, is that one product scaled by minutes and rate. The desktop count never enters the multiplication; it sits underneath as the baseline the gap was measured against, and it is the input you cannot drop, because a 1.1% with no defined denominator leaves the 3.1 points undefined too. This arithmetic prices whatever gap the re-measurement produces. It does not validate it.

At 800 mobile and 4,000 desktop submissions a month, 800 × 0.031 = 24.8 excess errors per month, or 297.6 a year — call it 298. Note what that number is not: not 298 tickets, not 298 hours. It is the count of requests mobile produced above what the desktop baseline would have produced at the same volume.

Price each error at River's floor rather than at a full ticket. River's analysis of bad enterprise UX puts the invisible cost at more than 15 minutes per incident, hidden because nobody is measuring it. Use that floor: 297.6 × 15 minutes is 4,464 minutes, or 74.4 hours a year, which at a $45 loaded rate is roughly $3,350. That figure is a ceiling, and a generous one — it assumes a rebuild removes 100% of the excess errors, which no rebuild does. Realistic recovery is some fraction of $3,350 annually at this volume.

The break-even test kills most rebuild proposals outright. Divide the rebuild's total hours — design, build, QA, change management — by 74.4. That quotient is the payback period in years, stated at the impossible 100% elimination rate. For payback inside one year, total rebuild effort must come in under 74.4 hours. A rebuild consuming four times that needs four years at full elimination, and longer at any real one. Run the division before the design phase.

Sensitivity is linear, which is the useful part. Drop the reproduced gap to 1.0 point and the ceiling falls to about $1,080: 800 × 0.01 × 12 = 96 errors, 96 × 15 ÷ 60 = 24 hours, 24 × $45. Double mobile volume and the ceiling doubles. Change any one of the three inputs and re-run the arithmetic before committing hours to the form.

What 4.2% vs 1.1% cannot tell you

A raw 4.2% mobile error rate against 1.1% desktop looks like a mandate, but the honest question is what those two numbers survive once sampling noise is applied. Take one month: 800 mobile submissions with 34 errors, and 4,000 desktop submissions with 44 errors. The mobile rate is 4.25%, the desktop rate 1.1%, and the observed gap is roughly 3.15 points in mobile's disfavor. That is the point estimate. It is not the finding.

Computing the 95% confidence interval on that difference — using submission-level counts expanded for week-to-week variation, which real HR data always carries — yields a range of approximately 1.7 to 4.6 points. Read that carefully: the sign of the gap is stable. Mobile plausibly worse than desktop in every scenario the interval permits. The magnitude is not. An interval that includes 1.7 points also includes a gap barely above the 2.0-point action threshold, and one that includes 4.6 points implies nearly two and a half times that. The direction survives; the size does not.

That arithmetic is exactly what the rebuild rule exploits. A gap whose lower 95% bound clears 2.0 points is a gap you can budget against — you know the minimum problem you are fixing. A gap whose interval merely contains 2.0, as this one does, could justify a full Workday form rebuild or a two-hour triage session, and the data cannot tell you which. Only a re-measurement can: four weeks, one absence type, one time off plan, so the denominator cannot drift between populations.

Three conditions break the rule outright, and each is a check you can run before any re-measurement. First, if mobile accounts for under 10% of monthly requests, the mobile estimate is too thin for any interval to tighten enough. Second, if contract or temporary workers sit on a separate time off plan, their request flow can contaminate the mobile rate for reasons that have nothing to do with form usability. Third, if the numerator mixes policy rejections — a manager denying a request — with data-entry mistakes, the "error rate" measures approval behavior, not the interface.

The rule still wins when the same absence type and time off plan are measured on both channels, the mobile share of requests is healthy, and the numerator counts only submission failures. Under those conditions, a reproduced gap with a lower bound at or above 2.0 points converts a noisy operational statistic into a defensible trigger. Under any other conditions, the 4.2% figure is a hypothesis, not a business case.

So treat the existing numbers as a screening signal that has earned you one cheap next step: a four-week, single-plan, single-absence-type re-measurement with the interval arithmetic run on the result. Build only if that lower bound clears the threshold. The current gap's sign says mobile deserves scrutiny; only a reproduced, bounded gap can say it deserves a rebuild.

Three gates over 12 weeks

The schedule has three gates over 12 weeks, and the same copy-usable re-measurement worksheet feeds all three. Freeze the inputs before week 0: annual leave only, one population on one time off plan, and four weeks of baseline. The denominator is submitted requests; the numerator is requests with at least one validation or rejection event plus any manual correction logged within 24 hours. Desktop is counted in the same window, because a mobile change that lowers mobile errors while raising desktop errors is not a fix. The worksheet columns are fixed in advance so the week 4 and week 12 comparisons cannot be redefined after the numbers arrive.

Week 0 is the baseline gate. Compute the mobile-minus-desktop gap and its 95% interval from the four baseline weeks, then read the action threshold from the interval section. If the lower bound does not clear that threshold, stop and fix instrumentation before counting again; do not schedule a rebuild on the strength of a raw 4.2% versus 1.1% comparison. A 3.1-point observed gap with an undefined denominator is exactly the claim this gate is meant to test, not to repeat.

Weeks 1 through 4 are the live re-measurement, with no form changes, and week 4 is the reproduction gate. Run the same worksheet on the same absence type and time off plan. The rebuild is justified only if the lower 95% bound on mobile minus desktop is at least 2.0 points. If the lower bound falls short, the gap is not reproduced and the rebuild stops. If desktop’s interval moves outside its baseline interval in the same window, fix instrumentation before interpreting the mobile figure.

Weeks 5 through 12 belong to the rebuild and the post-fix gate. At week 12, rerun the identical worksheet. The mobile gap’s lower bound should now fall below the 2.0-point action threshold, and desktop must remain within its baseline interval. If mobile still clears 2.0 points, or desktop regresses, roll back and return to the worksheet rather than declaring the rebuild successful. The table is the checkpoint record: same fields, same population, same 24-hour correction rule, three decision points.

Five if/then gates before you rebuild

Before anyone touches the Workday time-off form, run five if/then gates in order. Gate one is the reproduction threshold: re-measure for four weeks on a single absence type and a single time off plan, compute the mobile-minus-desktop error gap, and authorize the rebuild only if the lower bound of the 95% confidence interval on that gap is at least 2.0 percentage points. If the lower bound lands below 2.0, stop. The original 4.2% versus 1.1% headline pair never authorizes spend on its own — fix the leading error code, wait out another four-week window, and re-run this gate on fresh data.

Gate two screens for error concentration. If one error code accounts for 50% or more of the excess errors, and that code traces back to a default absence type, an eligibility display rule, or an approver routing rule, change the configuration first and treat the form itself as unproven. A rebuild that ships on top of a misconfigured plan will reproduce the same errors on a new front end, and you will have paid for the privilege.

Gate three is a volume screen on the mobile side: if mobile accounts for under 10% of monthly time-off requests, the gap describes a small slice of your population, and a form rebuild is the wrong instrument. Gate four is the companion screen on absolute impact: if annual excess errors stay under the volume floor you set in advance — before looking at the data, so the floor is not negotiated to fit the answer — the exposure does not justify engineering time, whatever the percentage gap says.

Gate five is a validity check on the measurement itself. The re-measurement only counts if it was run on one absence type and one time off plan, with the numerator and denominator defined exactly as specified earlier in this piece. A four-week window that mixes vacation with a plan that has different approval routing produces a gap you cannot compare to anything, and a mixed-scope re-measurement fails this gate by definition.

The order matters. Gates two through four are screens that override the threshold: a re-measured gap can clear 2.0 points and still fail because one configuration rule drives half the errors, or because mobile is a rounding error in your request volume. Only when all four screens pass does the 2.0-point lower bound become an authorization. This mirrors the discipline the enterprise usability literature calls for — NN/g notes that group-level usability is less well defined and more variable than individual-task usability, and Microsoft's ESUS work exists precisely because enterprise metrics need specific, reliable evaluation. A variable metric demands a reproduction gate, not a single headline number.

What to do next

Frequently Asked Questions

What must the re-measured mobile-minus-desktop error gap clear before the mobile form is rebuilt?

The gap's lower confidence bound must be at or above 2.0 points.

Why can't the reported mobile versus desktop submission-error rates by themselves justify rebuilding the form?

The denominator behind those rates is undefined, so the raw gap alone cannot justify a rebuild.

How long must the re-measurement run?

The re-measurement must run four weeks.

What must the re-measurement be limited to?

It must be limited to a single absence type and a single time off plan.

How costly can poor enterprise usability be if it is not measured?

Per Nielsen Norman Group via River Group, poor enterprise usability costs more than 15 minutes per incident, and its cost is unseen without measurement.

Do poor-usability systems affect onboarding time?

Poor-usability systems show longer onboarding times than well-designed alternatives.

Quick answers

Also worth reading: How to pick budget planning software that fits your workflow: How to pick budget planning · Healthcare Staff Login Delays 2026: Workday Single Sign-On (SSO) Cuts 32% Tickets: Healthcare Staff Login Delays 2026: · Workday's 2026 CSM Ratio: 1:50 vs Industry 1:100 - Renewal Impact: Workday's 2026 CSM Ratio: 1:50