Workday · Antidote

HCM and finance system of record

Hire-to-pay workforce administration and adjacent finance on one cloud platform. The employer owns the records. Direct AI substitution of payroll and the org chart fails; Sana and Illuminate-era agents are an overlay. Oracle, SAP, and Microsoft are the named invasion surface.

As of 24 August 2026. This memo covers hire-to-pay workforce administration — core HR, time, payroll, and the manager or employee transactions that keep the org chart and pay file current. Recruiting, Financials close, Adaptive Planning, and the Sana or Illuminate-era agent overlay sit on the same platform and are discussed, not scored. No live-product test has been run.

The employer owns the records. Oracle, SAP, and Microsoft are the invasion surface.

Workday hosts hire-to-pay and adjacent finance on one cloud platform. Workforce and financial records are typically the employer's. There is no exclusive corpus. Direct AI substitution of the system of record fails: the job is statutory pay, tax, benefits, write-authority, and audit, not generation. The plausible bind is configured business processes, payroll calendars and history, tax and benefits integrations, identity, and a system-integrator rebuild. That bind is unmeasured until export and migration exist.

Sana, Illuminate-era agents, Paradox, and Flex Credits are a later treatment. Vendor hour-saved and '100% of the time' figures remain vendor statements. Workflow capability and operating economics remain open without an independent product test. The right comparison is a named HCM/payroll peer with a declared AI overlay and the same statutory budget — not a raw chat model. Oracle and SAP are the primary named enterprise peers. Rippling and Deel are the new-suite hire-to-pay peers; they are not ChatGPT. Microsoft is both a named competitor and a fiscal-2026 Agent System of Record partner. That is the bundling exposure, not a legal-model product replacing payroll.

Exposed vs scarce. Workforce and financial records are the employer's. Overlay labor — parse, commentary, contract extraction — is reconstructable. Statutory pay is not a generation task.
Still untested. Statutory-error rate and any Sana increment on the same tenant with agents off.
The comparison that would change the call. Workday versus Oracle Cloud HCM or SAP SuccessFactors with the same payroll budget, Sana recorded on or off — not a chat window.

Decision questions

If payroll already runs on Workday, the question is not whether a pay-period gate exists. It is whether Sana or Illuminate-era agents change statutory-error rate, exception load, or write-authority risk on the same tenant — and what a named HCM peer would do on the same cases.
What isolated increment, if any, comes from Sana or Illuminate-era agents versus Workday without those agents, each changed one at a time?
What is full cost per accepted pay run or HR transaction, including SI hours, Flex Credits, failed transactions, and payroll-ops minutes — not company-level subscription revenue?
Which ingredients are exclusive or costly to reconstruct, and which a competent Oracle, SAP, UKG, Dayforce, Rippling, or Deel setup already supplies?
Where can AI compress labor minutes on top of the system of record without impersonating pay, tax, or write-authority?

Hire-to-pay fails in specific ways: wrong net pay, missed tax or garnishments, failed time capture, a ghost or duplicate worker, a fluent exception that should have blocked the run, and a write that the operator did not authorize. Those classes stay separate. A correct paycheck with the wrong costing string is a different error than a missed payday with a good journal. Hosting the org chart is not payroll accuracy.

The job splits. Statutory pay, tax, garnishments, and write-authority have no public capability floor — no published pack measures a payday. They do have public labor-cost and cycle-time context: BLS May 2025 occupational wages, SHRM 2026 median nonexecutive time-to-fill of 39 calendar days, and Hackett 2025 relative close-cycle and planning-spend figures. The overlay labor also has published floors: parse, close commentary, contract extraction, conversational follow-up, and occupation-shaped finance and admin work. Issuer revenue, year by year, bounds peer scale; it is not HCM share. Those numbers bound the labor market and what a generic system already does on that task type. They are not a Workday score, not a Sana increment, and not payroll accuracy. Sana, Illuminate-era agents, Paradox, and partner agents are a later treatment. If they cannot be turned off in the comparison, say so and leave attribution unproven.

Workflow breakdown

Each step of the job is mapped to a published benchmark or another reputable quantitative measure only where the mapping is honest. A public figure is a floor or a labor-market context on that task type. It is never a score of this product.

Task	Published measure	Pack fit	What it does not measure
Statutory pay, tax, garnishments, and write-authority	None	No honest public floor	The scored hire-to-pay job. No public pack measures a payday.
Parse invoices, tax forms, contracts, and close packs	ParseBench	Pack fit strong	Five parse metrics. Not payroll accuracy and not Workday.
Close commentary and pay-file questions on a report	TAT-QA, FinQA	Pack fit strong	The reconstructable finance-ops layer. Public financial QA is not Workday days-to-close or journal write-authority.
Conversational HR or finance assistant on the same documents	ConvFinQA	Public floor only	Follow-up questions. Not manager self-service write-authority.
Planning workbooks and Adaptive Planning models	None	No honest public floor	SpreadsheetBench and AIDABench stay held out on license. Relative planning spend is not Adaptive Planning quality.
Recruiting screen, resume match, and conversational ATS	SHRM 2026 Recruiting Executives Benchmarking: median time-to-fill 39 calendar days for nonexecutive positions (4,600+ organizations)	No honest public floor	Industry cycle time, not Paradox or HiredScore quality. No license-cleared resume pack in this catalog.

Findings at a glance

Capability, cost, and defensibility are assessed separately. They are not averaged into a score. Where evidence is missing, the finding stays open.

Status	Area	Finding
Open	Workflow capability	Hosting the hire-to-pay record is not payroll accuracy, not recruiting quality, and not time-to-close. No independent Workday test has been run. Sana and Illuminate-era agent outcome claims remain vendor-declared.
Open	Operating economics	Company-level subscription and professional-services revenue are disclosed. Full cost per accepted pay run or finance close, including SI hours and failed transactions, is unmeasured. Flex Credits are a commercial term.
Judgment	Defensibility	Workforce and financial records are typically the employer's. Direct AI substitution of the system of record fails because the job is statutory pay, tax, audit, and write-authority. Reconstructing configured processes is a migration. Illuminate, Sana, and partner agents are a later treatment. Oracle, SAP, and Microsoft remain the named enterprise set. Rippling and Deel are new-suite hire-to-pay peers, not chat-model substitutes. No public bake-off against those suites exists here.
Vendor-stated	Vendor claims and open questions	The fiscal 2026 Form 10-K describes HCM, Financial Management, Spend, Planning, Sana, Paradox, Flex Credits, and a professional-services ecosystem. Illuminate is a fiscal 2025 announcement; later agent outcome claims are vendor declarations, not independent test results.

Workflow capability

Hire-to-pay is the job. Adjacent modules are not the scored workflow.

The priority workflow is workforce administration as the employer system of record: hire, core HR, time, payroll, and the manager or employee transactions that keep the org chart and pay file current. The fiscal 2026 Form 10-K describes that hire-to-retire HCM suite, plus Financial Management, Spend Management, and Planning on one platform. Those adjacent modules matter commercially. They are not the job this report scores.

Recruiting — HiredScore, Candidate Experience Agent, and Paradox Conversational ATS — is a sourcing and candidate-experience overlay on the same tenant. Financials close, consolidation, and Adaptive Planning are finance-ops jobs that share the platform. Scoring them as if they were hire-to-pay would hide the failure that matters: a wrong payday. No independent Workday test on any of those jobs has been run.

Hosting the record is not payroll accuracy

Users under contract, Fortune 500 presence, and transaction volume describe an input surface. They do not describe statutory-error rate, time-to-close, recruiting quality, or human correction on a frozen set. The fiscal 2026 Form 10-K reports more than 75 million users under contract, more than 65 percent of the Fortune 500, and more than 11,500 customers. The fiscal 2026 earnings release adds more than 7,000 core Financial Management and HCM customers and 1.7 billion AI actions. Those are issuer disclosures. They are not a pass rate.

A system of record can be the employer's write-authority and still fail the cases that matter: a pay variant that should have blocked the run, a configuration gap that only appears at quarter-end, or an exception the operator accepted because the agent sounded sure. Those classes stay apart.

Sana and Illuminate-era agents are a treatment, not the product

The fiscal 2025 Form 10-K announced Workday Illuminate as AI built into the platform. The fiscal 2026 Form 10-K does not use that name. It introduces Workday Sana as the knowledge, agent, and automation experience, and it records fiscal 2026 acquisitions of Flowise, Paradox, Sana, and Pipedream. The 16 September 2025 newsroom still sold Illuminate agents for performance, close, payroll, and planning, with new agents 'available in 2026.' Product pages now frame Sana as the orchestrator of agents such as Payroll Agent. Branding moved. The question did not: the overlay is still a later treatment on the system of record.

Vendor outcome claims remain vendor statements. The September 2025 newsroom states Contract Intelligence reduced execution time by 65 percent, Frontline cut staffing-change time by up to 90 percent, Financial Audit saved early-access customers up to 900 hours per year, Payroll Agent enabled compliance up to 4 times faster, and Planning Agent reduced exploration by 30 percent, or about 100 hours per month. The Payroll Agent product page says the agent 'gets payroll right 100% of the time' and cites McKee Foods figures. Those sentences are marketing. They are not an on/off comparison. The same product page still requires human oversight. Until the same tenant, same cases, and same graders exist with agents off, any Illuminate or Sana increment is not established.

Published benchmarks

Published packs bound parse and public finance-ops QA. They are not payroll accuracy, not a Sana increment, and not Workday.

ParseBench

Invoices, tax forms, contracts, and close packs fail first at parse. Tables, charts, faithfulness, formatting, and grounding stay separate. Parse quality is not payroll accuracy.

Third-party leaderboard · not an Antidote run · not a product score

Current frontier VLMs on ParseBench

Official ParseBench leaderboard, observed 13 August 2026. Five dimensions kept separate. Gemini 3 Flash is Thinking High. GPT-5.6 Sol and Fable 5 are Reasoning None / default API settings in the source.

Dimension	Gemini 3 Flash	GPT-5.6 Sol	Claude Fable 5	Claude Opus 4.8
Tables	91.5%	89.34%	89.79%	89.65%
Charts	64.79%	59.57%	52.21%	49.75%
Faithfulness	90.87%	86.17%	90.02%	89.02%
Formatting	68.31%	50.42%	72.62%	71.38%
Grounding	59.77%	25.09%	49.24%	18.69%

Third-party leaderboard · not an Antidote run · not a product score

Specialist parsers vs a frontier VLM

Same ParseBench leaderboard. LlamaParse Agentic, Pulse Ultra 2, and Reducto Agentic against Gemini 3 Flash Thinking High. The source Overall column is omitted.

Dimension	LlamaParse Agentic	Gemini 3 Flash	Pulse Ultra 2	Reducto Agentic
Tables	90.74%	91.5%	75.45%	80.42%
Charts	78.11%	64.79%	90.82%	73.4%
Faithfulness	89.68%	90.87%	79.49%	86.37%
Formatting	85.24%	68.31%	73.05%	57.6%
Grounding	80.62%	59.77%	66.56%	67.07%

TAT-QA

Hybrid table-and-text QA bounds close commentary and pay-file questions that sit on a report. Arithmetic is the weak slice. Not statutory pay.

Paper table · Zhu et al., 2021 · not an Antidote run · not a product score

TAT-QA test exact match and F1

Zhu et al., ACL 2021, Table 3, test split. TAGOP is the paper system. Human expert is 50 hybrid contexts (301 questions).

System	Exact match	F1
Human expert	84.1%	90.8%
TAGOP	50.1%	58%
NumNet+ V2	37%	46.9%
TaPas WTQ	16.6%	22.8%
BERT-RC	9.1%	18.7%
HyBrider	6.3%	7.5%

Paper table · Zhu et al., 2021 · not an Antidote run · not a product score

TAGOP by answer type on the test set

Zhu et al., ACL 2021, Table 4. Arithmetic is the weak slice. Metrics stay separate; there is no composite.

Answer type	Exact match	F1
Span	54.1%	67.9%
Spans	60%	75.1%
Counting	62.5%	62.5%
Arithmetic	42.5%	42.5%

FinQA

Executable numerical reasoning over financial reports bounds calculation-heavy close and planning work. Program accuracy stays separate from execution accuracy.

Paper table · Chen et al., 2021 · not an Antidote run · not a product score

FinQA test execution and program accuracy

Chen et al., EMNLP 2021, Table 2. FinQANet (RoBERTa-large) is the paper system. Human expert execution accuracy is 91.16%; the non-expert crowd is 50.68%.

System	Execution accuracy	Program accuracy
Human expert	91.16%	87.49%
FinQANet RoBERTa-L	61.24%	58.86%
FinQANet BERT-L	53.52%	51.62%
Crowd	50.68%	48.17%
NeRd BERT-base	48.57%	46.76%
Longformer base	21.9%	20.48%

ConvFinQA

Conversational follow-up over the same report is closer to an HR or finance assistant than single-turn FinQA. Still not write-authority.

Paper table · Chen et al., 2022 · not an Antidote run · not a product score

ConvFinQA execution and program accuracy

Chen et al., EMNLP 2022, Table 3. FinQANet (RoBERTa-large) is the paper system. Human expert is 200 sampled questions. GPT-5.6 and Fable 5 are not in this table.

System	Execution accuracy	Program accuracy
Human expert	89.44%	86.34%
FinQANet-Gold RoBERTa-L	77.32%	76.46%
FinQANet RoBERTa-L	68.9%	68.24%
FinQANet RoBERTa-base	64.95%	64.16%
FinQANet BERT-L	61.14%	60.55%
T5 large	58.66%	57.05%
GPT-2 medium	58.19%	57%
FinQANet BERT-base	55.03%	54.57%
Crowd	46.9%	45.52%

Paper table · Chen et al., 2022 · not an Antidote run · not a product score

FinQANet RoBERTa-large by conversation slice

Chen et al., EMNLP 2022, Table 4. Hybrid second-part turns are the weak slice. Metrics stay separate.

Slice	Execution accuracy	Program accuracy
Number selection	82.54%	82.34%
Simple conversations	72.37%	72%
Full test	68.9%	68.24%
Hybrid first part	68.11%	66.54%
Program questions	62.14%	61.26%
Hybrid conversations	60.99%	59.7%
Hybrid second part	52.38%	51.43%

How to read the published measures

Published packs bound parse and public finance-ops QA. They do not measure statutory pay and they are not a Workday score. Issuer revenue bounds peer scale over time; the calendars are not aligned. Sana and Illuminate-era agents, if later measured, still need the system of record underneath and an on/off comparison.

The comparison is Oracle or SAP, not a chat window

The failure mode that matters for Workday is treating an overlay pack — or an occupation Elo — as payroll accuracy. Statutory pay is not on those charts. The comparison is a named HCM peer, Sana recorded on or off.

The interesting paired tests are the processor-and-spreadsheet floor versus a declared generic HCM, and that generic HCM versus Workday, each with the same pay-period set, statutory key, and payroll budget. A product that only ships the last comparison cannot claim the increment. UKG and Dayforce are admissible named peers for workforce administration. Rippling and Deel are the admissible new-suite peers for mid-market and global payroll; they are not a straw man and they are not ChatGPT. Sana, Illuminate-era agents, Paradox, Flowise builders, and partner agents are later treatments. If they cannot be turned off, say so and leave attribution unproven.

Operating economics

Company revenue is not cost per accepted payday

The fiscal 2026 earnings release reports $9.552 billion total revenue, $8.833 billion subscription revenue, and $719 million professional-services revenue, against $728 million professional services in fiscal 2025. Those are company-level figures. They do not isolate cost per accepted pay run, HR transaction, or finance close. National occupational wages give labor-cost context, not tenant unit cost. BLS May 2025 Occupational Employment and Wage Statistics put U.S. payroll and timekeeping clerks at a mean of $59,630 a year (153,140 employed) and human resources specialists at a mean of $81,990 (912,430 employed). Those are labor-market means. They are not Workday seat price, not cost per accepted payday, and not a Sana increment. Flex Credits, introduced in fiscal 2026 and sold in the September 2025 newsroom as a fungible AI and platform allowance, are a commercial term. They are not a unit-cost measurement.

The 10-K states that third parties provide a majority of deployment services and that a professional-services ecosystem of Workday consultants and system integrators is part of adoption across the subscription life. Company professional-services revenue is not the customer's SI bill. Full cost still has to include subscription, Flex Credits, partner payroll where used, SI and internal payroll-ops hours, and failed transactions in the denominator. Any leverage case that treats AI as seat compression remains modeled and assumption-bound. Do not infer NPV, churn, or adoption from subscription scale. Issuer revenue, year by year: Workday rose from $6.216 billion in fiscal 2023 to $7.259 billion, $8.446 billion, and $9.552 billion in fiscal 2024–2026. Paycom rose from $1.694 billion in 2023 to $1.883 billion in 2024 and $2.052 billion in 2025. Paylocity rose from $1.175 billion in fiscal 2023 to $1.403 billion, $1.595 billion, and $1.771 billion in fiscal 2024–2026. Those calendars are not aligned, and none of them is HCM share. Website-visit series are not used here as a share proxy.

Defensibility

The employer owns the records. Direct substitution fails. The overlay is the opportunity.

Workforce, payroll, and financial records are typically the employer's. Workday is a custodian. Responsible AI pages say Workday never shares customer data to train third-party public models, and they claim ISO 42001 and NIST AI RMF certifications. The administrator guide last updated 12 June 2026 requires feature-level opt-in of tenant data for named ML features — time anomaly, pay anomalies, skills, recruiting, Career Hub, journal insights — and states that Pay Anomalies models fail to build without historical payroll contribution. Planning Agent telemetry, Workday says, does not retrain or fine-tune the pre-trained foundation model and can be disabled in Adaptive Planning. That narrows reuse. It is not a default cross-customer flywheel, and it is not 'never uses customer data.' Executed MSA, DPA, and tenant opt-ins remain controlling. A trillion annual transactions, repeated on the Responsible AI page and in the September 2025 newsroom, is volume. Volume is not a measured quality loop.

Direct AI substitution of hire-to-pay fails because a generic model cannot assume statutory pay, tax, benefits, or write-authority. The 10-K itself names competition from generic LLMs and general-purpose agents and says domain and industry expertise in financial management and HCM may inhibit new entrants. Inhibition is not impossibility. Reconstructing configured processes, payroll history, and integrations is a migration. Hard is not shown to be impossible.

Reproducibility is a named HCM peer. Microsoft is partner and exposure.

The right comparison is Oracle Cloud HCM or SAP SuccessFactors — the 10-K's named primary competitors — or UKG or Dayforce, with the same payroll and tax budget and a declared generic AI overlay. ADP is both a named competitor and a disclosed global payroll partner, with Remote and Strada. A raw chat window is the wrong floor. The floor is spreadsheets plus a standalone processor. The interesting paired tests are that floor versus a declared generic HCM, and that generic HCM versus Workday, each with Illuminate or Sana recorded on or off.

New-suite peers are a third comparison, not a substitute for Oracle or SAP at the Fortune 500 core. Rippling ships a unified HR, IT, and finance workforce platform and announced Rippling AI on 18 March 2026 as an overlay that can query live company data and stage payroll and access changes for approval. Deel's 20 October 2025 Series E post says it had already surpassed $1 billion ARR, processes $22 billion in payroll a year, and serves 35,000-plus customers and 1.5 million-plus workers in 150-plus countries. Those are vendor statements of scale. They are not a Workday bake-off, not Fortune 500 share, and not payroll accuracy. Paycom reported $2.052 billion of FY2025 revenue; Paylocity's fiscal 2026 10-K reports $1.771 billion of revenue and about 44,400 U.S. clients. Issuer revenue is a size floor. It is not HCM market share. Oracle and SAP do not break out Cloud HCM in a way this report can treat as a share cell.

Microsoft is named as a competitor in the same 10-K sentence as ServiceNow, and as a fiscal-2026 partner that lets customers register Microsoft-built agents in the Workday Agent System of Record. That split is the invasion surface: Viva, Dynamics, Copilot, and ServiceNow HR can attach generation and case management to seats the employer already pays for, while the Microsoft partnership is also a distribution channel into Workday's agent register. Platform bundling is the material exposure. Customer insourcing is staying on Oracle or SAP, or keeping payroll on ADP. AI opportunity is labor-minute compression on exceptions, recruiting screen, HR case, close evidence, and planning commentary — on top of the system of record. It does not retire the system of record. Do not price that opportunity as NPV from attach rates.

Defensibility

Workday's load-bearing questions are hire-to-pay write-authority, export, and Oracle, SAP, or Microsoft bundling. Workforce records are usually the employer's. That is not a data moat. Sana is an overlay.

Dimension	Rating	Judgment	Missing evidence
Data rights	Exposed	Workforce, payroll, and financial records are typically the employer's. Workday is a custodian. Published policy bars sharing customer data to train third-party public models; feature-level opt-in still contributes tenant fields to named ML features. Custodianship without exclusive reuse rights is not an exclusive asset.	Executed MSA, DPA, and the target tenant's ML data-contribution opt-ins.
Exclusivity	Exposed	Hire-to-pay, core HR, time, payroll, and adjacent finance are category features shared with Oracle, SAP, UKG, Dayforce, ADP, Rippling, and Deel. There is no exclusive corpus. The differentiator, if any, is configured processes, payroll history, and integrations, which are customer-specific.	What, if anything, cannot be reproduced in a named HCM/payroll peer.
Rebuild difficulty	Conditional	A generic HCM can host workers. Reconstructing configured business processes, payroll calendars and history, tax and benefits integrations, and identity is a migration. Hard is not impossible. Sana and Illuminate-era agents are a different treatment and are not this reconstruction.	Export completeness, payroll-history fidelity, integration rebuild, and timed SI hours.
Freshness	Open	Pay-file and org-chart currency are controlled inside each customer's tenant. Opt-in ML contribution and Planning Agent telemetry that Workday says does not retrain the foundation model do not establish a cross-customer quality loop. Transaction volume is not freshness.	Statutory-error latency, opt-in rates, and any measured quality increment from contribution.
Switching cost	Conditional	Manager self-service, payroll calendars, benefits, and identity integrations can bind daily HR and finance paths. That is the plausible switching-cost story. It stays contingent until export and migration are observed.	Switching interviews, migration time, and metadata or history loss on export.
System of record	Conditional	For many employers Workday is the hire-to-pay write-authority. Payroll may sit on ADP, Remote, or Strada. Finance may remain on Oracle or SAP. Sana and agents must not be allowed to impersonate record authority.	Write-authority map across Workday, partner payroll, general ledger, and identity.
Integrations	Conditional	Identity, tax, benefits, and payroll-partner connectors create gravity. Microsoft is both an Agent System of Record partner and a bundling threat. ADP is both a payroll partner and a named competitor.	Integration inventory and a named-peer paired comparison.
Network effects	Exposed	An HCM does not get better because another employer uses it unless lawful cross-tenant reuse and a measured quality increment exist. Feature-level opt-in narrows that path. A trillion transactions is volume, not a flywheel.	Rights, contribution rates, and a paired quality result when contributed fields are added.
Trust and compliance	Conditional	Employers pay for statutory pay, tax, audit, and access control. Responsible AI pages claim ISO 42001 and NIST AI RMF. The 10-K records FedRAMP authorization. Those claims need an audit of the target tenant, not a brochure.	Current report scope, tenant configuration, statutory-error history, and incident history.
Distribution	Conditional	Direct enterprise sales, Workday GO for mid-market, and an SI ecosystem are a specialized channel. They do not block Oracle, SAP, Microsoft, or ServiceNow from attaching HCM, finance, or agent features to existing seats, and they do not block Rippling or Deel from winning a mid-market or global-payroll bake-off that has not been run here.	Channel concentration and bundling exposure, separately from payroll quality.
Services dependence	Open	The 10-K states that third parties provide a majority of deployment services and that consultants and system integrators remain part of adoption across the subscription. Fiscal 2026 professional-services revenue was $719 million. Specialist services are structurally part of deploy and switch; customer-level intensity is unmeasured.	Target SI hours, payroll-ops FTEs, escalation rate, and five-year total cost.
Rate of improvement	Open	No measured improvement curve on statutory-error or time-to-close rates is in this report. Weekly updates and major releases are a delivery cadence, not a quality series.	Dated payroll-error and close-cycle series on a frozen key.

How AI can compete

Replacement, a new AI product, a platform bundle, cheaper labor, and a customer building it themselves are different risks. They are not averaged into a score.

Mechanism	Assessment
Direct replacement	A generic model cannot assume statutory pay, tax, or write-authority. Substituting the interface on exported extracts is easier than substituting the pay engine. Substituting configured hire-to-pay is a migration, not a prompt.
New AI product	Two different invasions. Recruiting agents, HR-case assistants, close-evidence collectors, and planning commentary attack labor on top of the system of record; Paradox and specialist recruiting AI are this surface. Rippling and Deel are a different object: new-suite hire-to-pay and global payroll that can host the record. Neither invasion is a chat window running payroll, and neither has a public bake-off here.
Platform bundle	The material exposure. Oracle and SAP are the named primary competitors. Microsoft is a named competitor and a fiscal-2026 Agent System of Record partner. ServiceNow HR can attach case and generation to seats the employer already pays for.
Labor compression	The AI opportunity. Agents can compress payroll-exception, recruiting-screen, HR-case, close-evidence, and planning minutes. That is value-creation on top of the system of record. It does not retire the need for write-authority, and it is not priced as NPV here.
Customer builds it	A large employer can remain on Oracle or SAP, or keep payroll on ADP, and attach its own agents. The binding constraint is migration: configuration, payroll history, integrations, and SI hours — not model skill.

Data flywheel

A flywheel needs every step below. On this evidence it is unproven. Having data is not the same as being allowed to reuse it, and reuse is not the same as a measured quality gain.

Link	Status	Note
Unique data	Vendor-stated	Workforce exhaust is the employer's. Workday's exhaust, if any, is telemetry, configuration, and opt-in ML fields.
Permission to reuse	Open	Published policy bars sharing customer data to train third-party public models. Feature-level opt-in contributes tenant data to named ML features. Executed terms remain controlling.
Reliable feedback	Open	No audited statutory-error or exception labels.
Proven quality gain	Open	No paired SoR-only versus Sana or Illuminate result.
Better outcomes	Open	No frozen payroll statutory-error or time-to-close result.
More usage or data	Open	Retention from process embed is unobserved. Customer and user counts are scale, not a data loop.

What public sources resolve

Public filings and vendor documentation can settle some structural questions. They cannot replace a test on the live product.

Question	Conclusion	Still open
Does Workday own an exclusive workforce or financial corpus?	Workforce, payroll, and financial records are typically the employer's. Workday hosts hire-to-pay and adjacent finance. Hosting is not exclusive ownership.Form 10-K for the fiscal year ended January 31, 2026; Responsible AI	Executed MSA and DPA language on custody versus ownership for the target tenant.
Can a generic model replace Workday as the hire-to-pay system of record?	Direct substitution of statutory pay, tax, benefits, and write-authority fails. The 10-K names generic LLMs and agents as a competitive vector and says domain expertise may inhibit new entrants. Inhibition is not a measured test. The invasion path is a named HCM peer or a platform attaching an AI overlay, not a chat window running payroll.Form 10-K for the fiscal year ended January 31, 2026	A paired Oracle, SAP, UKG, or Dayforce cell on the same pay-period set, and any Microsoft or ServiceNow bundle the target already pays for.
Is customer tenant data used to train shared or third-party models?	Workday says it never shares customer data to train third-party public models. The administrator guide requires feature-level opt-in of tenant data for named ML features, including pay anomalies, and states that Planning Agent telemetry does not retrain the foundation model. Executed terms and tenant opt-ins remain controlling.Responsible AI; Reference: Workday AI Data Contributions	The target's executed agreement, Innovation Services or UMSA elections, and which ML categories are opted in.
Do Sana or Illuminate-era agents produce a measured increment over Workday without agents?	Illuminate was announced in the fiscal 2025 10-K. Fiscal 2026 introduced Sana and recorded Flowise, Paradox, Sana, and Pipedream acquisitions. The September 2025 newsroom and Payroll Agent page publish hour-saved and '100% of the time' claims. No independent on/off comparison has been run.Form 10-K for the fiscal year ended January 31, 2025; Form 10-K for the fiscal year ended January 31, 2026; Workday Illuminate Expands with New AI Agents for HR, Finance, and Industry; Payroll AI Agent	The same tenant, same cases, and same graders with Sana or Illuminate-era agents off, then on.
Is switching cost low because records can be exported?	The 10-K describes configured applications, a professional-services and system-integrator ecosystem, and third parties performing a majority of deployment. That makes a low-services, copy-the-records exit an unsupported assumption. Migration completeness is unmeasured.Form 10-K for the fiscal year ended January 31, 2026; Fiscal 2026 Fourth Quarter and Full Year Financial Results	A timed migration covering configuration, payroll history, benefits, integrations, identity, and SI hours.
What is full cost per accepted pay run or HR transaction?	Fiscal 2026 company reporting gives $9.552 billion total revenue, $8.833 billion subscription, and $719 million professional services. Flex Credits are a commercial AI and platform term. None of those figures is cost per accepted payday.Fiscal 2026 Fourth Quarter and Full Year Financial Results; Form 10-K for the fiscal year ended January 31, 2026; Workday Illuminate Expands with New AI Agents for HR, Finance, and Industry	Subscription, Flex Credits, partner payroll, SI hours, payroll-ops minutes, and failed transactions in one denominator.
Do a trillion annual transactions create a cross-customer AI flywheel?	Transaction volume is a vendor declaration. Feature-level opt-in and a bar on training third-party public models narrow lawful reuse. No measured quality increment from contribution is in this report. Volume is not a flywheel.Responsible AI; Reference: Workday AI Data Contributions; Workday Illuminate Expands with New AI Agents for HR, Finance, and Industry	Opt-in rates, contributed fields, and a paired quality result when those fields are added and nothing else changes.
Is Microsoft only a partner, or also an invasion surface?	The fiscal 2026 10-K names Microsoft as a competitor alongside ServiceNow and as a partner that can register Microsoft-built agents in the Workday Agent System of Record. Partnership does not cancel bundling exposure.Form 10-K for the fiscal year ended January 31, 2026	Named bake-offs and the target's existing Microsoft 365, Dynamics, or ServiceNow estate.
What public labor-cost and cycle-time figures bound hire-to-pay and the adjacent overlay?	BLS May 2025 occupational wages bound U.S. labor-cost context for payroll clerks, HR specialists, accountants, financial managers, compliance officers, and administrative services managers. SHRM 2026 reports a 39-day median nonexecutive time-to-fill. Hackett 2025 reports Digital World Class close cycles 35–57% shorter than peers and 57% less planning-and-forecasting spend. Those figures are not Workday scores, not payday accuracy, and not a Sana increment.Occupational Employment and Wage Statistics, May 2025, Table 1; 2026 Recruiting Executives Benchmarking: Attracting Critical Talent; Digital World Class Finance Teams Operate at 45% Lower Cost and Deliver Faster, Smarter Insights	Tenant-level payroll-ops minutes, statutory-error rates, and an on/off Sana comparison. National wages and industry cycle times do not substitute.
Do Rippling, Deel, or other new-suite peers displace Workday as the hire-to-pay system of record?	Rippling and Deel are admissible new-suite hire-to-pay and global-payroll peers, not chat-model substitutes. Deel states $1B-plus ARR and $22B payroll processed as of its October 2025 Series E post. Workday issuer revenue rose from $6.216B in fiscal 2023 to $9.552B in fiscal 2026. Paycom rose from $1.694B in 2023 to $2.052B in 2025. Paylocity rose from $1.175B in fiscal 2023 to $1.403B, $1.595B, and $1.771B in fiscal 2024–2026. No public payday bake-off exists. Web traffic is not market share.Our Series E: Building the global infrastructure of work; Rippling AI: Built to Do the Work, Not Just Talk About It; Paycom Software, Inc. Reports Fourth Quarter and Year-End 2025 Results; Paycom Software, Inc. Reports Fourth Quarter and Year-End 2024 Results; Paylocity Announces Fourth Quarter and Full Fiscal Year 2024 Financial Results; Form 10-K for the fiscal year ended June 30, 2026; Semrush Traffic Analytics for myworkday.com, workday.com, rippling.com, gusto.com, and hibob.com; Archived Semrush monthly snapshots for myworkday.com and rippling.com; Form 10-K for the fiscal year ended January 31, 2026; Fiscal 2026 Fourth Quarter and Full Year Financial Results; Fiscal 2024 Fourth Quarter and Full Year Financial Results	A frozen pay-period set against a declared Rippling or Deel tenant, and a rights-cleared HCM share series that is not website traffic.

Required comparisons

A decision-grade test would run these side by side. None has been run here.

Comparison	Note
Current workflow	Requires imported observations of the incumbent hire-to-pay and payroll-ops process. Not assumed from an HRIS brochure.
Spreadsheet and processor floor	Spreadsheets, a standalone payroll processor, and mailbox workflows without a unified HCM/finance platform.
Named competitor	A declared Oracle Cloud HCM or SAP SuccessFactors setup, or UKG, Dayforce, Rippling, or Deel, with the same payroll and tax budget. Not a raw chat model.
This product	Needs authorized Workday tenant access and exportable transaction traces. Illuminate on or off must be recorded.
Isolated changes	Illuminate or Sana on versus off; Workday payroll versus an ADP or other partner payroll; Financials attached versus HCM-only. Bundled agents stay unlabeled.

Open claims

Each claim stands on its own. They are not averaged into a moat score. Hire-to-pay outcome and price wait on a live-product test. Illuminate and Sana stay unproven without an on/off comparison. The employer owns the records. Vendor hour-saved figures are not measured results. Subscription scale is not adoption, churn, or NPV.

Claim	Status	What is still missing
Workday payroll is more accurate than a generic HCM or a processor-only floor.	Open	A frozen pay-period set, independently recalculated net-pay keys, and a paired named competitor.
Illuminate agents produce a measured increment over Workday without agents.	Open	An on/off comparison on the same tenant, same cases, same graders. Vendor hour-saved figures do not substitute.
Workforce and financial records are typically the employer's, not Workday's exclusive corpus.	Judgment	Executed MSA, DPA, and any ML data-contribution opt-ins for the target tenant.
Switching cost is low because records can be exported.	Open	A timed migration covering configuration, payroll history, benefits, integrations, identity, and SI hours.
Oracle, SAP, or Microsoft bundling risk is low.	Open	Named competitive bake-offs and channel exposure. The FY2026 10-K names those vendors as competitors; Microsoft is also a disclosed Agent System of Record partner.
Rippling or Deel is displacing Workday as the hire-to-pay system of record.	Open	Win/loss on a frozen pay-period set, segment (mid-market versus Fortune 500), and a named-suite paired comparison. Web traffic is not that test.
A trillion annual transactions create a cross-customer AI flywheel.	Open	Lawful reuse of contributed fields, a measured quality increment from contribution, and opt-in rates. Volume is not a flywheel.

What this report does not conclude

Payroll accuracy, statutory-error rate, or replacement of the current hire-to-pay process
A measured Sana or Illuminate increment over Workday without agents
Full cost per accepted pay run, HR transaction, or finance close
A data moat; workforce and financial records are typically the employer's
A cross-customer AI flywheel from transaction volume or opt-in ML contribution
A switching-cost, Microsoft-bundling, or services-dependence figure
AI opportunity as NPV, seat compression, or revenue-at-risk
Win/loss versus Oracle, SAP, UKG, Dayforce, ADP, Rippling, Deel, or a Microsoft or ServiceNow bundle
HCM market share from website visits

What would settle these questions

Authorized Workday tenant access with exportable transaction traces and a frozen price card, including Flex Credits.
A sealed pay-period and HR-transaction set with independently recalculated net-pay, tax, and write-authority keys.
Imported observations of the current hire-to-pay process; not a product tour.
A declared Oracle Cloud HCM or SAP SuccessFactors setup — or UKG, Dayforce, Rippling, or Deel — same cases, same payroll and tax budget.
Sana or Illuminate on versus off on the same tenant; Workday payroll versus partner payroll; Financials attached versus HCM-only.
Export and migration completeness: configuration, payroll history, benefits, integrations, identity, and SI hours.
Executed MSA, DPA, and ML data-contribution opt-ins before any reuse or flywheel claim.
Payroll-ops, HR-ops, and reviewer-minute measurement, including failed transactions.

Sources

Form 10-K for the fiscal year ended January 31, 2025Workday, Inc. · Public-company disclosure · 11 March 2025
Workday announced Workday Illuminate in fiscal 2025 as AI built into the platform and described role-based agents and a Workday Assistant expected to become generally available in fiscal 2026, plus a February 2025 announcement of agents for contracts, payroll, financial auditing, and policy, and the Workday Agent System of Record.
Superseded on branding and scale by the fiscal 2026 10-K, which does not use the Illuminate name. Useful for the overlay's announcement history, not as current scale or an independent test.
Form 10-K for the fiscal year ended January 31, 2026Workday, Inc. · Public-company disclosure · 6 March 2026
Workday describes a cloud suite of HCM, Financial Management, Spend Management, and Planning with more than 75 million users under contract, more than 65 percent of the Fortune 500, and more than 11,500 customers. Fiscal 2026 introduced Workday Sana, Flex Credits, Workday Build, Workday GO expansions, and a Microsoft partnership to register Microsoft-built agents in the Workday Agent System of Record. Fiscal 2026 acquisitions were Flowise, Paradox, Sana, and Pipedream. Payroll partners include ADP, Remote, and Strada. Primary competitors are Oracle and SAP; the named set also includes Anaplan, ADP, Coupa, Dayforce, Microsoft, NetSuite, ServiceNow, and UKG, plus generic LLMs and agents. Third parties provide a majority of deployment services. Domain expertise in HCM and financial management 'may inhibit new entrants.'
Issuer disclosure. Scale, partnerships, and competitive naming are not payroll accuracy, switching cost, or a measured AI increment.
Fiscal 2024 Fourth Quarter and Full Year Financial ResultsWorkday, Inc. · Public-company disclosure · 26 February 2024
Workday reports fiscal 2024 total revenues of $7.259 billion and subscription services of $6.603 billion, against fiscal 2023 total revenues of $6.216 billion and subscription services of $5.567 billion.
Company-level results. They are not HCM-only share and not cost per accepted payday.
Fiscal 2026 Fourth Quarter and Full Year Financial ResultsWorkday, Inc. · Public-company disclosure · 24 February 2026
Workday reports fiscal 2026 total revenue of $9.552 billion and subscription revenue of $8.833 billion, against fiscal 2025 total revenue of $8.446 billion and subscription revenue of $7.718 billion. Professional-services revenue was $719 million against $728 million in fiscal 2025. It reports more than 11,500 customers, including more than 7,000 core Financial Management and HCM customers, general availability of Sana Core and Sana Enterprise, and 1.7 billion AI actions across the platform in fiscal 2026.
Company-level results. They do not isolate cost per accepted pay run, SI hours at the customer, or any agent increment. AI-action counts are a vendor activity metric.
Paycom Software, Inc. Reports Fourth Quarter and Year-End 2024 ResultsPaycom Software, Inc. · Public-company disclosure · 12 February 2025
Paycom reported full-year 2024 total revenues of $1,883.2 million, up 11.2 percent from $1,693.7 million in 2023.
Issuer results. Calendar year, not Workday's January fiscal year. Revenue is not HCM market share.
Paycom Software, Inc. Reports Fourth Quarter and Year-End 2025 ResultsPaycom Software, Inc. · Public-company disclosure · 11 February 2026
Paycom reported full-year 2025 total revenues of $2,051.7 million, up 9.0 percent from $1,883.2 million in 2024.
Issuer results for a public HCM/payroll peer. Different fiscal year and mix than Workday. Revenue is not HCM market share.
Paylocity Announces Fourth Quarter and Full Fiscal Year 2024 Financial ResultsPaylocity Holding Corporation · Public-company disclosure · 1 August 2024
Paylocity reported fiscal 2024 total revenue of $1,402.5 million, up 19 percent from $1,174.6 million in fiscal 2023.
Issuer results. Fiscal year ended 30 June. Revenue is not HCM market share.
Form 10-K for the fiscal year ended June 30, 2026Paylocity Holding Corporation · Public-company disclosure · August 2026
Paylocity reports fiscal 2026 total revenues of $1,771.3 million, up 11 percent from $1,595.2 million in fiscal 2025, and says it provided software to approximately 44,400 U.S. clients as of 30 June 2026, excluding clients acquired through acquisitions.
Issuer results. Client count is not Fortune 500 share and not a Workday comparison. Typical client size is 10 to 5,000 employees.
Occupational Employment and Wage Statistics, May 2025, Table 1U.S. Bureau of Labor Statistics · Official statistic · April 2026
May 2025 OEWS reports 153,140 payroll and timekeeping clerks (mean $59,630; median $28.01 an hour), 912,430 human resources specialists (mean $81,990; median $36.51), 1,449,500 accountants and auditors (mean $94,750; median $40.23), 841,710 financial managers (mean $186,910; median $80.08), 417,070 compliance officers (mean $88,400; median $38.81), and 263,960 administrative services managers (mean $129,870; median $54.87).
National occupational means. They are labor-cost context around hire-to-pay and adjacent finance, not Workday tenant cost, not payroll accuracy, and not a Sana increment.
Responsible AIWorkday · Official policy
Workday says it never shares customer data to train third-party public models, gives customers visibility and control over data use, and claims ISO 42001 and NIST AI Risk Management Framework certifications. The same page says its AI is fueled by more than 1 trillion transactions a year.
Vendor policy and certification claim. It does not establish tenant configuration, executed terms, or a quality increment from transaction volume. 'Third-party public models' is narrower than all model training.
Reference: Workday AI Data ContributionsWorkday Administrator Guide · Official product documentation · 12 June 2026
Workday documents feature-level opt-in of tenant data for named ML features across HCM, payroll, financials, spend, and people experience. Pay Anomalies requires contribution of historical payroll categories or the models fail to build. Planning Agent operational telemetry, Workday says, is not used to retrain or fine-tune the pre-trained foundation model and can be disabled in Adaptive Planning.
Administrator documentation, not executed customer terms. Opt-in is not a measured cross-customer quality flywheel, and it is not a blanket bar on all uses of customer data.
Workday Illuminate Expands with New AI Agents for HR, Finance, and IndustryWorkday · Official product documentation · 16 September 2025
Workday announced Illuminate agents for HR, financials, and industry, said new agents would be available in 2026, and introduced Flex Credits as a subscription consumption model for agents and platform innovations. It published vendor outcome claims including Contract Intelligence minus 65 percent execution time, Frontline up to 90 percent less staffing-change time, Financial Audit up to 900 hours per year, Payroll Agent compliance up to 4 times faster, and Planning Agent minus 30 percent exploration or about 100 hours per month. It repeated a trillion-transactions claim and more than 11,000 organizations.
Vendor newsroom. Outcome figures are early-access or unspecified customer claims, not an independent test. Customer count is superseded by fiscal 2026 reporting. Unreleased features are subject to the release's own forward-looking disclaimer.
Payroll AI AgentWorkday · Official product documentation
Workday markets Payroll Agent as identifying data and configuration issues inside the customer's Workday environment, with human oversight and control. The page claims the agent 'gets payroll right 100% of the time,' frames Sana as the orchestrator of expert agents including Payroll Agent, and cites McKee Foods figures of $2.6 thousand per ad hoc report and root-cause analysis cut to under a minute.
Marketing page. '100% of the time' is not an independent test. Named-customer figures are vendor-presented and not independently adjudicated. Human oversight remains required on the same page.
Rippling AI: Built to Do the Work, Not Just Talk About ItRippling · Official product documentation · 18 March 2026
Rippling announced Rippling AI on 18 March 2026 as an overlay on its unified HR, IT, and finance platform. The post says the agent writes queries against live company data, inherits existing roles and permissions, and stages payroll, access, and org-chart changes for human approval rather than writing them unchecked.
Vendor product page. It describes a new-suite overlay, not a measured increment over Workday, and it is not issuer financials. Third-party Rippling ARR estimates are not used in this report.
Our Series E: Building the global infrastructure of workDeel · Vendor-controlled profile · 20 October 2025
Deel announced a $300 million Series E at a $17.3 billion valuation and said it had earlier in 2025 surpassed $1 billion ARR, processes $22 billion in payroll annually, and serves 35,000-plus customers and 1.5 million-plus workers in 150-plus countries.
Vendor blog. ARR and payroll volume are not GAAP revenue, not Workday HCM share, and not a payday bake-off. Employer-of-record mix is a different job than enterprise hire-to-pay.
Digital World Class Finance Teams Operate at 45% Lower Cost and Deliver Faster, Smarter InsightsThe Hackett Group · Third-party assessment · 9 June 2025
Hackett's 9 June 2025 release states that Digital World Class finance organizations have 35–57% shorter close cycles than peers and spend 57% less on planning and forecasting, while investing more in business analysis.
Relative industry study of a self-defined top-quartile cohort versus peers. It is not Workday Financials days-to-close, not Adaptive Planning quality, and not an independent product test.
Semrush Traffic Analytics for myworkday.com, workday.com, rippling.com, gusto.com, and hibob.comSemrush · Third-party assessment · 12 August 2026
Semrush reports myworkday.com visits of 62.59 million in May, 63.33 million in June, and 69.16 million in July 2026. The public six-month chart tooltip on the same page is labeled February 2026 and reads 69,467,761 visits (42,292,650 desktop and 27,175,111 mobile). workday.com was 5.76 million in May and 5.35 million in June. rippling.com was 11.67 million, 12.33 million, and 15.27 million across May, June, and July. gusto.com was 9.20 million in May and 9.32 million in June. hibob.com was 4.17 million in May and 4.61 million in June. July cells for workday.com, gusto.com, and hibob.com were not shown on the public pages. Earlier months in the public month picker require a Semrush login.
Third-party modeled visits, not a census. The February 2026 figure is a chart tooltip on the July 2026 public page, not a dedicated February overview. Marketing sites and tenant hosts are different objects. Login-mixed domains are not comparable 1:1 with workday.com. Visits are not seats, not payroll volume, and not HCM market share. SimilarWeb's live pages are captcha-walled.
Archived Semrush monthly snapshots for myworkday.com and rippling.comSemrush, retrieved via Internet Archive · Third-party assessment · 24 March 2023
Archived Semrush public pages report myworkday.com at 70.8 million visits in January 2023, 66.1 million in February 2023, 86.06 million in January 2025, 76.54 million in February 2025, and 59.3 million in December 2025. The February 2025 page's trailing three months include December 2024 at 77.07 million. workday.com was 4.16 million in December 2023, 5.95 million in January 2024, 5.28 million in June 2024, 5.67 million in August 2024, 5.06 million in December 2024, 5.15 million in June 2025, and 5.5 million in August 2025. rippling.com was 4.01 million in June 2024 and 8.66 million in December 2025. gusto.com was 9.71 million in December 2025. February 2024 for myworkday.com, and June 2023 and June 2025 for rippling.com, were not on a public snapshot retrieved for this report.
Third-party modeled visits captured on later archive dates. Semrush revises its model, so year-over-year cells are same-month estimates, not a census. Visits are not seats and not HCM market share. The June 2026 rippling.com cell and the February 2026 myworkday.com chart tooltip are on the current Semrush pages, not these archives.
2026 Recruiting Executives Benchmarking: Attracting Critical TalentSociety for Human Resource Management · Third-party assessment
SHRM's 2026 recruiting executives brief, based on data from more than 4,600 organizations, reports a median time-to-fill of 39 calendar days for nonexecutive positions and says 97 percent of nonexecutive positions were filled externally, up from 93 percent in 2025.
Industry association survey on the public brief. Time-to-fill is not Paradox or HiredScore quality, not quality-of-hire, and not a Workday recruiting score. Dollar cost-per-hire figures are not taken from this public page.