# SAP Plans to Buy Workday's Skills Partner as Both Map Jobs > SAP's planned TechWolf acquisition meets Workday's new task-level Work Intelligence product. Their overlapping maps could guide hiring and training, but a task score alone cannot prove a worker moved or a budget paid off. - Published: 2026-10-15 - Author: Gene Dai - Canonical: [https://digidai.github.io/2026/10/15/sap-techwolf-workday-task-map/](https://digidai.github.io/2026/10/15/sap-techwolf-workday-task-map/) - Topics: Artificial Intelligence, Workforce, Organizational Design, Skills, SAP, Workday, Deep Investigation --- ![A layered paper worker holds five large task cards between rust and blue abstract maps, beneath the words INSIDE THE JOB and a small Gene Dai mark.](/images/articles/sap-techwolf-workday-task-map/cover-v1.jpg) _AI-generated editorial illustration._ At Workday Rising in Las Vegas on October 13, Workday [introduced Work Intelligence](https://en-gb.newsroom.workday.com/2026-10-13-Workday-Introduces-Work-Intelligence-to-Help-Customers-Design-Work-for-AI). The new product breaks jobs into tasks and connects those tasks to skills. It is available to early-access Workday HCM customers. Seven days earlier, SAP and TechWolf had [announced an agreement](https://news.sap.com/2026/10/sap-to-acquire-techwolf-evidence-based-work-age-of-ai/) for SAP to buy the Belgian startup, which also maps tasks and skills. They expect the deal to close in the fourth quarter if the required conditions are met. The purchase price was not disclosed. TechWolf was already working with Workday. In January 2025, Workday [said it would roll out TechWolf's skill inference](https://investor.workday.com/news-and-events/press-releases/news-details/2025/Workday-to-Roll-Out-TechWolfs-AI-Powered-Skills-Intelligence-Across-its-Workforce-Making-it-Easier-for-Employees-to-Drive-their-Career-Growth-01-29-2025/default.aspx) across a workforce then numbering more than 20,400, drawing on systems such as Salesforce and Jira. TechWolf became a Workday Innovation Partner. SAP's stated plan is to keep TechWolf available to customers using other software after the acquisition closes. For an employer with personnel records in Workday and work signals elsewhere, that future access is a contract question, not a footnote. Both vendors want to answer a decision that an org chart cannot: which work should a company hire for, teach to a current employee, or give to software? Their products put more detail on the table. Then the manager has to prove that the software works in this team. The reclaimed time needs a use. The employee needs an approved place to move. Each step sends a bill to a different desk. ## Workday knew the startup before SAP called Workday's 2025 partnership announcement described an employee-facing skills project. TechWolf would infer skills from existing work systems, let employees review suggested skills inside Workday HCM, and help match people to learning and opportunities. The release described a planned rollout to Workday's own workforce, not an audited count of employees who used it or moved jobs. Workday also put TechWolf's Skill Engine API on its marketplace. That history makes SAP's October purchase plan more than a routine addition to SuccessFactors. TechWolf's [Workday customer story](https://www.techwolf.ai/customer-stories/workday) reports that an early skills-based hiring effort cut time to hire by 15%, reduced the share of new hires performing below expectations by 39%, and shortened the time for account executives to reach a first deal by 14%. The supplier selected and published those figures. It does not provide an independent counterfactual for TechWolf's contribution. The three measures also stop at different points in a person's employment. A buyer can ask how Workday defined each cohort and measured each outcome; it cannot copy the percentages into its own investment case. The relationship creates a competitive complication. Workday has long owned the employee record and a large skills catalog; TechWolf offered inferred skills from the systems in which employees actually worked. In September 2025, Workday [listed TechWolf's Workforce Intelligence Agent](https://newsroom.workday.com/2025-09-16-Workday-Expands-Agent-Partner-Network-With-New-Ventures-Partners-to-Deliver-Enterprise-Ready-AI-Agents) among its partner agents for tasks, skills and jobs. Now Workday has a product addressing much the same question. Its public release does not say what happens to existing TechWolf contracts, marketplace listings or integrations if SAP completes the purchase. The buyer caught between these companies is not an abstraction. It may have payroll and worker profiles in one suite, project records in another, and a business-unit leader who trusts neither system's old job descriptions. The buyer wants the data joined because the systems answer different questions. The personnel record says who occupies a role. The workflow record shows some of what happened. The manager and employee can explain why it happened and which work would break if it stopped. A vendor can connect the first two. The third requires people to look at the map and disagree with it where necessary. SAP's [agreement](https://news.sap.com/2026/10/sap-to-acquire-techwolf-evidence-based-work-age-of-ai/) describes a current plan to keep TechWolf independent under founder Andreas De Neve and to serve non-SAP customers, subject to closing and required consultation. It also describes TechWolf as a future core of SAP SuccessFactors. A procurement team has to pin down today's product and access terms before debating that roadmap. The announcement contains no forced migration or exclusivity term. ## Two maps start from different systems Workday says Work Intelligence begins inside HCM, where companies already keep job, worker and skills data. It breaks a job into tasks, connects each task to needed skills, and tracks whether the mix changes as AI enters the work. Workday links it to Decision Intelligence for comparing internal development with outside hiring and to learning paths powered by Sana. The October announcement says the new product is available to early-access customers. It does not publish a customer count, price list or independently measured outcomes for Work Intelligence. Its larger numbers belong to an older foundation. Workday says more than **3,450 customers** use Skills Cloud, which has an ontology of more than **75,000 canonical skills**. Neither figure counts Work Intelligence deployments. This matters because a large skills dictionary and a current map of tasks performed by a specific team are different things. The dictionary can give the map a language. It cannot, by itself, reveal whether an analyst spends Tuesday closing books, correcting customer data or resolving an exception no job description mentions. There are at least three possible records under the label "job map." A job architecture says what the company expects a role to contain. A workflow log records activity in connected systems. A validated task sample asks the people doing the job what actually takes time and judgment. Their disagreements are information. If a role description says forecasting occupies half a finance analyst's time but the team's month-end evidence shows repeated reconciliation work, a budget proposal based only on the description begins in the wrong place. If a workflow log shows many clicks but the employee says they are repeated checks forced by poor data, a model may treat a defect as a durable task. TechWolf describes its own [Work Intelligence product](https://www.techwolf.ai/product/task-intelligence--2026) as joining HR records with workflow signals and external labor-market information. It says it can infer the tasks inside a role, estimate where AI may automate or assist, and return that view to existing systems, including Workday, SAP, ServiceNow and planning dashboards. These are product descriptions, not an independent audit of how accurately the models reconstruct work. They also raise a practical question for an employer: which workflow traces are relevant to a job, and which merely show activity inside a tool? An employee may use Jira to record an engineering issue but settle the hardest design choice in a conversation. A customer-service agent may close a ticket only after deciding whether a policy exception is fair. A task model that sees the ticket but misses the decision can overcount routine work. A manager may then pay to automate the visible step while leaving the hard part of the job untouched. This is a risk inferred from the data sources the vendors describe. It is not an observed defect in either named product. The difference between the approaches is also less clean than a buyer might expect. TechWolf says its insights can flow into Workday. Workday's new product starts from its own skills and worker context while describing changing tasks. A company could have both, use only one, or test them against a manually observed workflow. The public documents do not establish feature-by-feature parity, data portability or the eventual licensing terms after an SAP acquisition. A comparison should therefore begin with the buyer's existing data and a concrete decision, rather than a catalog of vendor nouns. For a pilot, the comparison can be narrower than an enterprise procurement. Select one recurring process and ask each system to identify the same ten or twenty high-volume tasks. Have workers mark missing tasks and false matches. Then have the process owner choose which tasks could be tested with a tool without changing a person's employment terms. This does not produce a universal vendor ranking. It reveals whether the two maps agree where the proposed investment actually lands, and whether either can show the source of a disputed inference. Without that check, a polished visualization may simply hide disagreement between systems. | Publicly described layer | Workday's October product | TechWolf in SAP's plan | What a buyer still has to verify | | ------------------------ | ---------------------------------------------------- | ------------------------------------------------------- | -------------------------------------------------------------- | | Starting record | Workday HCM people, jobs and skills context | Connected HR, business and workflow systems | Which events represent real work, and which teams are missing? | | Task view | Links tasks in a job to skills and changes over time | Models work, tasks, skills and the outside labor market | Can workers and managers correct an inferred task? | | Current status | Work Intelligence in early access | Existing TechWolf product; SAP acquisition pending | Which features and terms are available now? | | Claimed action | Compare hiring, learning and role redesign options | Support planning, redeployment and skill development | Did an approved role, pay or output actually change? | ## A task score cannot close a job A task map invites a shortcut: score five tasks for automation, add the scores, and turn the total into a headcount target. The available research does not support it. Theoretical exposure describes what a system might do under stated conditions. Usage shows that someone tried a tool. Reliable completion asks whether the result was good enough; economic adoption asks whether the full process, including review and exceptions, costs less. A change in a person's job requires an employer decision after those tests. The [International Labour Organization's 2025 task-based index](https://www.ilo.org/publications/generative-ai-and-jobs-refined-global-index-occupational-exposure) estimated that about **one in four workers globally** held an occupation with some generative-AI exposure. Only **3.3% of global employment** fell into its highest exposure category. The report's unit is exposure, not layoffs. It says most occupations contain tasks requiring human input, making job transformation the more likely route than full replacement. Its global model cannot tell a particular bank that it should remove a specific analyst position. Anthropic's [June 2026 Economic Index](https://www.anthropic.com/research/economic-index-june-2026-report) compares different ways of describing the same task: what an LLM could theoretically do, what researchers observe being done with Claude, and what surveyed users believe AI can do. The measures do not coincide. Theoretical exposure tends to exceed reported exposure; reported exposure exceeds observed Claude use in its sample. The survey itself overrepresents Claude users and some occupations, so it is not a census of work. That disagreement is a useful warning for any vendor that turns a predicted task share into a staffing plan. The gap can be wider for physical work. Anthropic's [September 30 robotics study](https://www.anthropic.com/research/what-work-can-robots-do) estimates that existing robots can perform about **74% of physical tasks** under at least some conditions, but are cost-competitive with human labor for only **0.3% of job tasks**. Robots and language models are different technologies, and those percentages cannot be transferred to Workday or TechWolf. The study does show why task capability and a viable operating decision need separate columns. A machine that works in a specially prepared warehouse is not automatically a cheaper substitute at every site. The [ILO's June 2026 review of empirical studies](https://www.ilo.org/publications/impact-genai-jobs-productivity-and-work-organization-review-empirical) offers another restraint. Worker-reported time savings of a few percent of working hours had not consistently turned into higher measured output, earnings or employment in the studies reviewed. Productivity gains were uneven and often unverified. If an employer buys a map in order to capture savings, it must trace what happens after a task becomes faster: whether the backlog falls, customers wait less, errors decline, workers handle more cases, or the released time is simply absorbed by other work. This distinction changes the executive conversation. A task map may be good enough to choose a pilot and still be too weak to approve a staffing reduction. The latter requires evidence about the whole process, including the work that arrives after automation makes one step faster. A manager may discover that the review queue grows or that a specialist becomes the bottleneck. The apparent saving then shifts labor rather than eliminating it. A credible business case counts that labor at its new location. It also records when the comparison period began, because a busy season or changed customer mix can make a pilot look successful or unsuccessful for reasons unrelated to the tool. ## The employee still owns an incomplete picture The employee's profile is a good place to test the map. An inferred skill can be wrong, stale or too broad. Workday's [TechWolf rollout description](https://investor.workday.com/news-and-events/press-releases/news-details/2025/Workday-to-Roll-Out-TechWolfs-AI-Powered-Skills-Intelligence-Across-its-Workforce-Making-it-Easier-for-Employees-to-Drive-their-Career-Growth-01-29-2025/default.aspx) gave employees a way to review and update suggested skills. TechWolf's [acquisition statement](https://www.techwolf.ai/press/sap-to-acquire-techwolf) says employees can validate inferred skills. The releases do not say how often a worker makes a correction or whether the correction changes a later hiring or training decision. The stakes rise when a task map becomes a resource map. A worker who does informal mentoring, catches a safety concern, or repairs a customer's trust may leave little trace in a ticket system. Another worker may generate many visible tickets because a broken process forces repeated corrections. Counting the first person as underused and the second as an easy automation target would confuse record density with value. A buyer needs a way for teams to contest task definitions before they become the basis of a role redesign. It also needs to tell workers which data systems feed the model and which decisions the outputs may influence. This is a business issue as well as an employee issue. If employees do not trust the profile, managers may maintain separate spreadsheets and the system becomes another expensive layer. If they cannot see an internal opening or obtain a manager's release, a correct skill match still fails to produce mobility. Workday's [October 5 workforce report](https://en-sg.newsroom.workday.com/2026-10-05-Workday-Global-Workforce-Report-AI-Is-Rewriting-Jobs-More-Than-Its-Cutting-Them) said internal moves fell at **57% of employers** in its matched customer analysis. That figure is not a result of Work Intelligence, which was announced later. It is a reminder that visibility and approved movement are different denominators. The same report described workers naming paused hiring, unsupportive managers and unfair selection as obstacles to moving internally. That is a reminder to ask what happens when the map finds a match. Will the worker see an open role? Can a manager release them? Is there paid time to learn a skill that the target role needs? Will a rejected candidate be told which part of the profile was wrong? Each answer belongs to an operating policy outside the task-inference model. A company can improve the model while leaving these constraints untouched; it should measure both changes rather than credit one for the other. ## From a map to a budget decision _Illustrative scenario: this analyst example adapts Workday's product description. It is not a reported customer case or meeting._ Imagine a finance analyst role with reconciliation, exception handling, forecasting and manager explanation. Workday itself uses a financial analyst as an example of a person who could shift from manual reconciliation toward forecasting. A company could test the narrower proposal: can a tool complete a defined set of routine reconciliations at acceptable error rates, and what becomes of the analyst's hours after the pilot? The project has several bills. IT pays for integration and software. The business unit pays for changed processes and error review. The learning team pays for forecasting training. A manager must open or approve a role in which the new skill is used. Finance has to choose a result it can inspect, such as a shorter reconciliation cycle or a lower contractor bill. If those owners never meet, the dashboard may show a promising task score while the payroll and customer numbers stay where they were. ![On a worktable, a worker and manager examine oversized task cards on one paper island; an incomplete bridge leads to separate training and open-role folders.](/images/articles/sap-techwolf-workday-task-map/interior-v1.jpg) _AI-generated editorial illustration. The gap between the task cards and the training and role folders represents a decision that still needs approval._ The first purchase question is therefore not which vendor has the most skills in an ontology. It is whether the provider can show, for one role and one site, the source of each important task inference, the worker's corrections, the real software deployment, and the downstream business measure. The second question is commercial: what does the employer pay for the task map, what additional systems or services are required, and what happens to access to that data if the supplier changes ownership? Neither Workday's announcement nor SAP's acquisition release provides a general price. A buyer must get it in the actual proposal. The proposed TechWolf transaction makes an exit question unusually concrete. SAP says TechWolf should keep serving non-SAP customers, but future terms have not been published in the acquisition announcement. A Workday customer that has already invested in TechWolf skill profiles can ask whether its current integration, correction history and inferred task data can still be used and exported under the contract it signs. That is a due-diligence question, not an allegation that SAP intends to cut access. The answer can affect the cost of a workforce plan even if the underlying model stays accurate. Procurement should also avoid a false binary. SAP says TechWolf should continue serving non-SAP customers. Workday still has its own HCM and task product. A company running Workday might reasonably test TechWolf, Workday Work Intelligence, or a manual baseline against the same role before changing a workforce plan. A customer of SAP SuccessFactors may make a different integration choice. The acquisition does not, by itself, prove a product winner or a migration deadline. ## A ledger for actual role change A useful pilot file can fit on one page if every column has an owner. It starts with a job that exists now, records the tasks a worker and manager agree are real, and ends with a dated outcome. The middle matters most. A capability prediction needs a real deployment; a deployment needs review; a course needs an open role if the employer intends to count mobility. Finance should recognize savings only when a comparable cost actually falls or output rises at acceptable quality. | Decision stage | Evidence to keep | Owner and denominator | Failure signal | | ---------------- | --------------------------------------------------------------------------- | ----------------------------------------------------------- | --------------------------------------------------------- | | Define the work | Observed task sample, source system, worker correction and manager sign-off | Role owner; workers sampled / workers in role | The most consequential exceptions never appear in the map | | Test the tool | Completed tasks, error and escalation rates, time including review | Process owner; eligible tasks, not all job hours | Fast completions require a larger repair queue | | Decide the route | Approved automation, augmentation, training or hiring choice and budget | Business unit and finance; roles reviewed | No approved opening, trainer or process change exists | | Move the worker | Training time, verified skill use, job change, pay and retention | HR and manager; eligible workers / workers who changed role | A match or course is counted as a filled role | | Count the return | Output quality, cycle time, customer result and actual spend | Finance and process owner; comparable pre/post cohort | Vendor usage is reported as booked savings | This ledger is an editorial operating proposal, not a feature claimed for either product. It is designed to keep units from leaking across a presentation. If the task model identifies 1,000 exposed tasks, the buyer still needs to know how many were tested. If 100 were tested, how many passed quality review at the actual site? If work hours were released, who used them and for what? A buyer cannot answer those questions with a single exposure percentage or a skills-catalog count. The file should also carry dates. The task sample belongs to a specific week, not to the job forever. The tool test needs a beginning and end, a comparison workload, and an exception log. The training measure needs an enrollment date and a later check of actual skill use. For an internal move, count the worker only after an approved role change appears in the personnel record; keep a second measure for employees who learned a skill but did not move. This prevents a vendor dashboard from treating a prediction, a completed course and a payroll change as one continuous success. It also lets the employer see where the chain breaks. The Workday and TechWolf materials point to different possible first tests. Workday's own case centers on skills-based hiring and a sales ramp. A customer that wants to replicate it should ask for the cohort definition behind the **15%**, **39%** and **14%** figures, the earlier comparison group, and how Workday measured performance. A manufacturer considering task redesign should instead start with a specific operation and include rework, safety and supervisor time. A service team might count resolved cases and customer complaints. The right denominator follows the work, not the vendor's favored chart. There is also a sequencing question. A task map can expose a gap before the company has budget to respond. Publishing an employee-facing career path without an approved role may raise expectations that the manager cannot meet. The October 5 Workday mobility findings describe paused roles and manager support as obstacles reported by workers. The employer should set the approval path before advertising a role match as a career outcome. An employee who learns a new skill and waits behind a hiring freeze has gained something, but the company has not completed the promised move. ## A filled role is the test SAP has announced a conditional acquisition, not a completed integration. Workday has introduced Work Intelligence to early-access customers, not reported a broad deployment result. TechWolf has published selected customer outcomes, but those figures do not determine what a different employer will save. The public record is strongest on product direction and weakest on the conversion from inferred work to approved, paid change. The contest will become easier to judge when a customer can show the whole chain for a defined group: tasks confirmed by workers, a tool that works at acceptable quality, a budget that survives review, and people who move into roles where their new skills are used. The next decisive meeting is mundane. A manager will have to approve a vacancy or a redesigned position, and finance will ask whether the pilot changed a real cost or result. The map can inform that meeting. The signed decision and the worker's next job will show what it achieved. ## Continue reading - [Workday's Mobility Agent Meets a Hiring Pause](https://digidai.github.io/2026/10/09/workday-mobility-agent-internal-career-freeze/): Follow the gap between an internal skill match and an approved role change. - [Inside NEC's AI Department, Where People Still Decide](https://digidai.github.io/2026/08/31/nec-ai-department-four-management-layers/): See how a named employer reorganized work and retained human management decisions. - [When internal mobility can beat another external search](https://digidai.github.io/2026/03/27/from-talent-acquisition-to-talent-readiness-why-internal-mobility-is-overtaking-external-hiring/): Compare a task map with the older internal mobility decision and its cost tradeoff. - [AI Builder Ads Grow While Early Careers Move to Contracts](https://digidai.github.io/2026/10/13/ai-builder-ads-early-career-contract-paths/): Separate new skill demand in job ads from paid work and career progression.