WOAAS: Workforce Outcomes as a Service

A company can spend years buying and configuring an HR system, then discover that the difficult questions are still waiting. Can its people deliver the strategy? Where does capability fall short? What would actually improve performance? The implementation may be complete while the business problem remains open.

That gap is a business opportunity. Employers could buy a continuously managed workforce capability service, informed by their own context and judged through demonstrated results. Call it WOAAS: Workforce Outcomes as a Service.

The radical premise: between 2026 and 2046, traditional HCM SaaS begins a long decline as AI makes more software functionality inexpensive to create, adapt and replace. This is a scenario explored here, not a forecast established by the evidence. The commercial proposition can also work if large SaaS platforms endure.

The decline of the configurable application

Enterprise software gave employers structure. It encoded processes, maintained records and reduced the need to resolve every operational detail themselves. Large configurable suites offered a way to cover many possible requirements before the customer knew which would matter.

The radical scenario is that AI progressively weakens the economics of that arrangement. If functionality can be generated and adapted around a requirement, buying a large application and then configuring the business into it becomes harder to justify. Interfaces and workflows could become temporary expressions of a need, assembled above durable services.

This is a claim about where value resides. Customers still need reliable execution. The commercial premium could move from the application bundle towards maintained business knowledge, trusted records, tested capabilities and responsibility for their operation. Subscription revenue may survive even as the product and basis of pricing change substantially.

Assured capabilities: integrity without unnecessary rigidity

Generating a useful interface is different from guaranteeing what happens behind it. An AI-built leave-request tool must still apply the correct entitlement, preserve permissions, update balances consistently and handle payroll consequences. The attractive destination is adaptable software built on components whose behaviour someone is accountable for maintaining.

Those components could combine authoritative records, versioned policy rules, access controls, tested transactions, audit trails and recovery mechanisms. AI could assemble or extend the experience around them within defined limits. Providers would sell maintained capabilities and operational assurance, with explicit responsibilities for changes and failures.

A prompt or reusable agent skill is not sufficient assurance. The service must establish which rules apply, what evidence supports an action and whether the assembled workflow preserves the required guarantees. Testing components individually does not prove that every composition is safe. Consequential changes need validation, monitoring and an accountable release process.

Legal and policy frameworks can inform these components, but vary by jurisdiction, contract and circumstance. An LLM's interpretation cannot itself establish compliance. Qualified owners must approve applicable rules and maintain them as conditions change.

This offers a plausible route to integrity with less application rigidity. It is not yet proof that arbitrary enterprise systems can be generated safely on demand. Persistent data models, integration obligations and coordinated transactions may keep substantial shared platforms valuable.

Three commercial destinations

These models can coexist. Buying assured capabilities does not automatically buy the management and delivery needed to improve workforce performance. A platform provider could supply excellent infrastructure while leaving the employer to diagnose needs, coordinate interventions and demonstrate value. That remaining responsibility is the WOAAS opportunity.

SAP illustrates the direction, not the conclusion

In its 6 October 2026 announcement, SAP describes Joule Assistants executing work using established data, authorisations and audit trails. Joule Studio lets customers build and extend skills and agents, while SAP Business Data Cloud and Knowledge Graph provide business context. SAP also describes connections with third-party agents. SAP's announcement.

This supports the proposition that incumbents are developing an intelligence and governance layer around business capabilities. It does not establish that their applications will disappear, that the announced services deliver every claimed benefit, or that they will take responsibility for workforce outcomes. SAP may retain a strong platform business underneath the new experience.

Over ten to twenty years, the radical possibility is that employers increasingly buy trusted execution and business capability, with software assembled as required. Whether incumbents capture that value through subscriptions, usage charges, managed services or performance-linked contracts remains an open commercial question.

Employers differentiate through context and execution

An exotic payroll configuration is rarely a persuasive competitive strategy. A reward design that attracts scarce capability, or an operating model that helps ordinary people perform exceptionally well, can be.

The employer contributes something no common platform can supply automatically: its priorities, customer needs, operating constraints, local relationships and choices about how work should be done. The provider must understand that context and translate it into practical support.

HR is critical where skills are scarce and distinctive. It is also critical where coordination, safety, trust, industrial relations or organisational change determine performance. Standardising repeatable work does not make those responsibilities disappear. It creates room for deeper expertise and more attention to the choices that matter.

The buying question becomes: what must this workforce be able to do, under which conditions, and what would count as convincing evidence?

What WOAAS would actually sell

Building on assured capabilities, the service would take responsibility for a defined cycle: establish the business objective, diagnose the constraint, develop or acquire capability, support its application, verify the result and revise the intervention.

A useful first contract might cover time to independent competence in a particular customer-service operation. A poor first contract would promise to transform the capability of every employee everywhere. The latter is easy to announce and almost impossible to price honestly.

The catalogue becomes a supporting resource

A catalogue can contain excellent material. Its breadth can also serve as insurance against uncertain needs: buy enough subjects and assume the required answer will be somewhere inside. That is a plausible buying logic, not evidence that all catalogue purchases failed. Coverage is easy to purchase; relevance, application and demonstrated capability remain largely the customer's problem.

AI creates an opportunity to replace that insurance with contextual support, provided the service can diagnose the need and test whether its intervention helps. A larger catalogue with better search alone leaves much of the original responsibility in place.

WOAAS starts with that problem. An AI coach uses approved knowledge and the person's demonstrated capability to explain, provide practice and offer feedback. A human specialist brings judgement, challenge or assessment when needed. Support appears within the task, with a clear route to help from a real person.

Evidence supports the possibility in bounded settings. A study of 5,179 customer-support agents found that a generative AI assistant increased issues resolved per hour by about 14%, with greater benefits for less experienced workers. It demonstrates a particular performance improvement, not proof that any AI coach will produce the same result. Generative AI at Work, NBER working paper.

A capability record should distinguish what has been inferred, practised, observed, assessed and authorised. It should also say where and when the evidence applies. Course completion is useful administrative information. It cannot establish that a person can safely perform an unfamiliar task.

The orchestration layer needs a delivery organisation

The technology connects objectives, workflows, knowledge, people, AI agents and evidence. It routes support and preserves permissions, source ownership, audit trails and escalation. It should work across the customer's existing systems wherever practical.

But orchestration software alone cannot carry the commercial promise. A delivery team must diagnose problems, assess practice, resolve exceptions and help managers change the work. Specialist analytics can reveal patterns. Interpreting their significance still requires operational knowledge and discussion with the people affected.

When performance falls short, the constraint may be staffing, poor equipment, conflicting incentives or an unreasonable process. A credible provider must be willing to recommend fixing those conditions. Selling more learning regardless of the diagnosis would defeat the model.

Start with one expensive, repeated problem

The most credible entry market is a recurring operational problem affecting a substantial cohort, with observable work and a business owner who has authority to act. Customer-service induction is one candidate. Similar opportunities could exist in recurring technical support or the introduction of a new operating process.

The buyer should be the operational leader accountable for the result, supported by HR, IT and finance. HR contributes capability and employee expertise; IT validates integration and security; finance tests the value calculation. Procurement needs a service contract that names responsibilities rather than a vague promise of transformation.

A paid discovery phase establishes the baseline and checks whether the client can supply the necessary context. A bounded pilot then compares new cohorts with an appropriate baseline or comparison group. Scale follows a demonstrated benefit. Discovery should be capable of concluding that this customer is unsuitable.

Start in one sector and a small number of related workflows. Expansion within a customer and into similar employers should reuse the assessment method, integration patterns and delivery playbook. Each new sector may require substantial new expertise. It should earn its own investment case.

Price responsibility, scope and verified contribution

Pure payment by results sounds attractive until a provider is asked to carry the risk of poor management, absent data or a business decision it cannot influence. A pure seat licence, meanwhile, rewards access rather than contribution. A hybrid contract is more credible.

• An implementation fee pays for diagnosis, approved integrations, the evidence baseline and service design.
• A recurring operating fee pays for agreed cohorts, workflows, response times, coaching and assessment capacity. Population bands help estimate cost; the offer is the complete service within that scope.
• A capped performance payment rewards independently verified improvement above an agreed threshold, subject to quality and employee safeguards.

Keep the first two sufficient to operate the service reliably. The provider should not need an uncertain bonus to fund essential support. Clients need clear limits for unusually high human demand and a transparent price for extending scope.

Contracts should address data access, the client's implementation obligations, material changes to the work, measurement disputes and an orderly exit. The employer retains its records, capability evidence and service history. Commercial dependence should arise from continuing value, rather than trapped data.

An illustrative annual contract

Design assumptions, not market prices or observed results. Consider a service covering 2,000 people in a defined operation. A £120,000 annual operating retainer plus £240 per covered person produces £600,000 recurring revenue. A separate £100,000 implementation fee pays for an assumed £100,000 of setup work. A performance payment of up to £60,000 is excluded from the base economics.

This is contribution, not operating profit. Sales, product development, central management, financing and tax still need funding. Client-specific technology costs must be included or disclosed separately. Existing customer software costs do not disappear from the employer's total cost calculation.

The sensitivity is revealing. If human coaching and assessment cost rises from £180,000 to £315,000, contribution falls to £45,000, or 7.5%. If that cost doubles to £360,000, contribution disappears. The business succeeds only if AI and reusable methods allow scarce experts to support more people without weakening the service.

The provider therefore needs to measure human hours per case, assessment effort, exception rates, onboarding cost and renewal. More AI interactions are useful only if they improve the economics and the outcome. A popular service that repeatedly exceeds its support capacity may be an unprofitable one.

The customer's arithmetic must also work

Suppose the operation recruits 300 people each year. Bringing independent competence forward by 15 working days, valued at an assumed £200 per productive day, gives a headline benefit of £900,000.

That figure is potential productive capacity. It is not automatically cash saved, and the service may not deserve credit for all of it. If only half is reliably attributable and economically usable, the benefit is £450,000. It does not cover a £600,000 annual fee, let alone implementation and the employer's own time.

The client then needs additional evidenced benefits, a lower service price or a different intervention. Avoided overtime, reduced rework or additional contribution from fulfilled demand might qualify. Salary costs already counted as productive capacity should not be counted again as savings.

A prudent design objective could be at least twice the recurring fee in conservatively validated annual value, with an acceptable payback on implementation. That is a proposed buying rule, not a universal benchmark. Finance should distinguish cash released, additional margin and capacity that management still has to use.

Evidence is part of the product

Outcomes need a baseline, an assessment standard and an agreed comparison. Where practical, use a phased rollout or comparison cohort. Account for changes in recruitment quality, workloads, managers and operating conditions. Pre-agree how benefits are valued and how disputes will be resolved.

A provider paid for fewer errors must not simply encourage fewer reports. A provider paid for faster competence must not quietly lower the threshold. Independent sampling, durable quality measures and employee feedback should constrain the performance payment.

Speed must be assessed alongside quality, workload and retention of competence. Employees need understandable records, appropriate access and a route to challenge consequential errors. Sensitive employment decisions remain with authorised people. The service must improve work without turning every action into a judgement about the worker.

AI performance also needs repeated evaluation. METR's early-2025 study found experienced developers took longer with AI assistance in its tested setting. Its February 2026 update discusses newer experiments and measurement difficulties, so the earlier finding cannot be treated as a permanent verdict on current tools. The durable lesson is to measure performance in the actual work. METR's productivity experiment update.

The opportunity exists alongside established competitors

Learning managed services and HR outsourcing already exist. Accenture offers learning design, delivery, analytics and ongoing management. Workday operates an ecosystem of service and technology partners. These offerings establish that WOAAS would enter an occupied market; they do not establish that this particular contract model is already delivered consistently. Accenture learning managed services; Workday partners.

The proposed distinction is a coherent service boundary around work: company context, diagnosis, practical support, verified capability and an explicit commercial link to business improvement. Competitors could adopt that model. A new name cannot protect it.

The defensible assets would be validated assessments, trusted employer relationships, effective sector-specific delivery methods, integration expertise and a record of producing results at a sustainable cost. Shared learning should come from reusable methods and legitimately aggregated evidence. It should not require pooling confidential customer data.

Who could build it?

HCM providers have relationships, workflow knowledge and infrastructure. They could become the lead operator, or coordinate accountable specialist partners. Doing so would require commercial and organisational change: delivery responsibility, specialist people and incentives that reward customer improvement even when it reduces licence usage.

Education groups such as QA or Lyceum could enter through coaching, assessment and employer relationships. They would need stronger access to operational context, integration capability and evidence of business contribution. A catalogue with an AI search box would fall well short of the proposition.

Consultancies and managed-service firms bring delivery experience but may carry cost structures designed for bespoke projects. Independent providers could work across platforms and sectors selectively. No entrant automatically possesses the whole model.

The employer still needs strong internal ownership. Strategy, culture, employee relationships and choices about acceptable risk cannot be delegated to a remote performance contract. WOAAS reduces the need to reproduce every specialist function internally while leaving those choices close to the business.

What would make the thesis fail?

If reliable enterprise software remains expensive to build and operate, the SaaS decline may be much slower than assumed. WOAAS could still exist, but its advantage would need to come from better service rather than collapsing software prices.

If customers cannot describe their needs, share usable context or release people to practise, the service may never reach its intended outcome. If every integration and assessment is unique, costs grow with each client and the model becomes conventional consultancy. If experts carry excessive caseloads, apparent efficiency can conceal weaker judgement.

Finally, buyers may prefer advice and tools to a shared service, especially where workforce capability is strategically sensitive. Some enterprises will build this operating capability internally. The addressable market is employers for whom specialist external delivery is more effective and economical, not every employer by default.

Those are tests to run, rather than objections to explain away.

A practical route from thesis to business

Begin with two or three design partners in the same sector. Agree one repeated problem, a baseline, access rights, delivery scope and a buying decision before starting. Test whether the service improves performance and whether its human support cost remains within the planned economics.

Earn renewal through an independently credible result. Then test whether the second and third clients can adopt substantially the same method at lower setup cost. Only after that evidence should the provider expand into adjacent outcomes or claim a broadly repeatable model.

Over the next five years, WOAAS could develop as an accountable layer around existing systems. In the following five to ten, providers could integrate more services and take responsibility for broader capability outcomes. Over ten to twenty years, under the radical scenario, software increasingly becomes included infrastructure within that relationship. These are possible stages, not a timetable guaranteed by technology.

The central proposition is worth pursuing: employers buy the continuing ability to build, demonstrate and apply the capability their business requires. As functionality becomes easier to generate, enterprise value could move towards assured execution, maintained business knowledge and accountable delivery. Software, AI and specialist people form the delivery machinery. A repeatable method and credible evidence make it a business.

For the companies currently selling HCM SaaS, the choice would be whether to protect the licence relationship or develop the organisation needed to take responsibility for more of the result.

Evidence and assumptions

This is an original business-model thesis by Dominic Chiappe, developed with OpenAI. The 2026 to 2046 SaaS decline is an explicit scenario. Prices, staffing costs, benefit values and commercial thresholds are illustrative assumptions, not verified market benchmarks. The assured-capability model is a proposed architecture and commercial direction, not a claim that LLMs can currently guarantee arbitrary enterprise software. Linked research supports specific observations about AI at work; supplier pages establish the existence of adjacent services, not their effectiveness or proof of the WOAAS model. No illustrative customer example describes Shell or another named employer's actual spending.

Updated 8 October 2026: clarified assured capabilities and the three business models, added SAP evidence and sharpened the catalogue argument.