HR teams are being asked to do more than ever. Everest's Tech Provider Spotlight explores how HR Service Delivery is evolving. Read Now.
The average company in Okta’s 2025 dataset used 101 apps, a milestone that matters because employee experience breaks down not when one system is poor, but when ten systems each demand a different login, vocabulary, handoff, and proof of context. In parallel, Microsoft’s 2025 data showed workers being interrupted every two minutes during the working day, with 275 interruptions a day for the most pinged users, while Microsoft’s 2026 Work Trend Index argues that people’s agency is rising faster than organisations are redesigning work to support it. In other words: employees are ready for AI-enabled self-service, but most operating models still behave as if the portal is the product. It is not. The connected experience is the product (Okta).
That changes the brief for senior HR leaders. Integration is now an employee UX problem, an identity problem, and an AI measurement problem at the same time. If your payroll, benefits, knowledge, case management, and workflow tools are technically integrated but experientially disjointed, employees still experience failure. If your AI front door can answer but cannot authenticate, fulfil, and preserve context across handoffs, it is only a prettier dead end. And if you measure usage instead of outcomes, you can scale disappointment with impressive dashboards (NIST).
The strongest practical recommendation is this: stop organising integration around systems and start organising it around continuities. In this article, I support a six-step playbook for designing a single employee journey across many systems, with AI as an orchestration layer rather than another destination.
The future of HR service delivery is not a single front door, but a governed constellation of front doors. That is a more honest description of how work now happens across Teams, email, mobile, intranets, chat, case systems, and embedded workflow moments. The controversy is not whether there should be one entry point; it is whether employees should have to care which one they used. They should not.
If you ask most HR and IT leaders whether their landscape is integrated, they will point to middleware, APIs, SSO, maybe SCIM, and a tidy architecture slide. Employees will point to something else entirely: whether they had to re-explain themselves, whether the answer matched their country policy, whether the request actually went through, and whether they trusted the response enough not to raise a ticket anyway.
That is the central truth of modern HR architecture: employees do not experience platforms; they experience stitches.
That is why integration now sits on the critical path of employee UX. In large organisations, fragmentation is not an edge case. CIPD’s case study of “ChemicalCo” describes an organisation operating with nearly 200 payroll providers globally, alongside fragmented HCM, finance, identity and service processes; the point of the case is not that ChemicalCo was badly managed, but that this is what scale often looks like in reality. Even outside HR-specific examples, the wider enterprise software environment is now so dense that the average company runs 101 apps. Fragmentation is the default condition. Your job is not to eliminate it completely. Your job is to make it invisible to employees.
The AI shift makes this more urgent, not less. Microsoft’s 2026 Work Trend Index says employees’ ability to act is rising, but organisations are not redesigning work quickly enough to support that agency. Microsoft’s 2025 data also found a workforce running at the edge of fragmentation, with constant interruptions and overloaded attention. In that context, every avoidable seam in an HR journey is not just inconvenient; it consumes scarce cognitive bandwidth the organisation cannot afford to waste.
The old portal model is no longer enough. Microsoft now explicitly describes Copilot as the “UI for AI”, while Anthropic’s Model Context Protocol and Google’s Agent2Agent protocol are both attempts to normalise how AI systems connect to data, tools, and one another. The implication for HR is profound: the employee-facing layer is becoming conversational and agentic, while the back end remains stubbornly heterogeneous. That means integration design has moved out of the plumbing basement and into the experience strategy conversation.
The real measure of HR integration is not whether systems exchange data, but whether the employee can move from intent to outcome without losing identity, context, confidence, or momentum.
That is a higher bar than technical connectivity. It also produces better architecture decisions. Because once you define success that way, three things become obvious.
First, identity is part of UX. OpenID Connect gives you interoperable authentication. SCIM gives you a standardised way to provision and manage identities across systems. NIST’s 2025 Digital Identity Guidelines underline that authentication and federation are not side topics; they are core technical conditions for trusted access. And in an AI-enabled environment, that extends to non-human identities too. Okta’s 2026 reporting says 78% of organisations see controlling non-human identity permissions as a top concern, while only 10% have a strategy for governing them. If your AI assistant can read policies, open tickets, trigger workflows, or access employee data, identity architecture is now a security domain AND a design domain (OpenID).
Second, handoffs matter more than answers. An AI answer that cannot preserve context into a human case is a seductive failure mode. Google’s Dialogflow analytics foreground precisely the things mature teams need to watch: live-agent handoffs, escalation rates, no-match rates, and webhook failures. Microsoft’s agent analytics similarly centre conversation outcomes, generated answer quality, tool success, knowledge source errors, CSAT, and sentiment. These are not vanity dashboards. They point to a new operating truth: your employee experience is only as strong as the weakest handoff in the chain (Google).
Third, the front door is multiplying. That is not necessarily a problem. It only becomes a problem when every entry point starts its own journey from scratch. The smarter design principle is not “one door only”; it is “many doors, one memory”. That is how modern employees already work. They ask in Teams, search on mobile, click from Slack, complete in a portal, and follow up in email. Trying to force all of that back into a single old-fashioned HR homepage is not strategic discipline. It is nostalgia
The Four Continuities gives leaders a sharper way to assess integration quality than “how many APIs do we have?”
They are:
If one of those breaks, the seam becomes visible.
OpenID Connect and SCIM cover the basics of interoperable authentication and identity lifecycle. NIST’s AI RMF and playbook emphasise governed, measurable, trustworthy AI deployment. Microsoft’s recent guidance on ESS telemetry explicitly says leaders should move from raw usage to outcome-level signals, while product and service owners should treat retries, failure patterns, and regressions as signs of friction. That is effectively a continuity mindset, even if Microsoft does not use the phrase.
The old mental model said employees should enter through one sanctioned portal. The new model says employees will enter where work already is: Teams, the intranet, mobile, an email deep link, a manager workflow, a payroll reminder, a knowledge search, even an AI prompt. The job of architecture is to make those entry points behave like a constellation: many visible stars, one navigational system.
This is where AI genuinely changes the rulebook. Microsoft’s ESS pattern is designed as a unified interface for HR and IT tasks and has already been deployed internally to more than 300,000 employees and vendors. Microsoft’s Inside Track guidance repeatedly frames the ESS Agent as a “single pane of glass”, but its own adoption documentation is more nuanced and more useful: it emphasises telemetry, scenario-level success, reduction in assisted support, evaluation coverage, and stakeholder-specific interpretations of performance. That is a constellation logic disguised as a product story. The interface may appear singular to the employee, but the operating model behind it is distributed, instrumented, and scenario-led (Microsoft).
A subtle but important design consequence follows. You do not need one UX surface. You need one service grammar across many surfaces. That means the same intent model, the same policy sources, the same fulfilment logic, the same escalation rules, and the same identifiers, regardless of channel.
Whether that service core is built internally, through a platform such as Microsoft’s ESS stack, or with a specialist employee service layer such as Applaud, the principle should be identical: channels may differ, but the service contract should not.
Learn how to transform HR into a people-first function that builds trust, designs better experiences, and drives real business results in this interactive, 10-minute guide. Read Now.
Map the top 25 employee intents by volume and friction, not your application inventory. Use actual query logs, tickets, portal searches, failed searches, and repeat contacts. CIPD’s research on workplace technology found only 20% of workers say they are consulted on technology introduction, even though consultation is associated with much better outcomes for job quality and change. In integration terms, that means architects should stop guessing which journeys matter most. Employees will tell you—if you instrument and ask (CIPD).
For each priority journey, define the employee promise in plain language: what should be recognised, what data should be prefilled, what policy source is authoritative, what action should happen, when human help must step in, and what evidence should be logged. This is where NIST’s Govern-Map-Measure-Manage framing becomes practical rather than abstract (NIST).
Use federated authentication and lifecycle standards as table stakes, then modernise the last mile. FIDO’s 2025 enterprise survey found 87% of surveyed organisations had deployed or were deploying passkeys for workforce sign-ins, with reported positive impacts on user experience (82%), security (90%), help-centre call reduction (77%), and productivity (73%). For HR leaders, that matters because the ugliest “integration” failures are often authentication failures wearing a different badge (FIDO Alliance).
This is where many AI pilots quietly fail. Microsoft’s latest internal ESS work in Europe North found that when local content was missing, the agent sometimes defaulted to irrelevant country policies; the fix was not “more AI”, but better country tagging (Applaud uses AI to do this for you), tighter data scoping, and content curation. In other words, the front door is only as trustworthy as the back-room filing system. Or, as one IBM strategy mantra puts it: “Eliminate, simplify, automate.” Do not automate chaos.
Design escalation deliberately. Preserve transcript, intent, employee metadata, previous steps, cited source, and confidence signals. Human review should not be improvised. The ICO’s guidance for AI-supported decisions stresses fallback options, documented review procedures, logged overrides, and the ability to move to hybrid or manual handling when performance drops below acceptable thresholds. Apply the same discipline to employee service: the moment of handoff is where trust is either preserved or destroyed (ICO).
Don’t wait for a grand transformation review. Microsoft’s guidance recommends regular review of deflection, resolution, engagement, topic-level failures, channel differences, and feedback; its newer documentation goes further by tying telemetry to scenario-level success, retries, failbacks, latency and evaluation regressions. Google’s Dialogflow likewise surfaces handoffs, no-matches, escalation rates and latency. This is the operational rhythm HR leaders need: smaller but more disciplined improvement loops.
Take one journey—say, “start parental leave”. Walk it through six lenses: discover, authenticate, understand, decide, act, confirm.
Then ask these blunt questions:
Any “no” reveals a seam worth funding.
The point of this metric is not mathematical elegance. It is to force the right conversation. If an employee can ask “What parental leave am I entitled to?” in Copilot, the portal, email, or chat, and the answer changes depending on venue, the organisation does not have a channel problem. It has a service architecture problem.
A practical executive scorecard would look like this:
| Metric | What it reveals | Why it matters |
| Consistency score | Whether the same intent gets the same approved outcome across channels | Measures architectural coherence, not just experience polish |
| Channel-hop rate | How often employees must switch channel to finish the same need | Exposes friction and hidden cost |
| Handoff integrity | Whether context survives from AI to human without repetition | Protects trust in mixed human-AI service |
| Action success rate | % of agent-triggered actions completed correctly without repair or reopen | Separates useful automation from noisy automation |
| Exception escalation rate | Where low-confidence or high-risk intents surface | Helps tune autonomy and staffing |
| Knowledge freshness risk | Share of high-use content overdue for validation or showing conflicting outcomes | Prevents drift, not just search failure |
The new HR front door is not a chatbot. It is an orchestration layer with taste.
By “taste”, I mean the ability to know when to answer, when to ask, when to act, and when to hand over. That is why the industry’s move towards open protocols matters. MCP aims to give AI systems a standard way to connect to data and tools; A2A aims to let agents coordinate with other agents securely across platforms. Senior HR leaders do not need to master the technical contents of those specifications. But they do need to understand the strategic implication: the next generation of employee service will not be built from a single monolith. It will be composed from interoperable services and agents that must still feel coherent to the employee (Anthropic).
That makes context the new experience battleground. The assistant that knows your role, location, leave balance, manager relationship, open cases, device posture, and policy scope, while staying within governance guardrails, will feel integrated even when it spans six underlying systems. The one that gives a generic answer and punts you to a form will feel broken even if the architecture team calls it “connected”.
Every human-facing metric should have a machine-facing mirror. Otherwise you only know what the user felt, not why it happened.
| Employee-centred metric | Machine-centred mirror | Why both matter |
| Seam-breach rate: % of journeys where users must re-authenticate, re-enter data, switch channel, or restate context | Identity failure rate, token expiry errors, connector failures, missing context events | Measures the visible seam and the likely technical cause |
| First-time completion: % of journeys completed without restart or repeat contact | Tool success rate, webhook success, workflow completion | Separates completion from underlying fulfilment reliability |
| Handoff preservation: % of escalations where transcript, policy source, and prior steps are preserved | Session payload transfer success, case enrichment completeness | Protects trust at the most fragile point |
| Authoritative answer rate: % of answers grounded in approved current sources | Knowledge source coverage, citation success, answer quality, eval pass rate | Prevents fluent but unsafe answers |
| Time to meaningful progress: time from first intent to first successful step | Authentication latency, orchestration latency, dependency timeouts | Focuses on momentum, not just total resolution time |
| Cost per resolved need: full cost of resolution regardless of channel | Agent savings, assisted support reduction, infrastructure and licence cost per successful completion | Stops channel-shifting from masquerading as efficiency |
This table is not lifted from a vendor dashboard. It is a synthesis of what AI leaders and regulatory guidance now expose: outcomes, answer quality, tool success, knowledge source performance, sentiment, handoffs, time/cost savings, and logged human overrides. The point is to give HR leaders a board-level measurement language that still points engineering teams to the right fix.
Use three linked formulas.
An illustrative example: suppose your HR service handles 60,000 annual needs across leave, payroll, benefits and letters. If improved integration and AI assistance lift first-time self-service resolution from 42% to 57%, that is 9,000 additional needs resolved without assisted contact. At 12 minutes of avoided HR handling time and a loaded HR operations cost of £38 per hour, capacity released alone is worth about £68,400 a year. If those same journeys also save employees four minutes each, and average employee time is valued at £32 per hour, the employee-time value is about £19,200. Total gross annual value: £87,600, before wider benefits such as lower ticket queues, better policy compliance, and higher trust. Those wider benefits are real, but should be treated as upside rather than stuffed into an inflated ROI claim.
That disciplined approach matters because the evidence base is finally getting sharper. In a large field study of more than 5,000 customer-support agents, Brynjolfsson, Li and Raymond found AI assistance increased productivity by about 14–15% on average, with much larger gains for less experienced workers. That suggests the strongest value case for HR AI is flattening the experience gap and raising the floor of service quality (NBER).
The strongest case studies all tell the same story: the breakthrough is not “we added AI”, but “we unified intent, context, and fulfilment around the employee”.
IBM AskHR. IBM reports that AskHR handled 10.1 million interactions in 2024, automating more than 94% of simple HR inquiries and about 80% of repetitive HR transactions. IBM also says the move helped reduce HR operating costs by roughly 40% while supporting a global workforce with more consistent service. The useful lesson is not the scale alone. It is the sequence: standardise process, simplify work, then automate (IBM).
Microsoft Employee Self-Service Agent. Microsoft rolled the ESS Agent to more than 300,000 employees and vendors and describes it as a unified experience for HR and IT tasks. Its more valuable lesson, however, comes from the hard parts: country mismatches, local content gaps, and the need to ground the agent in vetted knowledge sources. Microsoft’s own ESS telemetry model focuses on scenario success, retries, assisted support reduction, latency, and evaluation regressions, not merely clicks or active users. That is exactly the measurement posture HR should copy (Microsoft).
ACCIONA. Google Cloud’s published case shows an employee virtual assistant serving 41,000 employees across 40 countries. Between late 2022 and late 2023 it logged more than 41,000 conversations, with a 90.33% completion rate and only 0.89% requiring HR intervention. That is an important data point because it demonstrates something many HR leaders underestimate: completion is possible at scale when you narrow the assistant to high-value, high-frequency, well-grounded service moments (Google).
CIPD’s ChemicalCo case. ChemicalCo’s relevance is architectural rather than AI-centric. CIPD uses it to show how large global employers end up with deeply fragmented payroll, HCM and service landscapes. The value of the example is that it legitimises the real-world starting point: you are not “behind” because you have a messy estate. You are normal. The question is whether you design around that reality or pretend it is temporary.
A final evidence point matters because it connects integration quality to trust and adoption. FIDO’s 2025 enterprise research found passkeys are being deployed by 87% of surveyed organisations, with strong reported gains in UX, security, productivity, and help-centre reduction. That is a reminder that the best employee experience improvements are often unfashionable. A smoother, phishing-resistant login may do more for perceived integration than a dazzling generative interface layered on top of brittle authentication (FIDO).
How Applaud Helps You Make It Happen
At Applaud, we believe employees are a company’s most important customers. That’s why our technology is built entirely from the employee’s point of view—delivering more human, intuitive, and rewarding HR experiences that empower HR teams to do more for their people.
If you’re ready to turn employee-first HR from vision to reality, we’re here to help. Get in touch to see how Applaud can transform your HR Service Delivery and create a workplace where employees truly thrive.
Duncan Casemore is Co-Founder and CTO of Applaud, an award-winning HR platform built entirely around employees. Formerly at Oracle and a global HR consultant, Duncan is known for championing more human, intuitive HR tech. Regularly featured in top publications, he collaborates with thought leaders like Josh Bersin, speaks at major events, and continues to help organizations create truly people-first workplaces.