Microsoft 365 Copilot knowledge work visar divergent adoption patterns i Fraunhofer-studie (550 employees, non-university research organization): Administrative staff rapporterar higher usefulness (M=0.94 vs 0.42, d=0.39, p=.050) och output quality (M=0.88 vs 0.26, d=0.48, p=.018) vid T01. Men scientific staff utvecklar more positive assessments över tid – perceived usefulness ökar signifikant T01→T02 (M=0.42→1.09, d=0.42, p=.022), perceived ease of use (M=0.74→1.31, d=0.45, p=.016). För dig som projektledare betyder detta: implementation strategy måste vara role-specific med context-sensitive training. Quick wins i admin work, longer learning curve i complex scientific tasks men eventually higher acceptance.
Två employee groups, två adoption trajectories
Administrative staff (T01 baseline högt):
- Perceived usefulness: M=0.94 (redan positivt)
- Output quality: M=0.88 (signifikant högre än scientific)
- Reliability: M=1.13
- Minimal change T01→T02 (stabilized early)
Scientific staff (T01 baseline lågt, strong growth T02):
- Perceived usefulness: M=0.42→1.09 (+160% increase!)
- Perceived ease of use: M=0.74→1.31 (+77% increase)
- Output quality: M=0.26→0.73 (+181% increase, though not sig p=.067)
Förklaring: Administrative tasks (text-based, clearly structured, repetitive) = immediate Copilot fit. Scientific tasks (experimental research, prototype development, complex problem-solving) = requires learning, routinization, workflow integration before benefits manifest.
För projektledare: Don’t judge AI success på immediate metrics. Scientific/complex knowledge work shows delayed but eventually strong acceptance via learning effects.
Task-dependent usefulness: Where Copilot shines vs struggles
High usefulness tasks (clearly structured, text-based):
- Document drafting
- Email composition
- Meeting summaries
- Standard administrative workflows
- Repetitive text generation
Lower usefulness tasks (context-heavy, experimental):
- Experimental research design
- Novel prototype development
- Complex data interpretation requiring domain expertise
- Tasks demanding deep contextual understanding
Implication: Copilot är task-specific tool, inte universal productivity multiplier. Greatest added value för well-defined, pattern-based work. Struggles med novel, context-dependent challenges.
För projektledare: Map team tasks by structure/repeatability. Deploy Copilot strategically där det faktiskt adds value. Avoid forcing adoption för unsuitable tasks (creates frustration, undermines acceptance).
Learning & routinization effects: Why time matters
Key finding: Scientific staff acceptans ökar över tid without changes i baseline tool capabilities. Detta är pure learning/routinization.
Mechanism:
- Month 1-2: Struggle med prompt engineering, unclear use cases
- Month 3-4: Develop effective prompting strategies
- Month 5-6: Integrate Copilot into established workflows
- Result: Same tool, dramatically higher perceived usefulness
Supporting evidence: 8 documented M365 Copilot updates Nov 2024–Apr 2025 (new integrations, improved prompt interpretation, system stability). Men scientific staff improvement går beyond these technical upgrades – det är skill development.
För projektledare: Budget för learning period. Measuring ROI after 2 weeks = misleading. Complex knowledge workers need 3-6 months för routinization before true productivity gains emerge.
Context-sensitive implementation: Five critical actions
1. Role-specific training programs: Administrative staff: Focus på immediate productivity (templates, shortcuts, common scenarios) Scientific staff: Focus på experimentation, prompt engineering, iterative refinement
2. Differentiated success metrics: Admin: Immediate time savings, task completion speed Science: Long-term workflow integration, gradual acceptance increase
3. Task mapping exercise: Before rollout: Audit team tasks by structure/repeatability Identify high-fit tasks (deploy aggressively) vs low-fit tasks (deploy cautiously eller not at all)
4. Phased adoption timeline: Admin areas: 1-2 month onboarding Scientific areas: 3-6 month learning period med ongoing support
5. Continuous feedback loops: Regular surveys capturing usefulness, ease of use, output quality, reliability Track temporal changes (T01, T02, T03…) för identify learning curves
Scale quality & methodological insights
Internal consistency excellent: Perceived usefulness scale: α=.97 (T01), α=.95 (T02) Item-total correlations: .85–.96
Sample limitation: 550 licensed users, ~20% response rate Valid population inference kräver ~380 cases (organization har 32,000 employees) → Results är exploratory, hypothesis-generating, not definitive
Design strength: Repeated cross-sectional (T01 Nov-Dec 2024, T02 Mar-Apr 2025) Captures temporal dynamics även utan full longitudinal tracking
För projektledare: Implement similar measurement approach. Baseline survey (T01), follow-up 3-6 months later (T02). Track group-specific trajectories.
Bottom line
Microsoft 365 Copilot knowledge work adoption följer role-dependent patterns. Administrative staff high immediate acceptance (M=0.94 usefulness, M=0.88 output quality) via task-fit med clearly structured work. Scientific staff lower initial (M=0.42 usefulness) men strong growth över tid (→M=1.09, d=0.42, p=.022) via learning/routinization. Task-dependent effectiveness: shines på text-based, structured tasks; struggles på experimental, context-heavy work. Implementation requires context-sensitive strategy, role-specific training, differentiated success timelines. Learning effects means 3-6 months before full benefits manifest i complex knowledge work. Measuring ROI after 2 weeks = misleading. Scale quality excellent (α=.95-.97), methodology exploratory men rigorous.
Källa: “Generative AI in Knowledge Work: Perception, Usefulness, and Acceptance of Microsoft 365 Copilot” av Carsten F. Schmidt et al., Fraunhofer Institute for Industrial Engineering IAO, publicerad 20 februari 2026.