The AI That Knows You Better Than Your Boss Does

Microsoft’s Work IQ promises to make Copilot feel “personal.” Marketers should be asking: personal for whom?

Two weeks ago, Microsoft unveiled Work IQ at Ignite 2025. The promotional video—58 seconds of sleek animation and warm voiceover—promises an AI that “knows you, your work, and your company.” It builds “memory” of your “style, preferences and relationships.” It can “predict the next best action.”¹

The language is deliberate. Work IQ does not monitor; it learns. It does not track patterns; it understands. The framing positions surveillance architecture as personal empowerment.

For marketers evaluating AI tools—or building campaigns around them—this distinction matters enormously. Because the gap between what Work IQ promises and what it architecturally requires reveals something important about the AI productivity narrative we’ve been sold.

The Numbers Behind the Hype

Start with adoption. Microsoft claims more than 90% of Fortune 500 companies use Microsoft 365 Copilot.¹ That figure demands parsing.

According to sources with access to internal Microsoft sales data, as of August 2025, Microsoft had approximately eight million active licensed users of Microsoft 365 Copilot—a 1.81% conversion rate across its 440 million Microsoft 365 subscribers.² 

A CNBC survey of technology leaders found that when asked whether Copilot justified its $30 monthly cost, equal numbers said yes and no. The largest group—fully 50%—answered that it was still too soon to know.³

A separate Gartner survey of 215 IT leaders revealed a starker paradox: 94% reported measurable benefits from Copilot, yet only 6% had completed global rollouts. Seventy-two percent remained locked in pilot programmes.⁴

For a product Microsoft has positioned as transformational—the centrepiece of every major announcement for two years—these figures suggest something other than runaway success.

This pattern of inflated claims meeting tepid reality isn’t unique to Microsoft. The gap between vendor theatre and measurable outcomes has become a defining feature of enterprise AI—a theme I’ve explored in my analysis of generative AI marketing more broadly.⁵

The ROI Problem Nobody Can Solve

The measurement challenge is fundamental. According to Deloitte’s State of Generative AI research, more than 40% of companies struggle to define and measure the impact of generative AI initiatives. Fewer than half have developed KPIs to effectively measure AI returns.³

The UK government’s cross-departmental Copilot experiment found participants saved an average of 26 minutes daily. But the study also noted that Copilot’s output required editing in most cases, reducing claimed efficiency gains. Benefits varied significantly by job function.³

The question finance teams cannot answer: if Copilot saves employees 26 minutes daily, what does that translate to in business value?

Time saved doesn’t automatically equal cost reduction unless it results in fewer staff hours. Time saved doesn’t generate revenue unless employees use freed time for higher-value work.

Microsoft’s own Copilot Dashboard calculates “Copilot-assisted value” by multiplying assisted hours by an average hourly rate (defaulting to $72). This assumes all time saved has equal economic value—an assumption that doesn’t reflect how organisations actually operate.³

More troubling: the vendor controls both the product and the metrics used to justify purchasing more of it. No independent audits of Microsoft 365 Copilot productivity claims exist.

I’ve written previously about the productivity delusion in AI tool adoption—the uncomfortable truth that adding tools rarely adds clarity, and may subtract from the cognitive work that actually matters.⁶

Adding tools rarely adds clarity, and may subtract from the cognitive work that actually matters.

What Work IQ Actually Builds

Set aside the productivity question. Consider instead what Work IQ architecturally requires.

According to Microsoft’s announcements and technical documentation, Work IQ comprises three layers:¹⁷

Data: Every action you take in Microsoft 365 generates signals—emails sent, files updated, chats, meetings. Work IQ gathers these signals to understand “the flow of your work.”

Memory: As you work, Work IQ notices patterns—topics you deal with often, your writing tone, tools you prefer, people you collaborate with. These patterns become a “shared work memory” between you and Copilot.

Inference: Work IQ matches current activity against established patterns to “predict the next best action”—suggesting insights, selecting appropriate AI agents, surfacing relevant files before you ask.

Microsoft explicitly states that Work IQ understands not just your “org chart” but your “work chart”—the informal networks of influence and collaboration that define how organisations function.¹

This is not incidental data collection. It is comprehensive behavioural profiling—your communication patterns, professional relationships, work habits, decision-making tendencies—packaged as personalisation.

The Overpermissioning Problem

One of the most documented risks with Copilot deployment is overpermissioning. A 2025 analysis found that over 15% of business-critical files were at risk from oversharing—erroneous access permissions that Copilot inherits and exposes. Eighty-three percent of at-risk files were overshared with users or groups within companies.⁸

The US House of Representatives banned congressional staff from using Copilot over concerns about data leakage to unauthorised cloud services. Security researchers discovered a vulnerability in Copilot Studio enabling server-side request forgery that could leak sensitive information about internal cloud services.⁸

Microsoft’s position is that Copilot respects existing file permissions. But the more fundamental issue remains: systems designed to aggregate and cross-reference organisational data surface information employees can technically access, regardless of whether they should access it in practice.

The Dutch organisation Surf, working with the Privacy Company, issued a Data Privacy Impact Analysis recommending educational institutions cease using Microsoft 365 Copilot altogether, citing concerns around diagnostic data, telemetry, responsible AI practices, and “loss of control through lack of transparency.”⁹

AI Memory Meets EU Regulation

Work IQ’s memory architecture arrives at a fraught regulatory moment. On 2 February 2025, Article 5 of the EU AI Act came into effect, establishing strict prohibitions on AI systems deemed unacceptable due to potential harm.¹⁰

Prohibited practices include using AI to infer emotions of individuals in workplaces, profiling employees based on social media activity to assess job performance, and categorising individuals based on biometric data to deduce sensitive characteristics.

Researchers at Vrije Universiteit Amsterdam warn that the regulatory approach may leave workers unprotected: “The question is whether this approach provides sufficient safeguards for employees. Will supplier guidelines truly tackle the real risks and safety concerns, or will there be attempts to avoid liability?”¹¹

The AI Now Institute has called for policies mandating disclosures and prohibitions on workplace data collection, explicitly noting that consent frameworks are meaningless in employment contexts: “knowledge of a system is not tantamount to acceptance of its use given the risk that refusal to use a work-mandated system could lead to retaliation.”¹²

Knowledge of a system is not tantamount to acceptance of its use given the risk that refusal to use a work-mandated system could lead to retaliation

Work IQ does not currently conduct emotion inference or biometric categorisation. But its memory architecture—building profiles of work patterns, communication styles, and professional relationships—sits adjacent to precisely the capabilities regulators have identified as concerning.

The Productivity Paradox Nobody Discusses

The broader evidence on AI productivity tools challenges the narrative Microsoft’s marketing assumes.

Three major studies released in autumn 2025 reveal what researchers call a productivity paradox: while 85-95% of developers now use AI coding tools, actual productivity gains remain stuck at 10-15%.

A rigorous academic study found experienced developers were 19% slower with AI tools—despite forecasting they would be 24% faster.¹³

The perception distortion persisted even after measurement: developers who performed worse objectively still estimated a 20% improvement.

Bain & Company’s Technology Report 2025 explains why vendor claims of 30-55% productivity gains don’t materialise: code writing represents only 25-35% of the development lifecycle. When AI accelerates one phase but others remain constant, net productivity gains are 9-10% across total developer time.¹³

A 2025 McKinsey survey found that while AI adoption is broadening, only 1% of C-suite respondents describe their generative AI rollouts as mature—”fundamentally changing how work is done and driving substantial business outcomes.”¹⁴

Meanwhile, Wharton’s 2025 AI Adoption Report found that as usage climbs, 43% of leaders see risk of declines in skill proficiency.¹⁵

Cognitive Offloading: The Cost Nobody Measures

Research increasingly suggests that over-reliance on AI assistants may diminish critical thinking skills. Studies from IE University note that AI’s predictive capabilities in professional settings can lead to “weaker analytical abilities” and failure to critically assess algorithmic recommendations.¹⁶

Academic research published in 2025 found that AI’s learning and iteration speed “far surpasses that of an average employee,” potentially generating feelings of inadequacy among workers.¹⁷ The psychological implications of working alongside systems designed to outperform human cognition remain poorly understood.

If Copilot consistently predicts the “next best action,” what happens to the employee’s capacity to determine that action independently? Microsoft’s marketing doesn’t address this. Neither do most productivity metrics.

I’ve explored this tension elsewhere—in my analysis of Anthropic’s Claude marketing, which notably positions AI as a “thinking partner” rather than a replacement.¹⁸ The framing matters. Microsoft’s language of prediction and automation implies a fundamentally different relationship between human and machine than partnership or collaboration.

The Marketing Question

Microsoft’s 58-second Work IQ video makes no mention of employee consent mechanisms for behavioural profiling, data retention periods for memory constructs, individual access rights to AI-generated assessments, algorithmic transparency around inference processes, or opt-out provisions.

This is not unusual for enterprise marketing. What makes it consequential is the scale of Microsoft’s ambition and the intimacy of the data involved.

For marketers, the question is whether to perpetuate the framing or interrogate it. The productivity narrative—AI as friction-free empowerment—has commercial utility. It also elides genuine concerns that enterprise buyers, employees, and regulators are increasingly asking about.

Compare this approach to the marketing strategies I’ve analysed elsewhere. The Swiggy Wiggy 3.0 campaign succeeded precisely because it centred the humans behind the service—delivery partners became protagonists, not optimisation targets.¹⁹ Microsoft’s Work IQ does the opposite: employees become data sources in an intelligence layer that serves organisational efficiency.

The Economist predicts that 2026 will be the year AI’s true impact becomes apparent—”boom, bust or backlash.”²⁰ Marketing that treats AI productivity claims as self-evident rather than contested may not age well.

What This Means for Practitioners

If you’re evaluating AI tools for your organisation, several things follow from the evidence:

Budget for reality, not vendor claims. Expect 10-15% productivity gains from AI tools alone, not 30-55%. Achieving 25-30% gains requires lifecycle-wide transformation costing 3-5 times the tool licences.¹³

Trust measurement, not perception. Employees consistently overestimate AI productivity gains by 20-40 percentage points. Self-reported benefits are not reliable evidence.

Address permissions before deployment. Organisations with immature data governance face significant exposure risks from AI tools that aggregate and surface information across organisational boundaries.⁸

Ask what memory means. If an AI system builds persistent profiles of employee behaviour, understand where that data lives, who can access it, and how long it persists. The personalisation value proposition has a surveillance architecture prerequisite.

Prepare for regulatory attention. The EU AI Act’s workplace provisions are already in effect.¹⁰ UK and US legislative proposals addressing algorithmic management are advancing.¹² Memory-enabled AI systems will face increasing scrutiny.

These aren’t abstract concerns. Vanity metrics are the peacocks of the digital marketing menagerie: beautiful, useless, and everywhere at once.²¹ The same applies to AI productivity dashboards that measure activity rather than outcomes.

The Honest Question

Work IQ represents genuine technical innovation. The integration of organisational data, personalised memory, and predictive inference into productivity applications is architecturally significant. For organisations with mature data governance, clear consent frameworks, and robust access controls, it may deliver meaningful efficiency gains.

The concerns are equally genuine.

An AI that “knows” your communication patterns, professional relationships, and behavioural preferences is not merely a productivity tool. It is a comprehensive behavioural profile—one that exists whether or not its existence serves your interests.

Microsoft’s marketing assumes the answer to “should AI systems build persistent memory of how individual employees work?” is obviously yes. The evidence suggests the question deserves rather more consideration than a 58-second promotional video allows.

For marketers building campaigns around AI productivity tools, there’s a choice.

Perpetuate the framing that treats surveillance architecture as seamless empowerment. Or acknowledge the complexity—and perhaps earn the trust of audiences increasingly sceptical of AI promises that don’t survive contact with evidence.

The productivity claims will be tested. The memory architectures will be regulated. The question is whether marketing narratives prepared audiences for that reckoning or contributed to it.

---

Related Reading

• Tool Sprawl: The Productivity Delusion
• Selling AI Without Showing Product: Anthropic’s Marketing Restraint
• Claude Sonnet 4.5 Marketing Analysis
• ChatGPT Pulse: A Multi-Lens Critique
• Brand Anthem in the Age of Algorithms: Swiggy Wiggy 3.0

---

Footnotes

Footnotes

1. Microsoft, “Microsoft Ignite 2025: Copilot and agents built to power the frontier firm,” Microsoft 365 Blog, November 2025. https://www.microsoft.com/en-us/microsoft-365/blog/2025/11/18/microsoft-ignite-2025-copilot-and-agents-built-to-power-the-frontier-firm/ ↩ ↩² ↩³ ↩⁴
2. Perspectives+, “Microsoft 365 Copilot’s commercial failure,” October 2025. https://www.perspectives.plus/p/microsoft-365-copilot-commercial-failure ↩
3. SAMexpert, “Enterprises Are Still Deciding if Microsoft 365 Copilot Is Worth It,” October 2025. https://samexpert.com/microsoft-365-copilot-roi/ ↩ ↩² ↩³ ↩⁴
4. Louis Columbus, “Microsoft’s Copilot Paradox: 94% Report Benefits, 6% Deploy,” LinkedIn, June 2025. https://www.linkedin.com/pulse/microsofts-copilot-paradox-94-report-benefits-6-deploy-louis-columbus-fv4gc ↩
5. Suchetana Bauri, “Generative AI Marketing Archives,” suchetanabauri.com. https://suchetanabauri.com/tag/generative-ai-marketing/ ↩
6. Suchetana Bauri, “Tool Sprawl: The Productivity Delusion,” suchetanabauri.com. https://suchetanabauri.com/tool-sprawl-productivity-delusion-perplexity-ai/ ↩
7. AdminDroid, “What is Work IQ in Microsoft 365,” November 2025. https://blog.admindroid.com/work-iq-in-microsoft-365/ ↩
8. Concentric AI, “2025 Microsoft Copilot Security Concerns Explained,” September 2025. https://concentric.ai/too-much-access-microsoft-copilot-data-risks-explained/ ↩ ↩² ↩³
9. Albert Hoitingh, “Tackling Microsoft 365 Copilot data security and governance concerns,” January 2025. https://alberthoitingh.com/2025/01/20/tackling-microsoft-365-copilot-data-security-and-governance-concerns/ ↩
10. Bird & Bird, “AI & the Workplace: Navigating Prohibited AI Practices in the EU,” March 2025. https://www.twobirds.com/en/insights/2025/global/ai-and-the-workplace-navigating-prohibited-ai-practices-in-the-eu ↩ ↩²
11. Vrije Universiteit Amsterdam, “AI in the workplace: who protects the worker?,” April 2025. https://vu.nl/en/news/2025/ai-in-the-workplace-who-protects-the-worker ↩
12. AI Now Institute, “Algorithmic Management: Restraining Workplace Surveillance,” April 2025. https://ainowinstitute.org/publications/algorithmic-management ↩ ↩²
13. Byteiota, “AI Productivity Paradox: Why Gains Stay at 10-15%,” November 2025. https://byteiota.com/ai-productivity-paradox-why-gains-stay-at-10-15/ ↩ ↩² ↩³
14. McKinsey & Company, “The State of AI: Global Survey 2025,” November 2025. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai ↩
15. Wharton School, “2025 AI Adoption Report: Gen AI Fast-Tracks Into the Enterprise,” October 2025. https://knowledge.wharton.upenn.edu/special-report/2025-ai-adoption-report/ ↩
16. IE University, “AI’s cognitive implications: the decline of our thinking skills?,” February 2025. https://www.ie.edu/center-for-health-and-well-being/blog/ais-cognitive-implications-the-decline-of-our-thinking-skills/ ↩
17. Zhang et al., “Reflection or Dependence: How AI Awareness Affects Employee Behaviour,” PMC, January 2025. https://pmc.ncbi.nlm.nih.gov/articles/PMC11852158/ ↩
18. Suchetana Bauri, “Selling AI Without Showing Product: Anthropic’s Marketing Restraint,” suchetanabauri.com. https://suchetanabauri.com/claude-thinking-partner-ai-marketing-restraint/ ↩
19. Suchetana Bauri, “Brand Anthem in the Age of Algorithms: Swiggy Wiggy 3.0,” suchetanabauri.com. https://suchetanabauri.com/swiggy-wiggy-3-0-campaign-employee-advocacy/ ↩
20. The Economist, “AI’s true impact will become apparent in the coming year,” The World Ahead 2025, November 2025. https://www.economist.com/the-world-ahead/2025/11/10/ais-true-impact-will-become-apparent-in-the-coming-year ↩
21. Suchetana Bauri, “September Smartphone Marketing 2025 – Hype or Real Value?,” suchetanabauri.com, September 2025. https://suchetanabauri.com/the-september-siege-when-smartphone-brands-lost-their-collective-sanity-in-the-marketing-melee/ ↩