Research → Workforce

Human in the Loop: Where AI Tools Actually Shine

Brandon Sneider · March 2026

The companies capturing real value from AI share a specific budget allocation: 70% on people and workflow redesign, 30% on technology. Most companies invert that ratio — and the data shows exactly what happens when they do.

ActivTrak tracked 163,638 employees across 1,111 companies over 443 million hours of work activity. After AI tool adoption, email time increased 104%. Chat time increased 145%. Multitasking rose 12%. Focus efficiency hit a three-year low. The average uninterrupted focus session fell to 13 minutes and 7 seconds. Not a single work category decreased.

ActivTrak Workforce Insights, 163,638 employees, 1,111 companies, 443M hours tracked

AI did not eliminate work. It generated more of it. Every AI-drafted document needs review. Every AI-generated email gets a reply. Every AI-produced analysis requires validation. The downstream coordination cost is real, and almost nobody budgets for it.

The Three Intensification Patterns

A UC Berkeley study followed approximately 200 employees over eight months and identified three distinct ways AI increases workload rather than reducing it.

Task expansion. When AI makes you faster at your job, you absorb tasks from adjacent departments. The marketing manager who can now produce content in half the time gets asked to produce content for sales, HR, and investor relations. The output per person rises. The workload per person rises faster.

Boundary erosion. AI tasks fill the gaps that used to be breaks. The 10-minute wait for a document review becomes a prompt session. The commute becomes AI-assisted email triage. The lunch hour becomes "just one more Copilot query." The work expands into every available moment because the tool is always available.

Cognitive overload. Running parallel AI-assisted workflows degrades decision quality. The brain that is supervising three AI outputs simultaneously makes worse judgments on all three than it would have made on one manually. The speed gain in generation is offset by the quality loss in oversight.

UC Berkeley/HBR, 8-month study, ~200 employees, February 2026

The Three-Tool Threshold

BCG surveyed 1,488 employees and found a sharp inflection point. Employees using four or more AI tools report 19% more information overload, 14% more mental effort, 12% greater fatigue, and 34% intent to quit — compared to 25% quit intent among those using fewer tools.

Three tools is the threshold where productivity gains hold. Beyond that, the coordination cost of managing multiple AI interfaces, remembering which tool does what, and context-switching between AI-assisted and manual workflows overwhelms the speed benefit.

The implication for tool procurement is direct: deploying Copilot AND Cursor AND Claude AND ChatGPT AND Gemini is worse than deploying two of them well. The organizations capturing value are the ones that chose deliberately and trained deeply, not the ones that gave everyone access to everything.

BCG AI at Work 2025, n=1,488

The Trust Collapse

Deloitte's TrustID study tracked approximately 60,000 employees and found corporate AI trust fell 38% in three months — from May to July 2025. Trust in agentic AI fell 89%. Overall AI usage declined 15% despite increased tool access.

This is not resistance to change. It is an informed response to experience. Employees tried AI tools, found them unreliable or burdensome, and pulled back. The trust decline happened fastest among employees with the most AI exposure — the early adopters who had the most data to judge by.

The fix is known. Hands-on training produces 144% higher trust than no training. Interactive practice produces 72% higher trust. Manager weekly check-ins increase trust 60%. Five hours of training is the inflection point: 79% of employees with five or more hours become regular AI users, versus 67% below that threshold.

Most companies spend 93% of their AI budget on technology and 7% on people.

Deloitte TrustID, ~60,000 employees, Q3 2025; BCG, n=10,600, June 2025; Deloitte Tech Trends 2026

The Competitive Window

The research converges on a specific timeline. Organizations have approximately 18 to 24 months before the cost of catching up rises structurally. The advantage compounds through three mechanisms: data flywheel effects (AI improves as it processes more organizational data), workforce learning curves (teams that have used AI for 18 months operate fundamentally differently from teams starting fresh), and workflow redesign advantages (processes rebuilt around AI capabilities cannot be replicated by bolting AI onto old processes).

This does not mean "rush to deploy everything." It means the organizations that deploy the right three tools, train their people deeply, and redesign their workflows now will have a structural advantage that late movers cannot close with budget alone. The window is about capability compounding, not first-mover hype.

What This Means for Your Organization

The counterintuitive finding — that AI creates more work, not less — reframes the entire AI deployment conversation. The question is not "how do we give everyone AI tools?" It is "which three workflows, redesigned around AI, would most directly reduce coordination overhead rather than increase it?"

The organizations in the 5% share a specific pattern: they deploy fewer tools (three or fewer), they invest 70% of their AI budget in training and workflow redesign (not technology), and they measure coordination cost as aggressively as they measure speed. The companies deploying five tools with no training are the ones generating the 104% email increase. The companies deploying two tools with deep training and workflow redesign are the ones generating the ROI that UPS, JPMorgan, and Citi demonstrate.

If the data in this article raised questions about how your organization is deploying AI — whether the tools are generating speed or generating coordination overhead — that is a conversation worth having. brandon@brandonsneider.com.

Sources

I publish research on AI strategy for executives. Data, not hype.