Knowledge Workers Got Hours Back From Email With Copilot
Robauto Team · Published · Last updated
A large Microsoft randomized study found Copilot users spent about two fewer hours a week on email and opened more focus time—good news for small teams adopting AI assistants.
Knowledge work still runs on inboxes. Before AI assistants showed up inside everyday apps, the average worker in a large Microsoft randomized study spent about 11.7 hours a week in email sessions—more than a quarter of a typical workweek. The hopeful part is what happened next: when people actually used an integrated AI assistant, they got measurable hours back without dropping the ball on replies.
Microsoft Research and coauthors studied Microsoft 365 Copilot in a six-month field experiment across 66 firms and 7,137 knowledge workers. Half received access to Copilot inside the tools they already used for email, documents, and meetings. The rest continued as usual. The result is some of the clearest real-world evidence yet that generative AI can reclaim time for ordinary office work—not just lab demos.
What the randomized study actually found
Focusing on the later months of the experiment (once novelty wore off), workers with Copilot access spent about 1.4 fewer hours per week in Outlook email sessions—roughly a 12% cut from the pre-study average. Workers who used the tool showed a larger local effect of about two fewer hours of email time each week (about 17%). They also consolidated email into fewer sessions, opening roughly 1.5 to 2.2 hours of longer email-free blocks during the workday—the kind of stretch where deeper projects can finally move.
Importantly, people were not ghosting their inboxes. Thread reply counts stayed about the same. Some of the saved time even showed up as a small reduction in after-hours digital work. In other words: same communication load, less time burned getting through it.
A companion Microsoft Research summary of early Copilot impacts also reported that workers who used the tool completed documents about 12% faster, with nearly 40% of workers given access using it regularly across the six-month window. Over 90% of treated workers tried Copilot at least once.
| Finding | Figure | Source |
|---|---|---|
| Study scale | 7,137 workers at 66 firms, 6-month RCT | arXiv:2504.11436 |
| Email time (intent-to-treat) | ~1.4 hrs less / week (~12% vs ~11.7 hr baseline) | arXiv:2504.11436 |
| Email time (regular users / LATE) | ~2 hrs less / week (~17%) | arXiv:2504.11436 |
| Email-free focus blocks | ~1.5 hrs (ITT) to ~2.2 hrs (users) per week | arXiv:2504.11436 |
| Documents (companion summary) | ~12% faster completion among users | Microsoft Research |
| Adoption | ~40% regular weekly use; >90% tried at least once | Microsoft Research / arXiv |
Why this matters for small and midsize teams
Large firms ran the experiment, but the pattern is especially relevant for smaller businesses where the same person wears product, support, and ops hats. When email eats a quarter of the week, every reclaimed hour is a chance to talk to customers, ship a proposal, or train a teammate.
Separately, a Forrester Total Economic Impact study commissioned by Microsoft modeled projected three-year ROI for Microsoft 365 Copilot at small and medium businesses ranging from 132% to 353%, with scenarios that also pointed to faster onboarding and lower operating costs. Those figures are projected ROI from a commissioned TEI, not results from the randomized trial above—so treat them as forward-looking modeling, not causal RCT evidence. Still, they show why SMB leaders are paying attention: the same assistant that trims inbox time is framed as a growth investment, not just a gadget.
What changed—and what did not
The study’s nuance is useful. Copilot clearly helped with solo-adjustable work like inbox triage. Meeting time and meeting counts barely moved on average—group norms are harder to rewrite when only a few people on a team have the new tool. That is a feature of early adoption, not a failure: individuals can save hours today while organizations learn how to redesign workflows together tomorrow.
{
"study": "Microsoft 365 Copilot field RCT",
"sample": {"workers": 7137, "firms": 66, "duration_months": 6},
"email_hours_saved_per_week": {"intent_to_treat": 1.4, "regular_users_approx": 2},
"focus_blocks_gained_hours_per_week": {"intent_to_treat": 1.5, "regular_users_approx": 2.2},
"documents_faster_pct_companion": 12,
"regular_use_share_approx": 0.40,
"takeaway": "AI assistants reclaim solo task time; agent-ready brands win when workers research and act"
}The Robauto takeaway: when workers use agents, findable brands win
Hours back from email are not only an HR win. They change how people research vendors, compare products, and decide who to trust. Assistants and agents increasingly summarize pages, policies, and offers on a worker’s behalf. Brands that are clear, structured, and usable by those agents get discovered; brands that are opaque get skipped.
Across the Robauto network, AI-related traffic already accounts for about 29.7% of page views over the last 90 days, and the live Agent Readiness Index sits at a median score of 20. That is opportunity framing, not a scoreboard dunk: many sites still have room to become easier for agents to understand—refund rules, product facts, and contact paths that an assistant can actually use when a busy human asks.
If your team is reclaiming time with Copilot-style tools, the next constructive step is making sure your own digital front door is ready for the same assistants. Run a free Agent Scan at https://robauto.ai/scan and see how findable you are when knowledge workers—and the agents helping them—go looking.
Sources: Microsoft Research / arXiv field experiment on M365 Copilot work patterns; Microsoft Research “Early Impacts of M365 Copilot” publication page; Microsoft blog on Forrester TEI for SMB Copilot ROI (commissioned, projected); Robauto network AI-search summary.
Sources
- Shifting Work Patterns with Generative AI (arXiv:2504.11436)
- arXiv abstract: Shifting Work Patterns with Generative AI
- Early Impacts of M365 Copilot — Microsoft Research
- Microsoft 365 Copilot SMB ROI — Forrester TEI (commissioned)
- Robauto AI search / network data
Where can an agent get this data live?
Robauto MCP — Agent Intelligence & Brand Discovery Infrastructure. A public, stateless Streamable HTTP endpoint at https://mcp.robauto.ai/mcp, callable with no key. Setup and the full tool list are in the Robauto MCP directory.