The Subtraction Discipline: Why the Highest-ROI AI Decision Is What You Stop Doing

Brandon Sneider | March 2026

Executive Summary

AI is adding work, not replacing it. ActivTrak’s analysis of 443 million work hours (n=163,638 employees, 1,111 companies, March 2026) finds that after AI adoption, time spent across every measured activity category increased — email up 104%, messaging up 145%, business management up 94%. Not a single category decreased. The organization got faster at producing more of everything, including things that should not exist.
The human brain defaults to addition. Adams et al. in Nature (n=1,585 across eight experiments, April 2021) demonstrate that people systematically overlook subtractive solutions. When asked to improve a situation, subjects default to adding features, steps, and rules — even when removing them is objectively superior. This cognitive bias explains why every AI playbook tells companies what to adopt and none ask what to eliminate.
BCG identifies the cost of additive overload. Their March 2026 study (n=1,488 U.S. workers) finds that beyond three AI tools, productivity collapses — major errors rise 39%, decision fatigue increases 33%, and 34% of affected workers intend to quit. Gartner projects that 20% of organizations will use AI to flatten their structures by end of 2026, eliminating more than half of middle management positions. The organizations moving fastest are subtracting layers, not adding tools.
The 5% that capture AI value practice subtraction before addition. McKinsey’s 2025 State of AI survey finds that workflow redesign — not tool deployment — is the single strongest predictor of enterprise AI impact. Shopify demonstrated the principle at scale: canceling all recurring meetings with more than two people freed 76,500 hours and produced a 25% increase in project completion. The tool was a calendar script. The discipline was deciding what to stop doing.

The Addition Instinct and Why It Fails

Every company deploying AI faces the same question: what should change? The universal answer has been “add.” Add tools. Add training. Add governance. Add dashboards. Add AI coordinators, steering committees, acceptable use policies, and quarterly reviews.

The result, measured at industrial scale, is that organizations are doing everything they did before AI — and more of it.

ActivTrak’s 2026 State of the Workplace report is the largest behavioral dataset on this phenomenon. Researchers analyzed 10,584 employees’ digital activity patterns 180 days before and 180 days after AI tool adoption. The headline finding is stark: not one measured work category showed a decrease. Email time doubled. Chat increased 145%. Business management tools saw 94% more usage. Time across every job responsibility rose between 27% and 346%.

The instinct to add is not a management failure. It is a documented cognitive bias. Gabrielle Adams, Benjamin Converse, Andrew Hales, and Leidy Klotz published their landmark research in Nature in April 2021. Across eight experiments with 1,585 participants, they found that people systematically default to additive solutions — adding features, components, and steps — and overlook subtractive alternatives that are equally or more effective. The bias persists even when subtraction is cheaper and faster. Participants needed explicit cueing to consider removing something; left unprompted, they added.

This explains a structural failure in enterprise AI strategy. The consulting playbooks, vendor roadmaps, and governance frameworks all start with what to adopt. None start with what to eliminate.

The Accumulation Tax

The cost of unchecked addition is now measurable.

Cognitive overload. BCG’s March 2026 study (n=1,488 U.S. workers, published in Harvard Business Review) identifies the threshold: productivity rises from one to two to three AI tools, then collapses at four. Workers experiencing “AI brain fry” report 14% more mental effort, 12% greater fatigue, 19% higher information overload, 33% more decision fatigue, and 39% more major errors. At a $5 billion-revenue organization, suboptimal decision-making costs an estimated $150 million annually (Gartner, 2018 baseline adjusted for inflation).

Focus destruction. ActivTrak finds focus efficiency has dropped to 60%, a three-year low. The average uninterrupted focus session lasts 13 minutes and 7 seconds — down 9% since 2023. Microsoft’s Work Trend Index (2025) documents employees being interrupted every two minutes during core work hours — 275 times per day. Sixty-eight percent say they lack sufficient uninterrupted focus time.

Work expansion, not compression. The workday itself shrank by only 2%, but productive hours increased 5% and weekend work surged — Saturday hours up 46%, Sunday hours up 58%. Saturday start times shifted to 7:11 a.m. from 8:35 a.m. AI made it possible to be productive earlier, later, and on weekends. No one decided whether the organization wanted that.

Collaboration bloat. Collaboration time increased 34%. Multitasking rose 12%. These are not signs of a more connected workforce. They are signs that AI made it easier to generate output that demands coordination. Every AI-drafted document needs review. Every AI-generated analysis needs validation. Every AI-composed email needs a reply. The downstream coordination cost of AI-generated output is the accumulation tax no one budgeted for.

Asana’s Anatomy of Work research quantifies the baseline: knowledge workers spend 58% of their day on “work about work” — communicating about work, searching for information, switching between apps, chasing status updates. AI has not reduced that 58%. The evidence suggests it has grown.

What the Subtraction Leaders Actually Cut

The organizations capturing disproportionate value from AI share a practice that is invisible in most implementation playbooks: they decide what to stop doing before deciding what to start.

Meetings

Shopify’s 2023 calendar purge remains the most documented case. COO Kaz Nejatian declared “meetings are a bug” and IT ran a script deleting every recurring meeting with more than two people. Results: 76,500 hours freed instantly, meeting time down 33% year-over-year, and a projected 25% increase in project completions across engineering, product, and UX teams. Critically, most deleted meetings never came back. The meetings existed because nobody had stopped to ask whether they should.

Shopify built a meeting cost calculator into its internal calendar to prevent reaccumulation. The tool computes the loaded cost of every attendee’s time for every meeting. Visibility creates friction. Friction creates discipline.

Microsoft’s data quantifies the opportunity: the average employee spends 57% of their time communicating (meetings, email, chat) and 43% creating (documents, spreadsheets, presentations). Sixty percent of meetings are ad hoc. Half cluster into the 9-11 a.m. and 1-3 p.m. windows — precisely when circadian rhythms favor deep work.

Status Reports and Update Rituals

Deloitte’s 2026 State of AI in the Enterprise report (n=2,770+ executives across 16 countries) segments organizations into three tiers: 34% are fundamentally transforming processes, 30% are redesigning key workflows, and 37% are layering AI onto legacy processes with no structural change. The gap between the top third and the bottom third is not which tools they bought. It is which processes they eliminated.

A financial services company profiled in the report built agentic workflows to automatically capture meeting actions from video conferences, draft follow-up communications, and track execution. The subtraction was not the agent — it was the weekly status meeting the agent replaced. The meeting had existed for eleven years. Nobody questioned it until someone asked what the meeting produced that a system could not.

Approval Chains

Oro Labs, whose clients include Coca-Cola, Pfizer, and Novartis, found that one Fortune 500 energy company with $40 billion in annual revenue had 20 million human touchpoints per year in its procurement process. The AI intervention was not adding a tool to the process. It was compressing a weeks-long approval chain into hours by eliminating steps that existed because of information asymmetry that AI resolved.

Gartner projects that by end of 2026, 20% of organizations will use AI to flatten their structure, eliminating more than half of current middle management positions. The positions being eliminated are disproportionately coordination roles — people whose primary function is aggregating status from below and reporting it above. When AI can aggregate and report, the coordination layer becomes the process to subtract.

The Reports Nobody Reads

McKinsey’s 2025 State of AI report identifies workflow redesign as the single strongest predictor of enterprise AI impact, ahead of tool selection, training investment, or leadership commitment. Their finding: high-performing organizations are 3.6 times more likely to pursue transformational change, and 55% fundamentally rework workflows when deploying AI.

The key word is “rework” — not “augment.” Augmenting a report nobody reads with AI produces a faster report nobody reads. Reworking means asking whether the report should exist. Deloitte’s segmentation data suggests that two-thirds of organizations have not asked that question.

The Subtraction Audit: Five Questions for Monday Morning

The subtractive discipline does not require a consulting engagement or a transformation program. It requires asking five questions about any process AI is being deployed to accelerate:

#	Question	Diagnostic Signal
1	If this process disappeared tomorrow, who would notice — and when?	If the answer is “nobody for two weeks,” the process is a candidate for elimination, not acceleration.
2	Does this process exist because of an information gap that AI now closes?	Status meetings, approval chains, and exception reports often exist because someone lacked visibility. AI dashboards may make the meeting unnecessary.
3	What downstream work does this process generate?	A weekly report that triggers three review meetings and eleven email threads has a coordination cost 5x the report itself. Eliminate the report and the downstream cost vanishes.
4	Is this a process or a habit?	Deloitte finds 37% of organizations are layering AI onto legacy processes with no structural change. Many of those processes are organizational habits — they persist because they have always existed, not because they produce value.
5	What would happen if you cut this process and reallocated the time to deep work?	ActivTrak’s data shows focus sessions have fallen to 13 minutes. Every eliminated meeting or report is an opportunity to reclaim blocks of uninterrupted thinking — the work AI cannot replace.

The audit is deliberately simple. The difficulty is not analytical. It is political. Every process has an owner, a constituency, and an institutional memory. Subtraction challenges all three.

The Subtraction-Before-Addition Sequence

Organizations that capture AI value follow a consistent sequence that reverses the default instinct:

Step 1: Inventory the accumulation. Before deploying any AI tool, catalog the processes, meetings, reports, and approval chains that touch the target workflow. Asana’s data says 58% of the workday is coordination. Start there.

Step 2: Apply the five-question audit. For each process in the inventory, run the diagnostic. Flag everything where the honest answer to question one is “nobody would notice.”

Step 3: Subtract before you add. Eliminate or consolidate flagged processes before deploying AI into the workflow. Shopify’s experience demonstrates the principle: the productivity gain from subtraction alone (76,500 hours, 25% more projects) rivaled what most organizations expect from AI deployment.

Step 4: Deploy AI into the simplified workflow. McKinsey’s data is unambiguous: redesigned workflows produce enterprise-level impact. Unredesigned workflows produce single-digit improvements or negative returns when implementation costs are included.

Step 5: Measure what didn’t come back. Shopify’s meeting cost calculator serves a second purpose — it prevents reaccumulation. Without active prevention, organizations will rebuild the processes they subtracted. Gartner’s 64% of CFOs planning SG&A below revenue growth rate in 2026 suggests the executive appetite for subtraction is real, but appetite without a measurement system becomes a one-time exercise.

Key Data Points

Finding	Source	Detail
No work category decreased after AI adoption	ActivTrak, March 2026	n=10,584 users, 180 days pre/post, email +104%, chat +145%, business management +94%
Humans default to additive solutions	Adams et al., Nature, April 2021	n=1,585 across 8 experiments; subtractive solutions systematically overlooked
Productivity collapses at 4+ AI tools	BCG/HBR, March 2026	n=1,488 U.S. workers; major errors +39%, decision fatigue +33%, quit intent 34%
Focus session now 13 min 7 sec	ActivTrak, March 2026	n=163,638 employees; down 9% since 2023; focus efficiency at 3-year low of 60%
58% of workday is “work about work”	Asana Anatomy of Work	Knowledge workers; communicating, searching, switching apps, chasing status
Shopify freed 76,500 hours via meeting purge	Shopify/WorkLife, 2023	Deleted all recurring 3+ person meetings; 33% meeting reduction; 25% more projects
Workflow redesign is #1 predictor of AI impact	McKinsey State of AI, 2025	Of 25 attributes tested; high performers 3.6x more likely to pursue transformational change
20% of orgs will flatten structure via AI by 2026	Gartner, October 2024	Eliminating 50%+ of middle management positions
37% of organizations layer AI with no process change	Deloitte State of AI, 2026	n=2,770+ executives, 16 countries
275 interruptions per day during core hours	Microsoft Work Trend Index, 2025	68% report insufficient uninterrupted focus time
Weekend work surged: Saturday +46%, Sunday +58%	ActivTrak, March 2026	Saturday start shifted to 7:11 a.m. from 8:35 a.m.

What This Means for Your Organization

The data presents an uncomfortable arithmetic. Every AI tool deployment increases the volume of output that needs coordination, review, and response. Without a corresponding subtraction of legacy processes, the net effect is more work — not less. The organizations in the 5% capturing real value from AI did not start by asking “what should we adopt?” They started by asking “what should we stop doing?”

The practical starting point is the five-question subtraction audit applied to any workflow where AI is deployed or planned. The questions are free. The discipline is hard. Every meeting you eliminate is a meeting someone scheduled for a reason — or at least a reason that existed once. Every report you cancel is a report someone spent years building. The political difficulty of subtraction is precisely why addition wins by default, and why the 95% keep adding tools to unredesigned workflows and wondering why the gains never materialize.

A mid-market company with 200-500 employees has a structural advantage here. Fewer layers of institutional memory. Fewer constituencies attached to legacy processes. A COO or CEO who can walk the floor and see, in real time, which meetings have three people in them who are checking email. The subtraction audit takes a week to run, costs nothing, and typically identifies 15-25% of coordination overhead as eliminable before a single AI tool changes. If this raised questions about where to start the audit in your organization, I’d welcome the conversation — brandon@brandonsneider.com.

Sources

ActivTrak Productivity Lab. “2026 State of the Workplace: AI Adoption and Workforce Performance Benchmarks.” March 2026. n=163,638 employees, 1,111 organizations, 443M+ hours analyzed. AI adoption subset: n=10,584 users, 180-day pre/post comparison. Credibility: High. Behavioral data from workplace monitoring, not self-reported surveys. Largest dataset on actual post-AI work patterns. https://www.activtrak.com/blog/2026-state-of-the-workplace/
Adams, G.S., Converse, B.A., Hales, A.H., & Klotz, L.E. “People systematically overlook subtractive changes.” Nature, Vol. 592, April 2021. Eight experiments, n=1,585 total. Credibility: Highest. Peer-reviewed, replicated across multiple experimental designs, Nature cover article. https://www.nature.com/articles/s41586-021-03380-y
BCG (Bedard, Kropp, Hsu, Karaman, Hawes, Kellerman). “When Using AI Leads to ‘Brain Fry.’” Harvard Business Review, March 2026. n=1,488 full-time U.S. workers. Credibility: High. Independent research by BCG’s behavioral science team with UC Riverside collaboration. https://hbr.org/2026/03/when-using-ai-leads-to-brain-fry
McKinsey & Company. “The State of AI: How Organizations Are Rewiring to Capture Value.” March 2025. Global survey. Credibility: High. Consulting firm survey with large enterprise sample; workflow redesign finding is the most actionable insight in their 2025 AI research. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
Microsoft Worklab. “Breaking Down the Infinite Workday.” Work Trend Index Special Report, 2025. Credibility: High. Based on M365 telemetry data from hundreds of millions of users. Note: Microsoft has commercial interest in AI productivity tools. https://www.microsoft.com/en-us/worklab/work-trend-index/breaking-down-infinite-workday
Shopify / WorkLife / Fast Company. “How to Create a 25% Productivity Hike: Lessons from Shopify’s Meeting Purge.” 2023. Company-reported data. Credibility: Moderate-High. Self-reported by Shopify executives; independently verified by multiple outlets but no independent audit of productivity claims. https://www.worklife.news/culture/how-to-create-a-25-productivity-hike-lessons-from-shopifys-meetings-purge/
Gartner. “Top Predictions for IT Organizations and Users in 2025 and Beyond.” October 2024. Credibility: High. Gartner’s annual prediction framework; middle management flattening projection based on current adoption trajectories. https://www.gartner.com/en/newsroom/press-releases/2024-10-22-gartner-unveils-top-predictions-for-it-organizations-and-users-in-2025-and-beyond
Gartner. “Survey Shows CFOs Are Trimming Overhead, But Not Revenue Growth Ambitions in 2026.” October 2025. Credibility: High. CFO survey data on SG&A planning. https://www.gartner.com/en/newsroom/press-releases/2025-10-15-gartner-survey-shows-cfos-are-trimming-overhead-but-not-revenue-growth-in-2026
Deloitte. “State of AI in the Enterprise, 2026.” n=2,770+ executives, 16 countries. Credibility: High. Large-sample global survey; process transformation segmentation is original analysis. https://www.deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/content/state-of-ai-in-the-enterprise.html
Asana. “Anatomy of Work Index.” Ongoing research series. Credibility: Moderate. Vendor-funded research; 58% “work about work” finding has been consistent across multiple years and aligns with independent data. https://asana.com/resources/why-work-about-work-is-bad
Fortune. “AI Isn’t Reducing Workloads — It’s Straining Employees.” March 13, 2026. Reporting on ActivTrak findings. https://fortune.com/2026/03/13/ai-isnt-reducing-workloads-its-straining-employees-time-spent-emailing-doubled-deep-focus-work-fell/

Brandon Sneider | brandon@brandonsneider.com March 2026