The CEO’s “AI Is Not Delivering” Diagnostic: Five Questions That Separate a Fixable Problem from a Fatal One

Brandon Sneider | March 2026

Executive Summary

Nine months and $400K-$1.3M into an AI program, the board asks why there is no visible impact. This conversation happens at 42% of companies that invested in AI in 2025 — the abandonment rate more than doubled from 17% the prior year (S&P Global, n=1,006, March 2025). The question is not whether AI works. The question is whether your deployment addressed the right problem with the right approach.
The three most common root causes at mid-market scale are workflow bypass (AI layered on top of unchanged processes), sponsorship fade (the executive champion moved on within six months), and measurement theater (activity metrics reported as business outcomes). McKinsey finds that of 25 organizational attributes tested, workflow redesign has the single largest correlation with EBIT impact — yet only 21% of organizations using generative AI have redesigned any workflows (McKinsey, n=1,993, July 2025).
The 6% of organizations that attribute more than 5% of EBIT to AI did not avoid setbacks. They diagnosed faster and acted sooner. Their median time from early-warning signal to intervention is six weeks versus five months for the 94% (Pertama Partners, n=2,400+ initiatives, 2025). Speed of diagnosis, not speed of deployment, is the differentiator.
Five diagnostic questions — answerable in a single executive session — identify whether the AI program has a fixable problem or a structural one. The persist/pivot/kill decision follows directly from the answers, and the data shows that projects with pre-defined success metrics at launch achieve a 54% success rate versus 12% without them.
PwC’s 2026 Global CEO Survey (n=4,454) finds only 12% of CEOs report AI has delivered both cost and revenue benefits. But those 12% are two to three times more likely to have embedded AI across products, services, and decision-making — not bolted it onto existing operations. The path from the 88% to the 12% runs through five specific diagnostic questions.

The Uncomfortable Math of Nine Months In

The typical mid-market AI program reaches its accountability moment between month six and month twelve. The budget has been spent. The tools are deployed. Individual employees may report that AI saves them time. But the P&L looks the same, customer satisfaction has not moved, and the board wants to know what they bought.

This is not unusual. It is the default outcome.

BCG’s Build for the Future study (n=1,250 CxOs, September 2025) segments the market into three tiers: 5% of companies are “future-built” and generate substantial value from AI. 35% are “scalers” generating modest returns. The remaining 60% generate no material value despite active investment. McKinsey’s State of AI survey (n=1,993, July 2025) narrows the aperture further: only 6% of respondents — approximately 120 out of nearly 2,000 — attribute more than 5% of EBIT to AI use and describe the value as “significant.”

For a 300-person company that invested $400K-$800K in AI tools, training, and consulting over the past year, the board conversation is inevitable. The question is whether the CEO walks in with a diagnosis or walks in with excuses.

The Three Root Causes That Explain 85% of Mid-Market AI Disappointment

The data from multiple independent studies converges on three failure patterns that account for the vast majority of mid-market AI underperformance. Most stalled programs exhibit at least two simultaneously.

Root Cause 1: Workflow Bypass

What it looks like: AI tools are deployed into existing processes without changing how work gets done. Employees use ChatGPT or Copilot as a faster way to do the same tasks in the same sequence. The organization measures tool adoption (seats activated, queries per user) and reports those numbers to the board as progress.

Why it kills value: McKinsey tested 25 organizational attributes against EBIT impact from AI. Workflow redesign had the single largest effect. Yet only 21% of organizations using generative AI have redesigned even some workflows. The other 79% are layering AI on top of processes designed for a pre-AI world.

Consulting Magazine’s analysis (February 2026) frames the mechanism precisely: “Organizations modernized intelligence without modernizing how work gets done.” Task-level productivity improved — individuals completed certain activities faster — while enterprise-level results stalled because the surrounding process was the bottleneck, not the task itself.

At mid-market scale: A 300-person company automates invoice processing with AI, cutting individual processing time by 40%. But the approval chain, exception handling, and vendor communication workflow remain unchanged. The 40% time savings disappears into the same bottleneck downstream. The board sees the same days-payable-outstanding metric and asks what AI actually accomplished.

Root Cause 2: Sponsorship Fade

What it looks like: The CEO or COO who championed the AI initiative with enthusiasm and budget encounters the first real obstacle — data quality, integration complexity, employee resistance — and attention shifts to the next priority. Without sustained executive authority, the project team cannot make cross-functional demands. Resources get quietly reallocated. The initiative drifts.

The data: 56% of failed AI projects lose active C-suite sponsorship within six months. Projects with sustained executive sponsorship achieve a 68% success rate versus 11% for those that lose it — a 6.2x differential (Pertama Partners, n=2,400+, 2025). Loss of executive sponsorship accounts for 21% of formal abandonment decisions.

At mid-market scale: At a 500-person company, the CEO who sponsors AI is also managing operations, clients, investor relations, and three other strategic priorities. AI is rarely the only transformation competing for bandwidth. McKinsey finds that AI high performers’ senior leaders “show clear ownership and long-term commitment” — role-modeling usage, protecting budgets, repeatedly sponsoring initiatives through obstacles. Only 16% of other organizations report the same behavior.

Root Cause 3: Measurement Theater

What it looks like: The AI program reports metrics that sound impressive but do not connect to business outcomes. “3,200 prompts processed this month.” “87% of licensed seats activated.” “Employee satisfaction with AI tools: 4.2/5.” The board hears activity metrics packaged as success metrics and eventually realizes no one defined what “delivering” means.

The data: 73% of failed AI projects lack clear executive alignment on success metrics at launch (Pertama Partners, 2025). Projects with pre-defined success criteria achieve a 54% success rate versus 12% without — the single largest controllable variable in project outcomes. Deloitte’s State of AI 2026 (n=3,235) finds only 40% of organizations rate their AI strategy as “highly prepared,” with governance trailing at 30% — numbers that actually declined from the prior year.

At mid-market scale: A 300-person company defines AI success as “increased productivity.” Nine months later, no one can agree on how to measure productivity, which departments should have improved, or what baseline they are comparing against. The CFO asks for the dollar impact. No one has an answer.

The Five Diagnostic Questions

These five questions can be answered in a single executive session. Together, they identify whether the AI program has a fixable problem, a structural one, or a combination. The answers determine the persist/pivot/kill decision.

Question 1: What specific business metric was this AI investment supposed to move — and did we define it before we started?

What a good answer sounds like: “We targeted a 15% reduction in customer response time in our support operation, measured against our Q1 baseline of 4.2 hours median.”

What a bad answer sounds like: “We wanted to increase productivity across the organization.” Or: “We wanted to stay competitive.”

Why it matters: The 54% vs. 12% success rate gap between projects with and without pre-defined metrics is the most actionable finding in the entire failure literature. If the answer to this question is vague, the problem began before any tool was purchased.

Question 2: Did we change any workflow, or did we add AI to the existing process?

What a good answer sounds like: “We eliminated the three-step manual review for standard invoices under $5,000 and replaced it with AI-flagged exception handling. The full approval chain now only engages on flagged items.”

What a bad answer sounds like: “We gave everyone access to Copilot and trained them on prompt engineering.”

Why it matters: The McKinsey finding is unambiguous. Among 25 tested attributes, workflow redesign is the single strongest predictor of EBIT impact. AI that accelerates a broken process produces a faster broken process.

Question 3: Is the executive who championed this initiative still actively sponsoring it — attending reviews, resolving cross-functional blockers, protecting the budget?

What a good answer sounds like: “The COO chairs a monthly AI steering committee, resolved a data-access conflict with IT last quarter, and rejected a proposal to cut the AI training budget.”

What a bad answer sounds like: “The CIO launched it, but they’ve been focused on the ERP migration since July.”

Why it matters: The 68% vs. 11% success rate differential on sponsorship is the second largest finding. An AI program without active executive sponsorship at a mid-market company is an IT experiment, not a business initiative.

Question 4: Can anyone in this room state the dollar impact of our AI investment — not the tool cost, but the business value generated?

What a good answer sounds like: “AI-assisted customer support reduced average handle time by 22%, which at our volume translates to $180K in annualized labor savings and a 9-point improvement in CSAT.”

What a bad answer sounds like: “We’ve had strong adoption — 87% of seats are active and employees report saving 30 minutes per day.”

Why it matters: PwC’s 2026 CEO Survey (n=4,454) finds only 12% of CEOs report AI has delivered both cost and revenue benefits. Those CEOs are two to three times more likely to have embedded AI across products, services, and decision-making. Embedding requires measurement. Self-reported time savings are not measurement.

Question 5: If we stopped this program today, what would break?

What a good answer sounds like: “Our customer support team now depends on AI triage for 60% of inbound tickets. Stopping would require hiring two additional support agents within 30 days.”

What a bad answer sounds like: “Nothing would break. People would just go back to doing things the old way.”

Why it matters: This is the acid test. If AI can be removed with no operational consequence, it has not been integrated — it has been overlaid. The 5% of organizations generating substantial value from AI (BCG, September 2025) have made AI load-bearing in at least one critical process. The other 95% are running a side experiment.

The Persist/Pivot/Kill Decision

The answers to the five diagnostic questions map directly to a decision framework.

Diagnostic Pattern	Answers	Decision	Next Step
Fixable: Wrong metrics	Q1 vague, Q2-Q5 reasonable	Persist with redefinition	Define 2-3 specific KPIs, set 90-day measurement sprint, report to board monthly
Fixable: Workflow bypass	Q2 is “no workflow change”	Pivot to process redesign	Select one high-volume process, redesign around AI, measure before/after
Fixable: Sponsorship gap	Q3 reveals disengaged sponsor	Persist with re-engagement	CEO re-commits or appoints a new sponsor with explicit authority and calendar commitment
Structural: Multiple failures	Q1 vague + Q2 no change + Q3 disengaged	Kill and restart	Terminate current program, conduct 30-day post-mortem, re-launch with metrics, workflow redesign, and active sponsorship as prerequisites
Working but unmeasured	Q5 shows dependency, Q4 cannot quantify	Persist with measurement	Instrument the process AI supports, calculate business impact, report in financial terms

The kill decision is not a failure. It is a diagnostic outcome. Companies that kill underperforming AI initiatives and restart with discipline recover faster than those that let stalled programs consume budget and credibility. Apple terminated Project Titan (autonomous vehicles) after a decade and billions invested, redirecting 2,000 engineers to generative AI. Tesla shut down its Dojo supercomputer project in August 2025 despite projections of $500 billion in valuation impact, reallocating engineers to higher-return AI priorities.

At mid-market scale, the stakes are proportionally smaller but the principle is identical: capital and executive attention are finite. A $600K AI program that is not delivering value is consuming resources that could fund a program that does.

Key Data Points

Finding	Source	Detail
60% of companies generate no material value from AI	BCG Build for the Future (n=1,250 CxOs, September 2025)	5% “future-built,” 35% “scalers,” 60% laggards
Only 6% of organizations see >5% EBIT impact from AI	McKinsey State of AI (n=1,993, July 2025)	~120 out of ~2,000 respondents qualify as “high performers”
Only 12% of CEOs report AI delivered both cost and revenue benefits	PwC Global CEO Survey (n=4,454, November 2025)	Those 12% are 2-3x more likely to have embedded AI enterprise-wide
56% of failed projects lose C-suite sponsorship within 6 months	Pertama Partners (n=2,400+ initiatives, 2025)	68% success rate with sustained sponsorship vs. 11% without
Workflow redesign is the #1 predictor of AI EBIT impact	McKinsey State of AI (n=1,993, July 2025)	Only 21% of gen AI-using organizations have redesigned any workflows
54% success rate with pre-defined metrics vs. 12% without	Pertama Partners (n=2,400+ initiatives, 2025)	Largest controllable variable in AI project outcomes
42% of companies abandoned majority of AI initiatives in 2025	S&P Global (n=1,006, March 2025)	Up from 17% in 2024 — rate more than doubled
Mid-market failed experimenter sunk costs: $630K-$1.3M	Pertama Partners, 2026	Large enterprise average: $4.2M per abandoned project
94% of CIOs say core data needs significant cleanup before AI	Consulting Magazine, February 2026	Only 7% believe historical data is truly AI-ready
37% of organizations use AI at surface level only	Deloitte State of AI 2026 (n=3,235, August-September 2025)	No process changes, minimal impact — the largest single segment

What This Means for Your Organization

The board conversation about AI ROI is not a crisis. It is a diagnostic opportunity — and one that 42% of your peers are having right now. The fact that the question is being asked means the organization is paying attention. Companies that never ask the question are the ones that spend 18 months in pilot purgatory before quietly shelving their AI program.

The five diagnostic questions take an hour to work through. The answers will fall into one of the patterns in the decision table above. For most mid-market companies — based on the data showing that 79% have not redesigned workflows and 73% launched without clear success metrics — the answer will be a pivot, not a kill. The AI investment is not wasted. The approach needs recalibration: define what “delivering” means in dollar terms, redesign one process around AI rather than bolting AI onto all processes, and ensure active executive sponsorship with protected calendar time and decision authority.

The companies in the 5-6% capturing real value did not get there by investing more or deploying faster. They got there by asking better questions earlier. If working through this diagnostic raised questions specific to your organization’s situation, I would welcome the conversation — brandon@brandonsneider.com.

Sources

BCG, “Build for the Future: The Widening AI Value Gap” (n=1,250 CxOs and senior executives, 68 countries, September 2025). Independent consulting research. High credibility for strategic segmentation. https://www.bcg.com/publications/2025/are-you-generating-value-from-ai-the-widening-gap
McKinsey & Company, “The State of AI: Global Survey” (n=1,993 participants, 105 nations, June-July 2025). Independent consulting research. High credibility; the workflow redesign finding is especially well-supported. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
PwC, “2026 Global CEO Survey” (n=4,454 CEOs, 95 countries, September-November 2025). Large-sample independent survey. The 12% dual-benefit finding is the clearest CEO-level AI ROI data available. https://www.pwc.com/gx/en/news-room/press-releases/2026/pwc-2026-global-ceo-survey.html
Pertama Partners, “AI Project Failure Statistics 2026: The Complete Picture” (n=2,400+ AI initiatives, aggregated from RAND, MIT Sloan, Deloitte, Gartner, and proprietary data, 2025-2026). Aggregated secondary analysis. Credibility is moderate-to-high; the individual source studies are independently verifiable. https://www.pertamapartners.com/insights/ai-project-failure-statistics-2026
Deloitte, “State of AI in the Enterprise 2026” (n=3,235 business and IT leaders, 24 countries, August-September 2025). Independent consulting research. One of the largest AI adoption surveys available. https://www.deloitte.com/us/en/about/press-room/state-of-ai-report-2026.html
S&P Global, “Voice of the Enterprise AI Survey” (n=1,006 IT and business leaders, North America and Europe, March 2025). Independent analyst research. Strong sample quality for enterprise technology adoption data.
Consulting Magazine, “Why Enterprise AI Stalled and What Is Finally Changing in 2026” (February 2026). Trade publication analysis drawing on Gartner, CFO survey data, and enterprise case studies. https://www.consultingmag.com/2026/02/04/why-enterprise-ai-stalled-and-what-is-finally-changing-in-2026/
Virtasant, “A Smarter Enterprise AI Strategy Starts With Knowing When to Stop” (2025). Practitioner analysis with case studies from Tesla, Apple, IBM, Amazon, and Volkswagen. Framework credibility is moderate; the corporate examples are independently verifiable. https://www.virtasant.com/ai-today/a-smarter-enterprise-ai-strategy-starts-with-knowing-when-to-stop

Brandon Sneider | brandon@brandonsneider.com March 2026