I just read the PwC Global CEO Survey… and one number made me pause: 56%.
That is the percentage of organizations that are getting nothing from their AI investments. Zero return.
And yet… here we are. At AI4SP, 58 agents run our entire global operation. Our Fortune 500 clients built 4,000 agents last year—generating $50 million in documented value, and are forecasting over $250 million this year.
Same technology. Same access to ChatGPT, Claude, and Gemini. Radically different outcomes. So I wanted to know: who’s crazy—us or the market?
We stress-tested the data. Cross-referenced PwC against IBM, Deloitte, Gartner, MIT, McKinsey. We found a systemic failure… and a massive opportunity for those willing to do things differently.
The winners aren’t building “Super Agents.” They’re building small, ugly, functional agents that solve actual problems.
📊 The Failure Patterns: What the Data Shows
A common pattern found in failed AI deployments is leadership that reads reports and signs checks—but has never felt the hallucination or experienced the magic firsthand. You cannot lead a transformation you don’t understand, and you cannot outsource it completely to consultants.
Some consultants are less equipped to help than your frontline employees, who are learning by experimentation with multiple AI tools every single day. Find those team members, engage them, and pair them with experts who transformed their operation and use AI every single day. That’s the heart of your change management success.
| Pattern | The Data | Source |
|---|---|---|
| FOMO-Driven Purchases | 64% of CEOs bought AI software before understanding what it would do for them | IBM 2025 CEO Study |
| C-Suite Misalignment | 65% of CEOs are not aligned with their CFO on how to measure AI value | Kyndryl Readiness Report 2025 |
| Leadership Doesn’t Use It | 9 out of 10 failing organizations have leadership not actively using AI tools | AI4SP Research |
| Inverted Investment Ratio | 93% of budget goes to technology; only 7% to people | Deloitte CTO Interview |
| Pilots That Never Scale | Only 25% of AI initiatives delivered expected ROI | IBM 2025 CEO Study |
💡 The Reality Check:
The 7% investment in people and change management reported by Deloitte stands in sharp contrast to successful implementations, which invest 3 to 4 times that figure in the first year.
🏆 What Winners Do Differently
The difference isn’t smarter people or better technology. It’s approach.
| Failing Organizations | Winning Organizations |
|---|---|
| Top-down mandates from IT | Bottom-up, grassroots adoption |
| 12-month “super-agent” projects | 100 small sprints, each teaching something |
| Buy the license, assign training, done | 3-4x more investment in people during Year 1 |
| Leadership observes from a distance | Leadership actively uses the tools weekly |
| Measure hours saved as hours productive | Understand that 72% of saved time flows elsewhere |
The 80/80 rule: Top-down AI programs fail 80% of the time. Bottom-up programs succeed 80% of the time. When frontline workers build agents, they solve actual problems, not what leadership thinks the problems are.
💁♀️ Bella Evolution: From One Email to 70,000 Tasks
Agent Bella is a perfect case study. Twenty-five senior leaders at a global tech firm said, “We’re drowning in meetings. No time to prepare. No time to follow up.”
The IT team proposed a massive Chief of Staff agent—12-month project, six-figure budget.
We said no. Start with one thing: A daily briefing. If it works, then you will figure out the next functionality, and the next one…
| Phase | Capability | Timeline |
|---|---|---|
| V1 | Daily briefing email: tomorrow’s calendar, attendee research, key context | 2 weeks to build |
| V2 | Ability to interact via email and draft follow-up emails and responses to priorities | User request – week 3 |
| V3 | Join Slack channels, track action items | User request – week 5 |
| V4 | Flag overdue commitments, ability to create, edit and manage documents and handle those interactions via email, Slack, and text message | User request – week 8 |
| V5 | Full Chief of Staff | 6 months |
It started with one email automation. Each new improvement sparked new ideas. By the time they needed better orchestration, they had battle-tested capabilities, not an untested monolith designed in a conference room.
🧮 A Realistic Goal? The AI ROI Calculator
Vendor ROI spreadsheets assume every minute saved converts to productivity. That’s not how humans work—and it’s why traditional calculations overestimate returns by 157% on average.
What Makes It Different
| Traditional ROI Models | AI4SP ROI Framework |
|---|---|
| All time saved as productive output | Only 28-35% converts to direct throughput |
| Ignore quality improvements | Count quality gains (fewer errors, deeper analysis) |
| Miss innovation and strategic work | Show these benefits without over-monetizing them |
| Overlook wellbeing and retention | Acknowledge reduced burnout as real value |
| Numbers executives can’t defend | Conservative projections that survive CFO scrutiny |
The Research Behind It
| Metric | Value |
|---|---|
| Real-world use cases in research base | 180,000+ |
| Data points powering calculations | 50+ million |
| Industries covered | 18 |
| Countries represented | 70 |
| Average time NOT converting to throughput | 72% |
Organizations consistently underestimate enablement. The calculator models realistic cost distributions based on what successful deployments actually spend.
⏰ The Clock Is Ticking
Two forces are converging:
Rapid embedding: By end of 2026, 40% of enterprise apps will embed AI agents—up from less than 5% in 2025.
Bubble risk: The AI bubble could deflate any quarter. A bad earnings report, another DeepSeek-style disruption, and boards will demand clear metrics.
If you’re not measuring value by now, you’re late. If you’re a leader who hasn’t used AI this week, start now.
The 56% getting zero return followed the old playbook. Don’t be them.
🔗 Resources
- AI ROI Calculator: roicalc.ai
- Digital Skills Compass: skills.ai4sp.org
- All Research & Insights: ai4sp.org/insights
Luis J. Salazar | Founder & Elizabeth | Virtual COO (AI)
Sources:
Our insights are based on 1-billion data points from individuals and organizations who used our AI-powered tools, participated in our panels and research sessions, or attended our workshops and keynotes.
