Skill. AI. Execute – The Only Words that matter in 21st Century
Why 95% of Enterprise AI Pilots Fail, and What SPeXecute Does Differently
An operating thesis on why most companies are buying AI subscriptions instead of capturing AI value and how the gap actually gets closed.
The companies SPeXecute works with already know their business.
For Ex: Real estate operators understand pipeline conversion. Talent agencies understand booking dynamics. Hospitality groups understand occupancy and ADR. Fashion houses understand seasonal margin. Founders in professional services understand client economics better than any consultant ever will.
That is not the problem.
The problem is everywhere else.
Things take longer than they should. Teams repeat work they have already done three times. Information lives in five different systems.
Leadership cannot see what is actually happening in the business because the reporting layer was never properly built.
This is operational drag, and it is what AI is supposed to fix. Mostly, it doesn’t, because most companies are buying tools instead of redesigning operations. The data on this is now unambiguous.
The 95/5 Problem: What the Hard Data Actually Says
In August 2025, MIT’s Project NANDA published The GenAI Divide: State of AI in Business 2025, the most comprehensive empirical study of enterprise AI deployments to date.
The headline finding is brutal: 95% of corporate generative AI pilots deliver no measurable impact on P&L. Only 5% of projects achieve rapid revenue acceleration, despite an estimated $30–40 billion in enterprise GenAI investment to date.
The cause is not the models. The MIT researchers were explicit about this.
The failure is organizational. What the report calls the “learning gap” between AI capability and the way companies actually integrate it.
Internal builds succeed about 1/3 as often as deployments built in partnership with specialized vendors, which succeed roughly 67% of the time.
Companies that staffed AI as an isolated lab, separate from operating teams, consistently underperformed companies that pushed adoption through line managers who owned the workflow.
McKinsey’s State of AI 2025, surveying 1,993 respondents across 105 countries between June and July 2025, tells the same story from a different angle. 88% of organizations now use AI in at least one business function. But only 6% qualify as “AI high performers” i.e companies attributing more than 5% of EBIT to their AI use and reporting significant enterprise value. 39% report any EBIT impact at all. Nearly two-thirds remain in pilot or experimentation mode.
The single strongest correlation McKinsey found with EBIT impact was whether the company had fundamentally redesigned at least some of its workflows around AI. Only 21% had done so. The other 79% were layering AI on top of existing processes without rethinking how work actually flows, which, predictably, produces marginal returns at best.
This is the 95/5 Problem. The technology is widely available. But not all are able to capture the value.
What’s the deal with it? Where the Drag Actually Lives
Before talking about AI, it helps to understand what AI is being deployed against. The operational substrate of most mid-sized businesses is leakier than leadership realizes.
ProcessMaker’s 2024 research, drawn from over four million data points across enterprise clients, found that a typical office worker performs more than 1,000 copy-paste operations per week or over 52,000 per year. The same study found employees spend roughly 10% of their time on manual data entry into ERP, CRM, and spreadsheet systems, and more than 50% of their time creating or updating documents.
A separate analysis cited by Harvard Business Review put the figure higher: knowledge workers spend an average of 41% of their day on activities that do not draw on their primary skills.
This is the actual surface area AI is supposed to address. Not “writing better emails.” The drag is structural: information is in the wrong places, workflows are not designed for the tools being used, and people are spending a quarter to a half of their day on coordination overhead rather than on the work they were hired to do.
A Claude subscription does not fix that even if it the the much rumoured Mythos. A workflow architecture does.
What Separates the 6% “AI High Performers”
The McKinsey data on high performers is the most useful diagnostic in the category, because it isolates what the top 6% are doing differently rather than describing what the average company says about AI.
The pattern across all five is the same: AI value capture is an operating-model problem, rather than a technology problem. The 6% that win are the ones that treat it that way.
Five patterns hold across the cohort:
Workflow redesign over tool deployment: High performers are nearly three times more likely than other companies to have fundamentally redesigned workflows around AI. They are not running AI as an overlay on legacy processes.
Line-manager ownership over central AI labs : Both McKinsey and MIT NANDA found that AI adoption pushed through operating managers, people who own the P&L the AI is supposed to affect outperforms adoption driven by a central AI team disconnected from operations.
Partnerships over internal builds: MIT NANDA found purchased and partner-built solutions delivered measurable results about 67% of the time. Internal builds, despite typically having more budget and senior sponsorship, delivered roughly a third of that.
Transformation ambition over efficiency targets : McKinsey’s high performers were 3.6x more likely to set transformative business change as the goal of their AI work. !Not Cost Reduction! Companies that aim only for incremental efficiency consistently get incremental returns.
Investment concentration: Over a third of high performers commit at least 20% of their digital budget to AI. The companies stuck in pilot purgatory typically allocate around 5%.
Where AI Genuinely Fits? And Where It Doesn’t
Someone in leadership read a thread, signed up the team for a productivity copilot, and assumed value would follow. This is what happens because of the noise and this is
decoupled from any specific operational outcome.
The integration that produces value tends to fall into a small set of categories, and they are roughly the categories where MIT NANDA found the highest success rates:
Back-office automation – invoice processing, expense classification, contract review, vendor onboarding, document extraction. The MIT report found ROI was meaningfully higher in back-office automation than in sales and marketing, which is the opposite of where most enterprise AI budget actually flows.
Reporting and decision infrastructure – replacing manual weekly reporting cycles with continuously updated dashboards leadership can actually look at without asking someone to “pull the numbers.”
Workflow orchestration – connecting systems that currently require human copy-paste between them, so the work moves automatically and the human only intervenes on exceptions.
Knowledge retrieval – making the institutional knowledge that currently lives in three Slack threads, two Notion pages, and one person’s head actually queryable.
Specialized classification and scoring — domain-specific judgment tasks that can be encoded into a model trained or prompted on the company’s actual operating standards.
What does not work, consistently, is buying horizontal AI tools and waiting for them to produce vertical results.
The MIT report calls the resulting pattern the “shadow AI economy” , employees individually subscribing to consumer AI tools because the company’s official deployments are not solving their actual problems.
That is a symptom of the underlying issue: AI was bought, not integrated.
The Execution Layer
SPeXecute’s role is the execution layer between industry knowledge and AI systems.
That means a few specific things in practice:
1. Mapping where operational drag actually lives in a business before recommending any tool, because we know that tool-first deployments are the failure mode, not the success mode.
2. Building the workflow architecture i.e what systems connect to what, where the human is in the loop, where automation runs unattended, what the audit trail looks like, before any model is fine-tuned, workflow is redesinged or prompt is written.
3. Choosing integration over invention. As we learned above from what experts at MIT say, most companies should not be building their own AI infrastructure from scratch. The actual enterprise advantage lies in assembling proven, pre-existing models and applying them precisely where the business needs them. Bypassing the multi-year cost of raw infrastructure development allows companies to bypass the lab and move straight to value capture.
4. Treating reporting and decision infrastructure as part of the AI deployment, not a separate IT project. We have experienced AI without a clean decision layer is just faster guessing.
5. Designing for the line manager who owns the outcome, not the central AI committee that owns the budget. This is the difference between AI as a product purchase and AI as a business infrastructure.
The companies that will benefit from AI over the next three to five years are the ones that treat it as the latter. Everyone else is funding the GenAI Divide from the wrong side of it.
Frequently Asked Questions
FAQs
Adoption is using AI in at least one business function, 88% of organizations now do this, per McKinsey. Value capture is generating measurable enterprise impact , only 6% of organizations qualify as high performers with more than 5% of EBIT attributable to AI use. The gap between the two is what MIT NANDA calls the “GenAI Divide.”
MIT NANDA’s 2025 data found ROI was higher in back-office automation than in sales and marketing, despite most enterprise AI budgets being allocated to the latter. The reason is that back-office workflows are more structured and the success criteria are clearer, making AI integration more measurable.
The MIT 2025 data is decisive on this: purchased solutions and partner-built deployments succeeded roughly 67% of the time, while internal builds succeeded about one-third as often. Most companies underestimate the engineering, data, and operating-model work required to build AI infrastructure from scratch.
We at SPeXecute can be your Partner in this.
SPeXecute operates across real estate, talent management and modeling agencies, hospitality and luxury, fashion and consumer brands, and professional services , selecting industries where the drag is significant and AI integration has clear leverage on commercial outcomes.
Sources:
- MIT Project NANDA, The GenAI Divide: State of AI in Business 2025 (2025). https://nanda.media.mit.edu
- McKinsey & Company, The State of AI in 2025: Agents, Innovation, and Transformation (November 2025). https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
- ProcessMaker, Repetitive Tasks at Work Research and Statistics 2024. https://www.processmaker.com/blog/repetitive-tasks-at-work-research-and-statistics-2024
- Qatalog and Cornell University, Workgeist Report (2023).
- Asana, Anatomy of Work Index.
- Harvard Business Review, on knowledge work and time allocation.
About SPeXecute
SPeXecute is a premium AI and data consulting firm specializing in the operational integration of AI systems into real businesses. We focus on workflow architecture, automation, and decision infrastructure across real estate, talent management, hospitality, fashion, and professional services. You have the skill. We know AI. Let’s execute.
You have the skill. We know AI.
Partner with SPeXecute
