#TheExecutionGap GOOD METRICS CAN HIDE BAD GOVERNANCE.
88% of companies use AI.
Under 6% see it move EBIT.
McKinsey's 2025 State of AI survey shows why: 39% report any EBIT effect, and fewer than 6% call it significant.
Deploying AI is easy.
Keeping the authority to question it is not.
✅ Shadow AI isn't a security lapse. It's the default state of adoption.
In 2023, Samsung engineers uploaded sensitive source code to ChatGPT, 3 leaks in 20 days.
The company banned generative AI on corporate devices within a week.
Microsoft and LinkedIn's Work Trend Index confirmed the pattern: 78% of AI users bring their own tools without IT approval, 85% among Gen Z.
The tool wasn't hidden. It outran approval.
Check: Name an AI tool your team uses that nobody approved.
✅ A pilot that never gets killed isn't caution. It's drift.
McDonald's tested AI voice ordering with IBM at over 100 drive-thrus in 2021.
By 2024, viral videos showed the system adding 260 nuggets to one order and nine sweet teas to another.
McDonald's ended the partnership that June, three years in, without ever scaling past a small fraction of its more than 13,000 US restaurants.
MIT's Project NANDA found in 2025 that 95% of enterprise generative AI pilots never reach measurable P&L impact.
Check: Name the pilot that has run over a year with no scale or kill decision. If it exists, ask who was supposed to decide.
✅ This is false adoption, the same failure mode this series named at the universal layer. Here it has its own evidence.
In February 2024, Klarna announced its AI assistant did the work of 700 agents, cutting resolution time from 11 minutes to under 2.
By May 2025, its CEO admitted cost had outweighed quality.
In 2026, Klarna was rehiring humans as gig workers, reframing human service as a premium tier.
The metric was real. What it measured wasn't.
Check: Pick your sharpest AI metric and name the quality it stopped measuring.
✅ Governance ambiguity isn't the absence of control. Sometimes control is switched off deliberately.
Zillow's home buying algorithm ran under the internal name Project Ketchup.
Pricing experts were explicitly barred from questioning its valuations.
The first three drifted. This one was designed.
No one held what auditors call the second line of defense: a supervisory layer, separate from the operating team, whose job is to challenge the model before it commits real money.
Zillow wrote down over $500 million and cut a quarter of its workforce before shutting the unit down in November 2021.
Check: Name who can override your best performing model, and when they last did.
None of these four systems failed on capability.
Each failed the moment someone needed standing to call it wrong, and didn't have it.
Adoption is not the signal.
The signal is who still has standing to say the system is wrong.
#TheExecutionGap · Layer 2, Digital Transformation · 2.2, DT How
#DigitalTransformation #AIGovernance #OrganizationalGovernance #OriacGimeno