Matt Fitzpatrick on Who Wins the Data Labelling Race & Why AI Needs Forward-Deployed Engineers?
•December 31, 2025
0
Why It Matters
Enterprise AI adoption hinges on disciplined, externally sourced engineering and CFO governance; without them, most generative‑AI projects will fail to deliver value.
Key Takeaways
- •Only five percent of GenAI deployments succeed in enterprises.
- •Externally built AI solutions are twice as effective as internal teams.
- •Synthetic data cannot replace human feedback for reliable AI performance.
- •Forward‑deployed engineers bridge data, trust, and workflow gaps.
- •CFOs must enforce clear data, milestones, and ownership for AI projects.
Summary
Matt Fitzpatrick discusses why the AI data‑labeling race is stalling in large enterprises, emphasizing that only a tiny fraction of generative‑AI projects ever reach production. He cites MIT’s finding that just five percent of GenAI deployments work in any form and Gartner’s forecast that 40% of enterprise AI initiatives will be cancelled by 2027, underscoring a widening gap between model performance and real‑world adoption.
The conversation highlights several root causes: external, specialist‑driven builds outperform internal teams by roughly two‑to‑one, synthetic data alone cannot substitute for human‑in‑the‑loop feedback, and successful deployment demands robust data pipelines, workflow redesign, trust, and observability. Fitzpatrick’s company, Invisible, offers a modular platform that combines reinforcement learning with human feedback, aiming to close the chasm between impressive benchmark gains and enterprise‑grade reliability.
He illustrates the stakes with a $25 million e‑commerce agent that was scrapped after months because it lacked proper evaluation and caused hallucinations. He also notes his own “free‑for‑eight‑weeks” proof‑of‑concept approach, and recalls advice from mentors that the biggest risk is not taking the plunge. These anecdotes reinforce the need for forward‑deployed engineers who can translate cutting‑edge models into disciplined, accountable production systems.
For CEOs and CFOs, the takeaway is clear: AI success will hinge on partnering with external experts, establishing rigorous data governance, setting measurable milestones, and assigning clear ownership. Forward‑deployed engineering talent and CFO oversight of ROI, risk, and compliance will become essential differentiators as enterprises move from experimental pilots to sustainable AI operations.
Original Description
Matt Fitzpatrick is the CEO of Invisible Technologies, leading the company's mission to make AI work. Since joining as CEO in January 2025, he has raised $100M, and accelerated AI adoption across industries from sports to consumer and government. Previously, Matt was a Senior Partner at McKinsey, where he led QuantumBlack Labs, the firm's AI R&D and software development arm.
-----------------------------------------------
Timestamps:
00:00 Intro
01:21 Career Journey and Leadership
08:36 The Single Biggest Barriers to Enterprises Adopting AI
10:41 It is BS That Enterprises Can Adopt AI Without Forward-Deployed Engineers
27:13 Are AI Talent Marketplaces Dead? What is the best model?
36:39 How Does the Data Labelling Market Shake Out: Who Wins/ Who Loses
50:01 Are Revenue Numbers for Data Labelling Real Revenue? Or GMV?
52:56 Best Capital Allocation Decision? What did Matt Learn from it?
55:22 How Important is Brand for AI Companies Selling Into Enterprise?
01:09:24 Remote Work vs. In-Person Collaboration
01:21:47 What Does No-One Know About the Future of AI That Everyone Should Know
-----------------------------------------------
Subscribe on Spotify:
Subscribe on Apple Podcasts:
Follow Harry Stebbings on X:
Follow Invisible Technologies on X:
Follow 20VC on Instagram:
Follow 20VC on TikTok:
Visit our Website:
Subscribe to our Newsletter:
-----------------------------------------------
#20vc #harrystebbings #mattfitzpatrick #invisibletechnologies #datalebelling #ai #engineers #saas
0
Comments
Want to join the conversation?
Loading comments...