The Twenty Minute VCMatt Fitzpatrick: Who Wins the Data Labelling Race & Why Al Needs Forward-Deployed Engineers
Harry Stebbings and Matt Fitzpatrick on why AI deployment fails: Invisible’s CEO on data, FDEs, trust.
In this episode of The Twenty Minute VC, featuring Matt Fitzpatrick and Harry Stebbings, Matt Fitzpatrick: Who Wins the Data Labelling Race & Why Al Needs Forward-Deployed Engineers explores why AI deployment fails: Invisible’s CEO on data, FDEs, trust Matt Fitzpatrick, CEO of Invisible and former McKinsey senior partner, explains why enterprise AI adoption badly lags model performance and how Invisible is positioning as both an AI training platform and enterprise deployment partner. He argues most GenAI projects fail because enterprises treat them like SaaS apps rather than workflow, data, and change‑management problems that require forward-deployed engineers (FDEs) and rigorous validation. Invisible’s model hinges on modular software, human-in-the-loop data labeling at scale, and a "prove it first" go-to-market where customers don’t pay until systems actually work. Fitzpatrick also challenges myths around synthetic data, remote work, and out‑of‑the‑box agents, while outlining why he’s long‑term optimistic about AI’s impact on healthcare, energy, and education.
At a glance
WHAT IT’S REALLY ABOUT
Why AI deployment fails: Invisible’s CEO on data, FDEs, trust
- Matt Fitzpatrick, CEO of Invisible and former McKinsey senior partner, explains why enterprise AI adoption badly lags model performance and how Invisible is positioning as both an AI training platform and enterprise deployment partner. He argues most GenAI projects fail because enterprises treat them like SaaS apps rather than workflow, data, and change‑management problems that require forward-deployed engineers (FDEs) and rigorous validation. Invisible’s model hinges on modular software, human-in-the-loop data labeling at scale, and a "prove it first" go-to-market where customers don’t pay until systems actually work. Fitzpatrick also challenges myths around synthetic data, remote work, and out‑of‑the‑box agents, while outlining why he’s long‑term optimistic about AI’s impact on healthcare, energy, and education.
IDEAS WORTH REMEMBERING
5 ideasEnterprise AI is failing not because of weak models, but because of weak deployment.
Despite huge improvements in LLM benchmarks and mass consumer usage, only ~5% of enterprise GenAI deployments work; most organizations underestimate the need for data infrastructure, workflow redesign, ownership, observability, and trust processes like model risk management.
External, outcome-driven builds often outperform internal AI teams.
MIT data cited by Fitzpatrick suggests externally driven builds are roughly twice as effective as internal ones, largely because vendors are forced into ROI, milestones, and accountability in ways internal teams typically are not.
Forward-deployed engineers are becoming essential for real enterprise AI impact.
You can’t just sell an agent and walk away; to change workflows and embed AI deeply, you need FDEs who configure modular platforms to each customer’s specific processes and keep models fine-tuned as reality (e.g., new drugs, regulations) changes.
“Pay when it works” will pressure traditional SaaS-style pricing in AI.
Invisible does free 8‑week solution sprints and only charges once software passes user acceptance and delivers operational KPIs, reflecting a shift from license-first SaaS toward performance- and outcome-based pricing for AI deployments.
Human feedback and expert data won’t be replaced by synthetic data anytime soon.
For complex, multimodal, domain-specific reasoning (e.g., law, medicine, underwater drones), synthetic data can’t capture real-world nuance; you still need highly specialized human experts generating and validating data, often in extremely niche domains.
WORDS WORTH SAVING
5 quotes“If there’s an app for everything, how come nothing works?”
— Matt Fitzpatrick (quoting Invisible founder Francis Pedraza)
“Externally driven builds are 2X as effective as internal team builds.”
— Matt Fitzpatrick
“Out-of-the-box software has always been a lie to some degree.”
— Matt Fitzpatrick
“In the AI world at least, strategy is a somewhat overrated concept.”
— Matt Fitzpatrick
“The only risk is if you don’t take this and the amount of regret you’ll have not giving it a go.”
— Somesh Khanna (as recounted by Matt Fitzpatrick)
QUESTIONS ANSWERED IN THIS EPISODE
5 questionsWhat concrete steps should a large enterprise take in the first 90 days to move from AI ‘science projects’ to a focused, outcome-driven roadmap?
Matt Fitzpatrick, CEO of Invisible and former McKinsey senior partner, explains why enterprise AI adoption badly lags model performance and how Invisible is positioning as both an AI training platform and enterprise deployment partner. He argues most GenAI projects fail because enterprises treat them like SaaS apps rather than workflow, data, and change‑management problems that require forward-deployed engineers (FDEs) and rigorous validation. Invisible’s model hinges on modular software, human-in-the-loop data labeling at scale, and a "prove it first" go-to-market where customers don’t pay until systems actually work. Fitzpatrick also challenges myths around synthetic data, remote work, and out‑of‑the‑box agents, while outlining why he’s long‑term optimistic about AI’s impact on healthcare, energy, and education.
How can a mid-size startup realistically build a forward-deployed engineering function without blowing up their unit economics?
Where is the real breaking point between what synthetic data can handle and where human-labeled or expert data becomes indispensable?
How might Invisible’s ‘pay when it works’ model pressure incumbents like Accenture, large SIs, or traditional SaaS vendors to change their pricing and delivery?
If strategy cycles are effectively quarterly in AI, what does good board-level governance and planning actually look like for AI-native companies?
EVERY SPOKEN WORD
Install uListen for AI-powered chat & search across the full episode — Get Full Transcript
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome