Matthew Fitzpatrick is an experienced operations and growth expert with deep expertise in scaling complex workflows and teams. His background spans consulting, strategy and operations leadership, and he is currently CEO of Invisible Technologies, where he focuses on designing and optimizing end-to-end business solutions. Matthew is passionate about combining talent with automation to increase efficiency scale, helping companies unlock transformative growth through process innovation.
Invisible Technologies is a business process automation company that integrates advanced technologies with human expertise to help organizations scale effectively. What is invisible is not to replace humans with automation, but to create custom workflows where digital workers (software) work seamlessly with human operators. The company offers services across areas such as data richness, lead generation, customer support and backend operations to enable customers to delegate complex, repetitive tasks and focus on core strategic goals. Invisible’s unique “work-service” model provides businesses with scalable, transparent, and cost-effective operational support.
You recently transitioned from McKinsey’s leading Quantumblack Lab to becoming CEO of Invisible Technologies. What attracted you to this role and what was most excited about Invisible’s mission?
At McKinsey, I have the honor to work at the forefront of AI innovation – building AI software products, leading R&D efforts, and helping businesses leverage the power of data. What attracted me to intangible technology is the opportunity to combine it with a unique and flexible AI software platform and expert market for human loop feedback to make it run at scale – I believe learning from human feedback (RLHF) is key to accurate and reliable Genai implementation. From data cleaning and data input automation to thoughtful reasoning and custom evaluation, Invisible supports AI throughout the value chain. Our mission is simple: combine human intelligence and AI to help businesses realize their AI’s potential, which is much more difficult in enterprises than most people would expect.
You have overseen over 1,000 engineers and expand multiple AI products across the industry. Have you applied for a McKinsey course from Invisible’s next growth phase?
Two courses stand out. First, successful AI adoption is as important as organizational transformation. You need the right people and processes – on top of a great model. Second, the companies that win AI are the ones that master the “last mile” – the transition from experimentation to production. In Invisible, we are applying the same rigor and structure to help customers move beyond pilots and bring them into production, thus bringing real business value.
You have already said: “2024 is a year of AI experimentation, and 2025 is about realizing ROI.” What specific trends do you see between businesses actually achieve ROI?
Businesses that see real ROI do a great job this year. First, they align AI use cases with core business KPIs such as operational efficiency or customer satisfaction. Second, they are investing in quality data and human feedback loops to continuously improve model performance. Third, they moved from universal solutions to tailor-made domain-specific systems that reflect the complexity of their environment. These companies are no longer just testing AI, but are scaling with purpose.
What is the need for domain specificity and PhD data labeling in basic model providers such as AWS, Microsoft, and Cohere?
We have seen a surge in demand for professional labels as basic model providers grow. If not visible, our expert pool has an annual acceptance rate of 1%, while 30% of coaches have a master’s or doctoral degree. This deep expertise is increasingly necessary – not only to accurately annotate the data, but also to provide subtle, context-aware feedback for improved reasoning, accuracy, and consistency. As models become smarter, the standards for training them become higher.
What is invisible is the forefront of proxy AI, emphasizing decision-making in the real world. What is your definition of proxy AI and where do we see the most promising?
Agent AI refers to systems that not only respond to instructions—they plan, make decisions and act within a defined guardrail. AI behaves more like teammates than tools. What we see is the most attractiveness in a large, complex workflow: customer support and insurance claims, for example. In these areas, proxy AI can reduce manual efforts, improve consistency and deliver results that otherwise require large human teams. This is not about replacing humans – rather, we are augmenting them with smart agents that can handle repetition and routines.
Can you share examples of the invisible train model for thought chain reasoning and why it is crucial for enterprise deployment?
Through thinking chain (COT) reasoning unlocks new potential for enterprise AI. At Invisible, we train models to reason step by step, which is crucial when the bet is high – whether you are diagnosing patients, analyzing contracts, or validating financial models. COT not only improves transparency, but also debugs, improvements and performance improvements without a large number of new data sets. We have seen leading models like Gemini, Sonnet, and Grok begin to disclose their inference paths, which allows us to observe not only what model outputs, but also how they get there. This bases on more advanced methods such as Tree Tree of Thought (the model evaluates multiple possible reasoning paths before solving the answer) and self-consistent (the multiple reasoning paths are explored).
Invisibly supports training across more than 40 coding languages and more than 30 human languages. How important is cultural and linguistic precision in building global scalable AI?
This is very important. Language is more than just translation, it’s about context, nuance and cultural norms. If the model misunderstands the tone or misses the area changes, it can lead to poor user experience and even compliance risks. Our multilingual coaches are not only fluent—they are embedded in the culture they represent.
What are the common points of failure when companies try to go from proof of concept to production, and invisible help how to navigate the “last mile”?
Most AI models have never been produced because the company underestimates the operational lifts needed. They lack clean data, strong evaluation protocols, and strategies to embed models into actual workflows. Invisible, we combine deep technical experience with production-grade data infrastructure to help businesses bridge the gap. Our symbiotic ability in training and optimization allows us to build better models and deploy them successfully.
Can you guide us through Invisible’s RLHF approach (learning from human feedback) and how it differs from other approaches in the industry?
In the case of invisibility, we believe that reinforcement from human feedback (RLHF) learning is more than just fine-tuning – it allows for more complex custom evaluation (“evaluation”) designs and moves towards training models with nuanced human judgments rather than binary signals (such as binary signals) to trained models instead of thumbs and thumbs. Although industry approaches often prioritize scale with large amounts of low-signal data, we focus on collecting structured high-quality feedback to capture reasoning, context, and trade-offs. This richer signal enables the model to be more efficiently generalized and more consistent with human intentions. By prioritizing depth over breadth, we are building infrastructure to keep more powerful AI systems consistent.
How do you envision the future of AI-Human collaboration, especially in high-risk areas such as finance, health care or the public sector?
AI has not replaced human expertise – it has become the infrastructure to support it. I envision a future where AI agents and human experts work in sync – clinicians are supported by diagnostic copilots, government agencies use AI to categorize more effectively, and financial analysts are free to focus on strategies rather than spreadsheets. Our focus is on designing systems that AI enhances human capabilities, rather than masking or vetoing.
Thanks for your excellent interview, and hopefully learn more about the readers who should access Invisible Technologies.
