When everyone is talking about ChatGPT, Cursor, and Sora, one company founded nine years ago has quietly tripled its annual revenue.
Its team has fewer than 30 people. Its customer list includes UBS, Lenovo, Dell, Comcast, and NBCUniversal. Its founder is a computer science professor at North Carolina State University who previously worked at IBM and Google.
The company is called InsightFinder. It builds “full-stack observability” for enterprise AI systems. In plain language: when an enterprise AI model goes wrong, InsightFinder helps engineers quickly determine whether the problem is in the model itself, data drift, or the underlying infrastructure.
This is not a sexy story. But it is a story builders should study carefully.
1. Why This Company?
From 2024 to 2025, the AI agent wave swept through the technology industry. But most people only saw the consumer-side excitement: chatbots, AI writing, AI video. In the deep waters of enterprise, a quieter revolution was happening.
InsightFinder is a representative of that deep water.
According to an exclusive TechCrunch report, the company recently raised a $15 million Series B led by Yu Galaxy. But the financing amount is less important than its revenue growth: founder Helen Gu personally confirmed that company revenue grew more than 3x over the past year.
The team size is even more striking. InsightFinder serves a group of Fortune 500 customers with fewer than 30 employees.
Interpretation: this means it is not scaling through headcount-heavy project outsourcing. It is scaling through a productized, high-gross-margin model.
2. What Pain Point Did It Hit?
To understand InsightFinder’s value, first understand a core contradiction in enterprise AI:
AI models are becoming more powerful, but enterprises have less control over AI systems.
A typical scenario: a credit-card company’s anti-fraud model suddenly behaves abnormally. The data scientists say the model is fine. The SREs say the infrastructure is fine. After several days of back-and-forth, the team finally discovers that cache expiration on certain server nodes caused the problem.
InsightFinder compresses that days-long investigation into a minutes-level loop handled by the platform.
Under the hood, it is not simply placing a large language model on top of machine logs. According to TechCrunch, it uses an engine called “Composite AI,” layering unsupervised machine learning, predictive AI, causal inference, and large language models while monitoring three layers at once: data, models, and infrastructure.
As Helen Gu put it: this is not something you can do by slapping a foundation model onto machine logs.
3. What Key Moves Did It Get Right?
There are several moments in InsightFinder’s story that builders should notice.
First, it did not chase every technology wave. It waited for the wave to come to it.
The company was founded in 2016 and has long worked on intelligent monitoring for IT infrastructure. The large-model explosion happened in 2022-2023, but InsightFinder’s revenue breakout came in 2024-2025, when enterprises finally began putting AI models into production at scale and observability became a must-have.
Interpretation: this is not luck. It is positioning. InsightFinder is not an “AI tool.” It is underlying infrastructure for AI systems. Tools get replaced. Infrastructure gets depended on.
Second, it turned academic accumulation into a product moat.
Helen Gu is a CS professor at North Carolina State University with 15 years of academic accumulation. InsightFinder’s core engine is not a call to the OpenAI API. It is a self-developed Composite AI system. That means competitors, whether Grafana, Datadog, or new AI monitoring startups, will struggle to copy its capability quickly.
Third, it proved that B2B AI is not about headcount. It is about density of domain knowledge.
Serving Fortune 500 customers with fewer than 30 people would have been unthinkable in the old software-outsourcing era. InsightFinder can do it because its product is standardized enough that most customer needs can be met through configuration rather than custom development.
4. What Can Others Copy? What Cannot Be Copied?
Moves others can copy:
-
Choose a vertical that is narrow enough and deep enough. InsightFinder did not try to build “general AI monitoring.” It focused on the intersection of AI systems and IT infrastructure. The cut is narrow enough that giants may not bother to specialize, but deep enough that beginners cannot master it.
-
Make the product a standard interface in the customer’s workflow. Do not sell a “better tool.” Make customers feel that without your product, the workflow cannot run. InsightFinder’s deep integrations with enterprise systems such as ERP and TMS help create this interface-level dependency.
-
Accumulate domain knowledge before the technology matures, then productize quickly once it does. InsightFinder waited eight years for the AI agent wave. Builders can ask themselves: in my vertical, what problems can today’s AI technology finally solve?
Advantages others cannot easily copy:
-
Fifteen years of academic research plus ten years of enterprise refinement. This cannot be bought with funding or hiring. InsightFinder’s moat is domain-knowledge density, and that density takes compound time.
-
Helen Gu’s academic reputation and IBM/Google credentials. In enterprise markets, founder background is a key chip for earning early large-customer trust. That is personal history and cannot be replicated.
-
The precise timing window for AI agent deployment. 2024-2025 was exactly the inflection point when enterprises moved from AI experiments to AI production. Two years earlier, the market was not ready. Two years later, competition would be more intense.
5. Lessons for Builders
InsightFinder’s story tells us something important: the largest AI commercialization dividend may not be in the flashiest consumer categories, but in the boring-looking deep waters of B2B.
Its success formula can be simplified as:
Vertical depth x AI technology maturity x productization capability = commercialization breakout
For builders searching for direction, here are three self-check questions:
-
Do I understand a vertical domain better than 90% of people? If not, AI is only a toy. If yes, AI becomes leverage.
-
Has my product become a standard interface in the customer’s workflow? If it is only a “better tool,” your ceiling is tool-level premium. If it is an irreplaceable interface, your ceiling is the customer’s business scale.
-
Can I serve a Fortune 500 customer with fewer than 30 people? If the answer is no, your productization is not strong enough yet, and you are still trading custom development for revenue.
6. Two Other Dark Horses Worth Watching
Before closing, here are two candidate cases from different markets for cross-reference:
Gizmo (education AI)
A four-year-old learning platform that uses AI to turn notes into interactive study materials, plus gamification such as leaderboards, streaks, and daily lives. Users grew from 300,000 in 2023 to 13 million in 2025, across more than 120 countries. It recently raised a $22 million Series A. Sources: TechCrunch and PR Newswire.
Loop (supply-chain AI)
A supply-chain forecasting platform founded by former Uber executives. It uses AI to structure fragmented data, including PDFs, paper documents, and emails, then provides predictive analytics. In 2025, it raised a $95 million Series C led by Valor Equity Partners, with participation from Founders Fund, Index Ventures, and J.P. Morgan. Source: TechCrunch.
Final Note
The AI industry is living through a strange paradox: the louder the category, the harder it may be to make money; the quieter the category, the earlier moats may form.
InsightFinder’s story is not a grand narrative about “using AI to change the world.” It is a practical manual for building AI into a good business in places others do not see.
For builders, that may be more valuable than any financing headline.
All data in this article comes from public reports by TechCrunch and other third-party media. InsightFinder’s revenue growth was confirmed directly by the founder and has not been independently audited. Loop’s founding time and commercialization details remain unclear and are based on public reporting.