AI was meant to boost profits but it costs about €35M for Enterprises. Why?
MIT found 95% of AI pilots fail to deliver business value.
The AI Gold Rush and the big Bubble
In 2025, every company – from Coca-Cola to NASA to your neighborhood bank - is running AI agents. These aren’t just chatbots. They book flights, approve loans, analyze medical scans, and even write code.
But there’s a problem.
AI often makes silly and costly mistakes. It can say 2+2=5 or tell a doctor a healthy patient is sick. Worse, these errors are invisible. AI doesn’t throw error messages like traditional software. It just hallucinates with confidence.
This is why 95% of corporate AI pilots fail.
Why 95% of Projects Fail
Silent failures: Unlike traditional software that crashes or throws errors, AI pretends it’s working. It lies with confidence, making mistakes invisible until damage is done.
Hallucinations: AI invents facts, and no one notices until it’s too late.
Compliance risk: EU AI Act fines run up to €35M or 7% of revenue.
Model decay: What worked in January quietly breaks by July.
Talent gap: 1.6M AI jobs are unfilled; enterprises don’t have enough engineers to monitor everything.
The result: wasted spend, reputational damage, and lost board confidence.
A recent MIT study found ~95% of GenAI pilots fail to deliver business value. At least $13.8B in AI spend is at risk.
It’s like buying fleets of self-driving cars and realizing 7 out of 10 crash within six months.
What Noveum is Building?
Noveum.ai is the control system that helps companies watch, test, and improve their AI agents so they don’t fail silently or cost millions.
A child becomes a good student by going through three steps:
First, a teacher watches how they solve problems.
Then, the teacher checks their homework and gives feedback.
Finally, a tutor or parent coaches them to fix mistakes and practice until they improve.
Exactly, Noveum.ai is like school for AI agents.
Noveum-trace (Watcher)
Developers plug Noveum’s SDK into their AI systems. It takes notes on everything the AI does, step by step.
👉 Without this, companies only see the final answer, not the messy steps in between.
NovaEval (Checker)
After the work is done, someone has to grade it.
Nova Eval Grades the AI’s work with a full report card (accuracy, safety, speed, rules). Enterprises can build test datasets from real-world usage. It shows exactly where and why an AI agent fails.
NovaPilot (Coach)
When mistakes show up, a tutor steps in to help the student improve.
For AI, the Coach(NovaPilot) suggests better prompts, new instructions, or small experiments to fix the problem.
Analyzes mistakes and teaches better prompts, running safe experiments until the agent improves. Helps the AI fix mistakes and get better over time.
👉 Without this, companies repeat the same problems forever, burning money and trust.
The Cycle:
Together, this creates a simple loop:
Watch → Check → Coach
Every time the loop runs, AI agents improve — just like kids moving from grade to grade, getting better each year.
Meet the Team Behind Noveum.ai
Shashank Agarwal (CEO): Built AI infrastructure at scale — AWS Sagemaker, API.market, Levity.ai. Deep expertise in observability systems with millions of API calls already under his belt.
Aditi Upadhyay (Co-founder, Growth): Product and GTM veteran. She was on the founding team of YC-backed Spenmo, led AI Studio at SambaNova, and scaled SaaS platforms globally.
Both founders carry YC-backed startup experience and know how to take products from zero to scale.
How They Make Money (Business Model)?
Please see the Pro-Zone to to check their business model and unit economy.
Their GTM:
Currently Noveum has 3 GTMs:
SEO (search-driven traffic)
Developer communities
Outreach & content
That led to 18% month-on-month growth without having high CAC.
Their competitors received millions in Funding from YCombinator. Will they fail?
On the surface, their product looks easy to copy. SDKs, metrics, dashboards, any well-funded startup can spin those up. Big players like Arize AI, Confident AI, and Braintrust already raised millions to do just that.
So why Noveum? And why now?
1. Already have the potential users in their bucket.
Before Noveum, the founders built API.market, a developer platform that:
Grew to 6,000+ organic users
Handled 4M+ monthly API calls
Achieved this with 100% organic growth
This is direct proof the team can build and scale infrastructure that developers actually adopt cheaply and effectively.
2. Closed-Loop System
Most of their competitors stop at one part of the loop like monitoring.
But Noveum goes further:
Observe with tracing SDKs
Evaluate with NovaEval’s 30+ metrics
Optimize with NovaPilot, the AI copilot that fixes problems automatically
This complete is much harder to replicate and delivers more business value.
3. Founders experience and low CAC.
Both of the founders have proven record of building infra developers actually adopt,
Without burning cash on marketing.
Anyone can copy the idea.
Few can:
- Build the closed-loop (observability, evals , optimization).
- Get developers organically to use their product with low CAC.
- Offer compliance-friendly deployment out of the box.
- Execute fast with proven infra founders in a market window that’s opening right now.
GVPs take
Noveum is solving the hardest problem in enterprise AI: Failure in AI.
Unlike traditional software, it doesn’t throw an error message when something breaks. It just hallucinates — creating wrong answers with confidence. Most enterprises never notice until money is lost or regulators come knocking.
The founders know this world inside-out. Both bring YC-backed startup experience, and together they built and scaled API.market (4M+ monthly calls, 6,000+ organic users). Shashank also led AI infra at AWS Sagemaker, while Aditi drove GTM at YC-backed Spenmo and led AI Studio at SambaNova.
Their product have massive moat above their competitors. Others stop at monitoring or evaluation. Noveum adds the missing third piece. (Monitoring, evaluation, tutoring)
The market is validating the need. YC itself has backed competitors, showing this is a growing, urgent, and expensive problem. But Noveum’s distinction is clear: multi-agent support, compliance-ready on-prem deployment, and a closed loop that makes AI continuously better.
Unit economy, traction records & full pitch deck download(Pro Zone)
Jay’s Note
Early conviction is already showing up in capital. Their first angels weren’t outsiders, they were ex-colleagues who saw Shashank execute firsthand and put in $250K of their own money.
