Tackling the Reality of 95% Failure in Enterprise Generative AI — Maisa AI Raises $25 Million to Implement Explainable Agents
A high failure rate of 95% has been reported for the adoption of generative AI in corporate settings. Amidst such difficult circumstances, Maisa AI, a startup aiming for leadership through highly reliable, "explainable," and "traceable" AI systems, is attracting attention. In this article, based on the company's latest funding and technical approach, we will provide a professional and easy-to-understand explanation of the challenges and future prospects of AI utilization in enterprises.
1. Background of Maisa AI's Emergence and Funding
1-1. The 95% Failure Rate in Enterprise AI Adoption
According to a report by MIT's NANDA, an astonishing 95% of generative AI pilots conducted by companies end in failure. In response to this situation, many advanced companies have begun to focus on building agentic AI that allows for both learning and monitoring, rather than just a "black box."
1-2. $25 Million USD in Seed Funding
Against this backdrop, Maisa AI has achieved notable results within one year of its founding. It successfully completed a $25 million seed round led by European VC Creandum, and simultaneously launched a new platform called "Maisa Studio" using those funds.
2. Technical Strategy: Giving Shape to "Explainable AI"
2-1. Criticism of the "Black Box": Introducing Chain-of-Work
Maisa's approach is distinct. While conventional AI tends to "generate answers," they advocate for the reverse concept of "building the process itself (chain-of-work) that leads to the answer." CEO David Villalón describes this point as follows:
"Instead of using AI to generate a response, we build the process that leads to the response—that is our 'chain-of-work'."
This method enables the visualization, verification, and tracking of processes, thereby boosting reliability for implementation in business environments.
2-2. HALP and KPU: Technical Foundations Supporting Reliability
HALP (Human-Augmented LLM Processing) is a system where the AI asks about user needs, and based on that, a digital worker presents the steps. It is an approach similar to a student explaining while writing on a blackboard.
KPU (Knowledge Processing Unit), as a deterministic system, plays a role in suppressing the "hallucination" risks inherent in generative AI. It is processed in a structure that can be logically traced, rather than just pasting results.
3. Practical Deployment and Competitive Advantage
3-1. Enterprise Use Cases and Deployment Forms
Maisa Studio is being adopted by major companies handling critical operations, such as banks, automakers, and energy companies. Unlike conventional rule-based RPA, its strength lies in the ability for non-technical users to train digital workers using natural language.
Furthermore, it supports deployment in data centers or on-premises, placing importance on security and regulatory compliance.
3-2. Positioning as "RPA 2.0" and Competitive Landscape
Maisa is viewed as a next-generation RPA ("RPA 2.0") that is flexible and process-oriented. Competitors it is compared to include CrewAI and other workflow-based AI tools, but Maisa aims for a positioning that reverses concerns about the lack of reliability and repairability that can result from "rapid deployment."
4. Global Strategy and Future Outlook
4-1. International Expansion Through Locations and Capital Structure
With headquarters in Valencia (Spain) and San Francisco (USA), the company is looking toward expansion in both Europe and the United States. Last year's pre-seed round raised $5 million from investors including NFX and Village Global (backed by Zuckerberg and others), and this seed round is a continuation of that momentum.
Furthermore, the participation of Forgepoint Capital International is seen as a strategic move to facilitate adoption in highly regulated industries such as finance and energy.
4-2. Organizational Expansion Plans and Growth Strategy
Currently a team of 35, Maisa plans to double its headcount to 65 by the first quarter of 2026. The company expects to begin addressing its waitlist by the end of the year, declaring its intent to demonstrate to the market that it is a company delivering the results expected.
Conclusion: The Challenge of Building Trustworthy AI Agents
In an era where generative AI is not trusted in business settings and 95% of projects fail, the challenge of introducing AI in a way that surpasses RPA by prioritizing the 'visualization of processes' and 'ensuring verifiability' is now accelerating. Maisa AI stands at the forefront of this movement, leveraging its funding, technology, and organization to pave the way for a future of 'explainable and trustworthy AI'.