Maisa AI Raises $25M to Tackle Enterprise AI’s 95% Failure Rate
▼ Summary
– 95% of generative AI pilots at companies are failing, but advanced organizations are turning to agentic AI systems that can learn and be supervised.
– Maisa AI has launched a model-agnostic platform called Maisa Studio, backed by a $25 million seed round, to deploy trainable digital workers using natural language.
– The startup differentiates itself by using AI to build the process (chain-of-work) rather than generating responses directly, aiming to reduce hallucinations and increase accountability.
– Maisa employs systems like HALP and KPU to enhance reliability and has attracted enterprise clients in banking, car manufacturing, and energy sectors.
– The company plans to expand its team and customer base, focusing on complex use cases that demand trustworthiness and auditability in AI automation.
A recent MIT NANDA initiative report reveals a startling reality: 95% of generative AI pilots within enterprises are failing to deliver expected results. Rather than abandoning artificial intelligence altogether, forward-thinking organizations are shifting toward agentic AI systems that emphasize learning capabilities and human supervision. This growing demand for reliable, accountable automation has created a significant market opportunity for specialized platforms designed to bridge the gap between AI potential and practical business application.
Maisa AI has entered this space with a clear vision. The year-old startup recently secured $25 million in seed funding, led by European venture capital firm Creandum, to support the launch of Maisa Studio. This model-agnostic, self-serve platform enables users to deploy and train digital workers using natural language commands. Unlike many contemporary tools that focus solely on generating outputs, Maisa’s approach centers on constructing what it terms a “chain-of-work”, a transparent, step-by-step process that ensures accountability and reduces unpredictability.
Founded by David Villalón and Manuel Romero, who previously collaborated at Spanish AI firm Clibrain, Maisa emerged from firsthand experience with the limitations of current AI systems. “We recognized that businesses could not fully rely on AI for critical operations without introducing serious risk,” Villalón explained. Their response was to develop a framework built around trust and precision, rather than pure automation speed.
Central to Maisa’s methodology is its proprietary HALP system, Human-Augmented LLM Processing. This interactive approach functions like a guided dialogue: digital workers outline each step of a process before execution, allowing users to validate and adjust in real time. Complementing this is the Knowledge Processing Unit (KPU), a deterministic engine designed to minimize AI hallucinations and enhance output reliability.
Though initially conceived as a technical solution, Maisa’s emphasis on verifiable results has attracted interest from sectors where accuracy is non-negotiable. Current production clients include a major banking institution, automotive manufacturers, and energy companies. These organizations are using Maisa to move beyond traditional robotic process automation, leveraging AI without being constrained by rigid rule sets or labor-intensive coding.
The platform offers flexible deployment options, including secure cloud and on-premise installations, catering to enterprises with stringent data governance requirements. This focus on complex, high-stakes applications distinguishes Maisa from freemium-oriented competitors targeting broader, less specialized audiences.
With dual headquarters in Valencia and San Francisco, Maisa is strategically positioned to serve global clients. The company’s pre-seed round, led by NFX and Village Global, already signaled strong U.S. interest. The latest funding round also attracted participation from Forgepoint Capital International, further underscoring the platform’s appeal within regulated industries.
Looking ahead, Maisa aims to nearly double its team size by early 2026, scaling operations to meet accelerating demand. “Our goal is to demonstrate that AI can indeed deliver on its promises, responsibly and reliably,” Villalón stated. As the company begins onboarding clients from its waiting list later this year, it seeks to establish a new standard for what enterprise AI can achieve when built around transparency and human oversight.
(Source: TechCrunch)
