Maisa AI Gets $25 Million to Fix Enterprise AI's 95 Percent Failure Rate
Maisa AI, a startup dedicated to improving enterprise automation through accountable AI agents, has secured $25 million in a seed funding round led by the European venture capital firm Creandum, aims to address the staggering 95 percent failure rate of generative AI projects.
Founded a year ago, the company seeks to address the staggering 95 percent failure rate of generative AI projects within enterprises by introducing Maisa Studio, a model-agnostic platform that facilitates the development of digital workers that can be trained using natural language.
A recent report from MIT’s NANDA initiative reveals that an astonishing 95 percent of generative AI pilots within companies are unsuccessful. However, instead of abandoning the technology entirely, leading organizations are experimenting with agentic AI systems that are capable of learning and being supervised.
With a new $25 million seed funding round spearheaded by European VC firm Creandum, it has introduced Maisa Studio, a self-serve platform that is model-agnostic and allows users to deploy trainable digital workers utilizing natural language.
Also Read: The Way We Use iPhones Could Change Due to EU's Pressure
The key architect behind this initiative is Manuel Romero, co-founder and Chief Scientific Officer of Maisa, who previously collaborated with Villalón at the Spanish AI startup Clibrain. In 2024, the two partnered to develop a solution addressing hallucinations, after realizing that “you could not rely on AI,” as Villalón stated.
The duo is not dismissive of AI; however, they believe it is impractical for people to evaluate “three months of work done in five minutes.” To tackle this challenge, Maisa utilizes a system known as HALP, which stands for Human-Augmented LLM Processing.
This tailored approach functions like students at a blackboard — it inquires about user needs while the digital workers map out each step they will undertake.
Additionally, the startup has created the Knowledge Processing Unit (KPU), a deterministic system aimed at minimizing hallucinations. Although Maisa commenced with this technical issue rather than a specific use case, it quickly discovered that its emphasis on trustworthiness and accountability appealed to companies wanting to implement AI for vital tasks. For example, its current production clients include a large bank and companies in the automotive and energy industries.
By catering to these enterprise clients, Maisa aims to establish itself as a more sophisticated form of robotic process automation (RPA) that facilitates productivity improvements without necessitating reliance on inflexible pre-defined rules or extensive manual coding. To cater to their requirements, the startup offers either deployment in its secure cloud or through on-premise solutions.
Also Read: Dolce & Gabbana: A Story through the Tests of Time
This enterprise-focused strategy means that Maisa’s customer base remains quite small compared to the millions attracted to freemium, vibe-coding platforms. Nevertheless, as these platforms seek ways to attract enterprise customers, Maisa is pursuing the opposite path with Maisa Studio, which is intended to expand its customer pipeline and facilitate adoption.
The startup also intends to grow alongside existing clients that have operations in several countries. With dual headquarters located in Valencia and San Francisco, Maisa already has a presence in the US, as shown by its cap table; its pre-seed funding round of $5 million last December was led by San Francisco-based venture firms NFX and Village Global.