SINGAPORE, Sept. 8, 2026 — Appier, an AI-native company delivering Agentic AI as a Service (AaaS), says its AI Research team has published two papers focused on making enterprise AI more reliable and more globally deployable.
As Agentic AI becomes more deeply embedded in core enterprise operations, Appier said the research examines two linked questions: how large language models can recognize when retrieved information does not support a valid answer, and how choosing the appropriate reasoning language can better serve users across linguistic backgrounds.
The company framed the work as part of a broader effort to advance AI innovation while expanding the frontiers of marketing and advertising technology. In its announcement, Appier said the research sets a more rigorous benchmark for the trustworthiness and global deployment of enterprise AI.
One of the papers, titled "None of the Above, Less of the Right: Parallel Patterns between Humans and LLMs on Multi-Choice Questions Answering," addresses a basic but important problem for Agentic AI: what happens when a model cannot find enough support to answer a question.
To complete tasks autonomously, Agentic AI typically retrieves information from enterprise knowledge bases, documents, or external sources before reasoning and responding. Retrieval-Augmented Generation, or RAG, is a common architecture for that process. But when retrieval fails to provide sufficient information, Appier said the model’s ability to recognize that gap directly affects the reliability of any later decision.
To study that challenge, Appier’s researchers used “None of the Above” options to simulate situations in which no valid answer was available. The team tested 28 leading large language models of varying sizes and found that model accuracy fell by 30% to 50% when “none of the above” was the correct response.
According to Appier, the results suggest that even when models possess relevant knowledge, they often choose suboptimal or incorrect options instead of proactively flagging insufficient information. The company said that ability to “know what you don’t know” is especially important in tasks such as business ethics, where models must assess several plausible options holistically rather than solve a problem with a directly verifiable answer.
Appier said it applied two training methods — Supervised Fine-Tuning, or SFT, and Direct Preference Optimization, or DPO — to teach models to recognize “none of the above” scenarios. SFT trains models using correct examples, while DPO exposes them to both correct and incorrect responses so they can learn to distinguish between them.
The company said DPO improved model accuracy in identifying questions with no correct answer by nearly 30 percentage points. Appier described that as evidence that targeted training can strengthen a model’s ability to withhold an answer when appropriate.
At the same time, Appier noted that “none of the above” is not suitable for every question. The company said it is most effective when answers are clearly defined and the options are mutually independent, implying that an AI system’s ability to hold back an answer should be trained for specific tasks rather than treated as a universal behavior.
Appier said that in enterprise knowledge scenarios, this means improving retrieval accuracy while also training models to recognize when information is incomplete. The result, the company suggested, is a system that is less likely to guess when the evidence does not support a confident response.
The second paper in the company’s announcement focuses on reasoning language. Appier said selecting the appropriate reasoning language can better serve users across linguistic backgrounds, though the announcement did not provide additional technical details in the portion made available.
Taken together, the two papers point to a common theme: enterprise AI should not only search and answer, but also evaluate whether it has enough evidence to answer at all. Appier said the research aims to help large language models respond more reliably in situations where uncertainty, incomplete retrieval, or cross-language use could otherwise reduce trust.
Appier’s announcement positions those goals within its broader business as an AI-native company offering Agentic AI as a Service. The company said it continues to advance AI innovation and research while expanding its marketing and advertising technology offerings.
For enterprises adopting AI agents in core operations, the central takeaway from Appier’s research is straightforward: better systems may depend not just on stronger answers, but on better judgment about when an answer is not supported, and on reasoning approaches that fit the user’s language context.
By testing whether models can recognize information gaps and by training them to avoid forced answers when evidence is missing, Appier said its researchers are trying to make enterprise AI more dependable in practice. That, the company said, is part of building AI that can recognize its limits and choose a more appropriate reasoning approach.