Look, let's be honest: keeping a game alive long-term requires a deep dive into user feedback. But manually sifting through mountains of player data is an Herculean task that would break anyone. Recently, many have turned to AI for efficiency, but without the right tools and a solid strategy, you're just inviting a catastrophe of wasted time and effort.

So, how do we actually pull off effective user research? We got some answers at the 16th NDC 26 (NEXON Developers Conference). Developer Lee Se-wang from the NEXUS Co., Ltd. Game UX analysis team took the stage to share how they are using their proprietary 'Insight Finder' to revolutionise the process.

The old way of doing things was a slog, dragging on from survey collection to final reporting. Statistical verification always required a specialist's touch, causing constant bottlenecks. It took two to three weeks to analyse just one survey, and the findings often ended up buried in some isolated document, never to be seen again. This led to the absurdity of different departments accidentally running the same research, a classic case of corporate inefficiency.

To break this cycle, they looked into general-purpose AI, but the risks of errors and hallucinations were too high for comfort. Instead, the team introduced 'Insight Finder' and established a strict division of labour between humans, AI, and the system. They offloaded the heavy lifting—statistical testing and data cleaning—to a deterministic system pipeline, effectively killing any chance of error. AI was then restricted to what it does best: grasping survey intent and translating complex statistical figures into plain language.

Human intervention is now kept to a minimum, reserved only for the final stages to avoid any bottlenecks. This lets researchers focus on what actually matters—picking the right targets and drawing final insights from the AI-processed middle ground.

Developer Lee Se-wang noted that this process achieved three key goals: standardisation, assetisation, and accessibility. Once survey data hits the system, it cleans the info using consistent criteria and feeds only the distilled statistical facts to the AI. Unlike the old days of scattered, fragmented files, all info is now linked and stored as a valuable asset. This means planners, directors, and PMs can now query the data in natural language without needing a specialist to hold their hand.

The results? A massive boost in efficiency for a team that was chronically short on time. Now, they can generate an AI report in under 30 minutes with a single click. Cross-analysis between different surveys and large-scale trend spotting is finally possible. Without any internal marketing, over 100 staff members jumped on board in just four months. Average cross-analysis per survey shot up from 3 or 4 to 19, and they've estimated a staggering 9,339 hours saved in the last four months alone.