Developing a serious game is a challenging multidisciplinary task. The risk of creating a game that does not meet the learning goals is considerable. Balancing gameplay and instructional design to make a fun and effective learning game requires multiple experts across various development phases. Academic literature acknowledges the need to involve different target groups but does not specify when the experts' scarce resources would be best utilized. The availability of subject matter experts is particularly challenging, and the timing of their involvement requires careful consideration. In this mixed-method research, a group of (n=32) key stakeholders – subject matter experts, target users, a pedagogue, and game developers tested a virtual reality (VR) fire extinguishing training application. The data for the study came from observing gameplay; collecting gameplay metrics; post-game questionnaires on user satisfaction, learning satisfaction and user experience; and focus group discussions. The results convey the topics that different stakeholder groups pay attention to. The material revealed that subject-matter experts have to be involved in the early stages of the design process. Involving these experts in the later development stages is no longer beneficial because the changes they typically propose would be too costly. Target users do not have the necessary ability to recognize or communicate fidelity flaws and teaching approaches. They cannot be used in the creation of learning goals as they do not know what they should master. This research concludes with a serious game development cycle model stipulating the effective and efficient involvement timing of the key stakeholders.