Why The Model Isn’t The Hard Part, The Workflow Is

Most companies chasing artificial intelligence are obsessing over the wrong thing. While leaders pour billions into training bigger, better models, the real bottleneck—and the biggest lever for success—is the workflow surrounding those models. A Forbes Tech Council article argues that 'the model isn't the hard part, the workflow is,' pointing to a fundamental shift in how organizations should approach AI deployment. The piece centers on a program that started with the team, not the tool, and saw dramatically different outcomes. This insight challenges the prevailing narrative that model performance alone drives value, highlighting instead the critical role of human processes, data pipelines, and cross-functional collaboration. Research from Gartner and McKinsey supports this view: up to 85% of AI projects fail to deliver expected value, not because the models underperform, but because organizations lack the operational workflows to integrate them effectively. The article underscores that companies often rush to build or buy a state-of-the-art large language model (LLM) while neglecting the infrastructure needed to clean data, manage prompts, handle feedback loops, and govern outputs. Without deliberate workflow design, even the most powerful model remains a proof of concept. The successful program referenced in the article prioritized assembling a multidisciplinary team—engineers, domain experts, product managers—before selecting a model. This approach allowed the team to define clear objectives, establish data hygiene standards, and create iterative testing procedures that turned a generic model into a tailored solution. The outcome was not just a deployed model but one that actually changed business metrics. Industry experts now argue that workflow design will become a core competency for enterprise AI. As organizations move from experimentation to production, the competitive advantage will come from how well they orchestrate people, processes, and technology—not from which model they choose. Looking ahead, we can expect more companies to invest in workflow automation tools, adopt MLOps platforms, and restructure teams around 'AI choreography' rather than model selection. The message is clear: in the race to leverage AI, the hard part isn't the algorithm—it's making that algorithm work within a messy, human-driven organization.