AI implementation is forcing marketing teams to rethink established ways of working. As organisations move from pilots to production, agile methodologies and continuous human oversight are emerging as the most effective frameworks for delivering lasting value.
Work methodologies have been used for decades to reinvent and improve ways of working – guiding teams to organise, manage, and deliver projects effectively and successfully.
Among the most widely used are agile and scrum, which favour short work cycles and continuous feedback; waterfall, which follows a strict, sequential path to the end result; and lean, which focuses on cutting waste and maximising value.
Marketing teams are still testing AI. Even though the narrative for some time has been “no one can claim to be an expert,” the tide is turning, as large companies begin to absorb the lessons learned from their early pilots.
What makes AI different is its versatility – it’s a technology that can be applied in countless ways, with no single “right” method for implementing or using it. So the question becomes: which approaches are actually working best right now?
Iteration, Iteration, Iteration
A recent report from digital experience platform (DXP) Optimizely looked into why AI isn’t lightening the workload of marketers. It found that most marketers feel they are in ‘survival mode’ often reacting to outcomes instead of intentionally reflecting. More than half (54 percent) shift between campaigns without evaluating what’s been done well, and 31 percent say their admin has increased “dramatically” in the last three years.
Almost all companies are in a process of piloting AI tools, seeing what’s working and what isn’t, and changing their approach underpinned by a culture of adaptability.
Because of this, agile ways of working have often been seen as best suited to the testing and interation needed for AI implementation, as teams can continuously reflect on what’s working and what isn’t. This agility is why small and medium-sized companies have typically been able to move faster with their AI projects and scale more quickly compared to large enterprises.
“We’ve found that waterfall approaches rarely work well for AI projects because you’re often dealing with an unknown outcome from the start,” says Jonathan Healey, Group Technology Director at digital marketing agency IDHL. “We favour an agile, human-in-the-loop approach that lets teams test quickly, learn from usage and refine based on feedback. This methodology works because it combines experimentation and oversight.
According to Healey, one of the biggest lessons from working with AI is that the most valuable use cases aren’t always identified on day one. He says that some of IDHL’s strongest results have come from running small pilots. He continues: “There’s a lot of focus on AI pilots that ‘fail’, but many are simply part of the adoption process, helping organisations understand where AI can create value and where it can’t.”
Making Space for the Human-in-the-Loop
The phrase ‘human-in-the-loop’ is commonplace in most AI conversations today, referring to AI-human collaboration and people working together with AI systems. Despite its common use – particularly since the dawn of agentic AI – this method can be traced as far back as the 1940s and the Cold War.
“We use a combination of agile and waterfall methodologies, always with a human-in-the-loop,” says Luke Budka, AI Director at B2B digital agency Definition. “The mixed approach is partly driven by cost and capability differences across AI models.”
Because AI hallucinates and makes mistakes, the human-in-the-loop is necessary. This becomes increasingly important when outsourcing work to AI agents, who can reason and act autonomously. That’s why this method is often used in conjunction with iterative ways of working.
Budka explains that from an agile perspective, for example, his team will run two-week sprints testing AI-generated copy variations, reviewing performance data and iterating quickly, while for more complex projects they use sequential phases with sign-off at each stage. “These often begin with a proof of concept,” he adds, “increasingly popular because capability is accelerating so fast that clients need help understanding what’s genuinely possible.”
While no single methodology guarantees AI success, the evidence increasingly points towards a combination of experimentation and oversight – with agile often providing the strongest foundation. Its emphasis on continuous experimentation, rapid feedback, and iteration aligns naturally with a technology that is evolving almost daily. Combined with human oversight where it matters most, agile gives organisations the flexibility to turn AI pilots into lasting competitive advantage.
