Beyond the AI bubble: Why marcomms teams need to take a step back to speed up
What began as experimentation—testing a few tools, experimenting with prompts—has rapidly escalated into an industry-wide expectation that every team should have an “AI strategy”. The pressure to adopt the latest platforms, plugins, and features ramped up has widened the gap between hype and reality. Unsurprisingly, adoption has become messy, uneven, and often confusing.
We’re constantly experimenting with how new technologies fit into real workflows. AI is no different. What we’ve learned is that we need to take a reality check. That means taking a pause, moving away from chasing the hype, and striving for meaningful integration that enhances—not erodes—human value.
The messy middle: What is actually happening within teams?
AI itself isn’t new. Most of us have used it quietly for years through predictive text, recommendation engines, or automation built into everyday apps. However, generative AI changed the mood. Suddenly, AI wasn’t just something running in the background; it was a creative force producing ideas, content, and visuals in seconds.
This pushed teams into reactive mode. Many were asked to outline an “AI strategy” before they had the basics in place: clear workflows, governance, data standards, risk frameworks, or even defined use cases.
As a result, many organisations find themselves in a “messy middle”—a tangle of disconnected pilots, random tools, no shared standards, no ownership, and overall dysfunction.
A study conducted by MIT Media Lab shows the consequence. 95% of AI pilots deliver zero measurable ROI, and 50-70% of investment goes into sales and marketing pilots, the areas with the least return. AI is often placed on top of broken systems, accelerating the wrong things faster.
Though it might be tempting to misinterpret this chaos as failure, it isn’t. It’s simply where organisations land when a new, powerful technology arrives faster than teams can absorb it. But it is a signal: teams need to slow down long enough to repair foundations before layering on more tools.
What AI is (and isn’t) good at
Despite the hype, AI is best used as an enhancer, not as a replacement. Human strengths still set the direction: judgment, taste, empathy, contextual intelligence, relationship-building, and the ability to read nuance. These are the levers of strategy, narrative, creativity, and trust.
AI excels in different ways: speed, pattern recognition, summarisation, recall of vast amounts of information, and the ability to analyse or translate in seconds.
The real opportunity lies in the space where human intelligence and artificial intelligence overlap.
This is where redesigned processes allow both sides to do what they do best. Getting to that middle requires rethinking how things are done. AI is neither a magic shortcut nor a tacked-on step at the end of a workflow. It becomes part of the system – but the system still needs human direction, clarity and structure to work.
A more grounded path forward
One good example of how designing a workflow with AI as a step, not a bolt-on, drives real value comes from a localisation project we completed with a global fintech company. By rebuilding the process so humans shaped the logic, structure, and narrative upfront, AI could then take the heavy lifting of scaling outputs across markets. The result was faster, more accurate localisation across markets, with human oversight still essential for refining the logic and spot-checking accuracy. It worked because the process was rebuilt so each side could do what it does best.
We’ve also seen AI level the creative playing field for organisations with modest budgets. With the right direction and guardrails, teams produced high-quality creative concepts that would typically be out of reach. AI didn’t replace creative thinking; it expanded the terrain. Human direction remained central, while AI enabled speed, quality, and experimentation.
Success comes not from early adoption or enthusiasm, but from intentional choices. AI only works if the foundations are right.
Four practical shifts to move from experimentation to impact
1. Start with one real problem.
Clearly define a pain point. It could be a labour-intensive localisation process or the challenge of producing competitive creative on a tight budget. If AI cannot solve a real business problem, don’t deploy it for the sake of it.
2. Design roles, workflow, and measures.
Success never comes from bolting AI onto an existing process. It requires rethinking who owns what, where AI sits in the workflow, and how success is measured.
3. Set guardrails for data and ethics.
Human oversight is a non-negotiable. Guardrails ensure tools don’t introduce errors, bias, or copyright risks. Humans should validate accuracy, protect confidentiality, and refine narrative choices.
4. Build a simple learning loop.
AI systems improve through iteration. The process isn’t “set and forget”; it’s test, refine, repeat. The right loop turns pilots into scalable performance.
Together, these shifts introduce a more grounded path forward—one rooted not in chasing the next shiny new tool but in building underlying structures needed for AI to deliver consistent, scalable value.
The bubble will burst, but value will remain
The AI hype cycle won’t last forever. Tools will consolidate. Noise will quieten. But the meaningful value – the kind that comes from intentional integration, clear foundations and human-led direction – will remain.
The teams that thrive won’t be the ones chasing every new feature or tool. They’ll be the ones who invest in the messy, unglamorous work of redesigning their systems, workflows, and roles so that AI amplifies the work rather than overwhelms it.
The future belongs to teams who can operate in the overlap: where generative intelligence meets human clarity, and where hype gives way to real, measurable impact.
