Extra fingers, broken workflows and ‘too good’ work: ad execs’ AI teething problems

Luke Bristow, chief executive officer, MNC: “We’ve thrown a lot of AI tools at our workflows as part of an intensive experimentation and road-mapping exercise. Microsoft Co-Pilot, Gamma AI, Canva, ChatGPT, Grok, Claude, Fathom, you name it. But there’s one thing AI still can’t do: make a pitch deck that doesn’t read like it was built by an 11-year-old who just discovered WordArt. Seriously. Every tool I’ve tried turns out something that looks like a school project. Whoever cracks that will be a billionaire. Until then, we’ll stick to making decks the old-fashioned way: last-minute panic and far too many Google Slides comments.”

Philipp Langer, principal creative, art & AI, Jung von Matt Spree: “In the early days of gen AI (so last year!), we hit a fundamental snag: the work was too good. Clients saw AI-generated visuals – tailored, high-impact and looking like a $250,000 David LaChapelle shoot – and naturally wanted exactly that. Back then, images weren’t high-res enough for production, legal uncertainties loomed and executing such a vision still required a six-figure budget. Clients rightfully asked: why present something that can’t be realized? Fair point. The fix? Backtracking to rough sketches and AI-assisted mood boards. Basically, what we did before. Today, resolution and detail are no longer issues, but the legal landscape remains murky. Will we still go all-in on costly productions, or is AI the future? Maybe. AI isn’t a magic wand that turns creative laziness into brilliance. ‘Shit in, shit out’ has never been truer.”

Rob Leeks, creative director, Imagination: “I’ve been using AI for about four years now. In that time its imperfections have shifted dramatically. In the early days, there were those extra digits on hands and all that wonderfully nonsensical prose. But now tools like Sora achieve convincing realism as default, so you have to be really vigilant with results: mistakes are masked by the sheer beauty of the content. For example, we created a video of a group of Gen Z friends dining alfresco on Italian food in Tuscany. It generated a beautiful, highly believable scene with lovely lighting and cinematography. It wasn’t until the fourth watch that I spotted one of the lead characters’ necks turning round the wrong way and snapping, The Exorcist-style. It didn’t make the cut.”

Adrienne Morgan, president & head of performance marketing, Stripe Theory:“A major AI teething problem is the time-intensive onboarding process. While AI promises efficiency, integrating it requires extensive human oversight – training it on refining outputs and ensuring alignment with objectives. It’s not plug-and-play. Another challenge is quality assurance. We can’t fully trust AI-generated data without verifying its integrity. For example, an AI-powered analytics tool once delivered completely misleading insights because it was working with outdated or biased data. It took extensive human intervention to catch the errors and correct them. AI is a partner, not a replacement. Until it can consistently separate signal from noise, human oversight is essential to make sure AI is adding real strategic value, not just more work.”

Camm Rowland, chief creative officer, Kepler: “For many creative leaders, including me, the AI conversation has shifted from primarily showcasing its potential to navigating the complex legal negotiations needed to implement AI solutions for clients. This critical phase introduces new risks in IP, likeness rights, privacy, and ethics. Some companies are charging ahead without any legal scrutiny. That’s a potentially catastrophic approach. Others remain paralyzed for various reasons, with equally severe consequences. The wise are finding balance and unlocking AI with both diligence and urgency. Because, even if minimal, AI-generated content carries inherent risks that need to be managed. But those checks and balances are worth the effort. There’s no one-size-fits-all solution or standard practice today that makes this process easy between brands and agencies. Legal departments across the industry are encountering these AI issues for the first time. So dig in, become best friends with your legal team and get ready for redlining.”

Kate Ross, co-founder & group MD, Eight&Four Group: “We’ve been on the AI journey for some time. In the early days, initial excitement quickly gave way to disillusionment as expectations clashed with reality. Some team members were perhaps slower to embrace change, while others saw AI as an existential threat rather than an opportunity. Tight deadlines led to creatives reverting to old methods rather than investing time in learning new tools. Adding to the frustration, nearly all the AI tools we tested weren’t fit for purpose. As we started building our own to combat this, we started to see far more interest. The real breakthrough was showing tangible results generated in our clients’ brand worlds. Once our generations were consistent, specific, and secure enough to be commercially viable, we saw a huge uplift. Our AI usage increased, as did our trust in our own tools and outputs.”

Mike Stefaniak, chief strategy officer, Hanson Dodge: “Sure, we’d love for our robot overlords to get us closer to the finish line on every query, with minimal need for human intervention. But we’re just not there yet. The AI tools we’re using routinely return extraneous, outdated or even dubious information. It takes significant effort to become a proficient user. Higher-order thinking (like the ability to infer an intended position in a competitive landscape audit) has yet to arrive. Ample human touch is still needed to get to usable outputs and defensible POVs. That’s to be expected this early in the game.”

David Kaganovsky, global chief technology officer, Landor: “One of the biggest challenges with AI is how quickly it evolves. It takes time to build AI-driven solutions that can quickly become redundant. We develop complex systems that chain multiple AI processes together, only to find that a new model update can now do the same thing in a single step. It’s frustrating, but it also forces us to think differently. Instead of focusing on just what AI can do today, we design solutions that are adaptable, knowing that the landscape will keep shifting. So, build for flexibility, not just functionality. AI is moving fast, and staying ahead means being comfortable with change.”

Alexandre Hoffmann, managing partner and growth director, Passion Digital: “Adopting LLMs into our workflows has brought undeniable efficiencies. I can’t see a world without it and it’s embedded in my day-to-day. Still, we’ve seen inconsistent output quality across teams without a clear framework guiding how we use AI in the broader business. Different tools and training levels give different results. Some team members lean on AI-powered platforms that deliver polished, nuanced content, while others get more generic, less refined responses. This creates a frustrating lack of consistency, making quality control more challenging and adding unexpected rounds of revisions. We moved quickly as an agency but underestimated the need for structured implementation, guidelines, best practices, and ongoing oversight. AI is an incredible tool, but like any tool, it’s only as good as how we use and train it.”

David DiCamillo, chief technology officer, Code and Theory: “We leverage a lot of Airtable’s AI tools for technical requirements and testing criteria. One team looked to have the AI write new criteria every time we changed a requirement. I quickly got a call from our Airtable team: ‘Hello, did you know that you used a year’s worth of AI credits in one week?’ Well, no, we didn’t, thanks for telling us. How much will this cost? Fortunately, our friends at Airtable appreciated our excitement and provided some extra credits with a friendly warning to ‘keep it cool next time.’”

Mark Sturino, vice-president, data & analytics, Good Apple: “The promise of AI-powered data visualization had us hoping for instant insights. But we quickly discovered the gap between algorithmic output and meaningful business storytelling. While we have tried using various platforms, and some have even been successful in plotting out the data, we’ve spent a lot of time only to find that the combined eyes of the analyst and media planner are still required to make the data speak to the intended audience. For instance, when presenting sales-trend data, AI could generate the chart but couldn’t identify which metrics deserved highlighting or when a different chart type would better reveal the pattern clients needed to see. For now, the human touch remains indispensable in transforming AI-generated charts into compelling narratives that resonate with clients.”

Todd Sussman, chief strategy officer, FCB New York: “For the most part, we’ve been good at growing with AI and having the right expectations but, like with any emerging tech, there are moments where it reveals itself as more of an intern than an expert. Sometimes, it’s simply too eager to please. I’ve seen many examples where AI has told someone that they had a ‘brilliant way to frame that solution’ or a ‘great and disruptive idea,’ which in reality were really boring and vanilla. Perhaps a more pressing issue is when AI is simply wrong, making up quotes and stats or relying on outdated clichés. With that comes my favorite AI quirk: when it tries to wiggle its way out. ‘Hmm, looks like you’re right. I was just trying to give an example…’ AI is just starting to get its adult teeth. The more we use it, the more it will learn, so it will be eating solid food in no time.”

Andy Rowe, chief marketing sciences officer, RAPP: “AI literacy is a problem. There are two parts to this. The first is ignorance: seeing people pull the covers over their heads to avoid its inevitable impact, blaming AI’s complicated nature (and their complicated business) for their inaction. The second is people who are aware but are not aware enough, aimlessly following AI trends because they think they should rather than understanding it and thinking of its use cases strategically. While AI is great at ingesting and understanding huge amounts of data at far greater speeds and nuances than humans can, the unpredictability of its outputs is a blocker for huge-scale adoption. Brands mustn’t let this deter them from preparing but do so at their own peril. The need for human interaction on the journey and in all use cases is essential. Controlling hallucinations and quality output is something still not yet mastered.”

Jeff Blackman, managing director, Barbarian: “The core principles of innovation are being tested in this new era. The paradox in AI is that you must move fast to compete, but the faster you go, the greater the risk of outpacing the tech. You can quickly find yourself with a poor product or experience because the tech doesn’t meet the vision. When building prototypes for EGOs, our dynamic persona platform, we learned that AI-driven products don’t follow traditional development processes. To scale effectively, the AI stack must be the foundation, with the entire experience shaped around it. With our AI Accelerator, an initiative to inject AI tools into workflows, we established governance, security, and policy to support experimentation. The shift from a practical, safe mindset to exploring all possibilities was tough but necessary for adoption. Learning by doing (and sometimes failing) in AI is difficult but essential to stay ahead of the change.”

Kate Tancred, CEO, Untold Fable: “While AI can be a powerful tool in the creative process, it’s not without its limitations. No matter how refined your prompts or how much time you invest, you’re ultimately at the mercy of the AI’s interpretation. The challenge is that AI doesn’t just illustrate your vision – it fills in creative gaps, sometimes in ways you didn’t intend. For our clients reviewing this work, it’s easy to get swayed by or distracted by those AI-generated details, with the potential to perceive those would carry through to final delivery, even when they weren’t part of the original concept or budget. Without the tools to finetune, image generation like this will always be limited in its use to pre-vis.”

Silvia Sparry, global chief transformation officer, GroupM Nexus: “A key challenge with generative AI is hallucination: gen AI models sometimes identify patterns and generate responses that sound plausible but aren’t factually accurate. Just as we might see shapes in the clouds – our brains are wired to find patterns. Gen AI is tuned to spot patterns in text character strings, and can find these patterns even where they don’t represent facts. To manage this, we’ve implemented strong guidelines validating and managing gen AI accuracy. We’ve also incorporated human oversight while new tools are introduced to continuously improve output quality. For instance, we’ve built a best-practice AI agent that offers instant insights. When the chatbot encounters a query it can’t accurately resolve; we added strong instructions to the prompt that ensures that we direct users to one of our Practice Leads for guidance.”

Ben Rooke, digital business director, Mediaplus UK: “Often, AI’s biggest hurdle isn’t the technology itself; it’s the people using it. Securing employee buy-in is crucial for successful adoption. People need the skills to maximize its potential and the motivation to integrate it into their workflow. They must see how AI can benefit them, and have the right support in navigating its ethical implications. We provide continuous training, including tailored workshops, to build confidence and competence and foster the skillset and mindset needed to make it a seamless part of their work. One example of this is prompting. Poor prompting leads to poor output. We’ve all experienced the frustration of asking a model for creative work only to realize we didn’t give it the right brief. That’s why we prioritize training on how to prompt effectively and develop frameworks to help the team craft better questions so they can get the answers they need.”

Sean Betts, chief AI & innovation officer, OMG UK: “The pace of change driven by AI has caused both incredible excitement and a fair amount of skepticism. A significant contributing factor to these varying responses is the differing levels of knowledge and confidence that people have in AI. That’s why our focus has been on education. In 2023, we ran a generative AI webinar series, and we’ve recently launched an AI Readiness Program for all our employees. The program is designed to create a common foundation of understanding, covering everything from the fundamentals to legal and ethical considerations, with specific usage policies and guidance on using AI. Responsible innovation is about ensuring we’re moving forward together with the right knowledge and common practices in place. Generative AI technology is amazing and has huge potential, but only when it enhances (rather than replacing) human creativity and strategic thinking.”

Henry Skelsey, vice-president, marketing, Songtradr: “We’ve tried various AI models for baseline market research, but they have a funny habit of misidentifying us. And when we’ve used LLMs to help shorten bios for things like conference submissions, the models seem incapable of staying within word or character count. It’s with these remedial tasks, which you hope AI would be able to take off your plate, that it seems to struggle with the most. A different complication with AI use is that people are becoming so trained to be on the lookout for AI-generated content that it’s become the new example of a spelling mistake in a presentation: grabbing the viewers’ attention and not letting go. We’ve had to update our design policies to be extremely limited in what AI imagery can be used in our decks, even for filler imagery.”

Peter McKenzie-Jones, technology & CX director, Grey London: “We’ve moved from a time of AI experimentation to adoption. With that comes familiar challenges, not just technical issues. One client wanted to use AI-generated imagery for commercial output, so we had to ensure the images were legally cleared for usage: relooking at the prompts and models and assessing how much human input had been added. Once we had all that information, we could then decide whether the images were OK to use. Working out the process around something seemingly simple like this takes time to get right. We need to guard against the mindset of ‘Can’t we just get AI to do that?’ The reality of using it may not be straightforward from a technical, commercial or ethical standpoint, especially with a technology moving as fast as AI.”

Mike Khouri, founder and CEO, Tactical: “As exciting as it’s been, our results with AI have been mixed so far, as they should be with any new technology. For every win, there is a setback – or, as we see it, a learning. Some tools and use cases have led to behavioral change across the agency, especially for designers and developers, which makes all our lives a little more efficient. But the flip side is also true. We’ve poured time and money into ideas that never made it past the testing phase, mostly because the quality didn’t hit internal benchmarks. And that’s OK. Sometimes, the tech isn’t ready; sometimes, the workflow changes are too extreme. But the important thing is to stay curious and keep playing the game. The point with any technology is to enable and simplify, not add complexity.”

Steph Hobart, managing director, Brothers & Sisters Sports Club and AI Studio: “Cracking brilliant ideas and executions in AI is about experimentation. It’s not an exact science. We’re used to 100% control; AI demands a shift of mindset. It pays to experiment using generative AI in concepting. But it’s critical to build testing into production, where you get roughly what you expect – but not exactly. New tools need new workflows. We discovered mid-animation the limits of how Runway could nail single frames. Post-production also needs the right blend of new and traditional tools. Retouching is especially complex: we’ve seen funky results tweaking lighting effects, then having an athlete’s face distorted in the process. And in quality control, hawk-eyed attention to detail is yet more important.”

Jocelyn Turlan, senior strategy director, Wolff Olins: “The biggest headache with AI in strategy work isn’t hallucination or visualization (those we can alleviate). It’s getting AI to stop being so AI-like. AI has crept up in nearly every corner of our craft. It writes for us, digs up info, and stress-tests our thinking. It’s blazingly fast, impressively scalable, and bends to our needs. Well, to a point. No matter how we prompt it, what data we feed it, what clever hacks we try, AI is still an averaging machine at heart. It lives in the land of ‘sounds about right’ and ‘that’s how it’s usually done.’ It doesn’t inspire or open new avenues. But strategy is about the moment when you leap from insight to idea. AI shows you what already exists, not what could be. Next time you’re on ChatGPT, ask if it can parallel think.”

Yang Wong, director of innovation (AI & creative technology), Iris: “GenAI is vast, hyped, real and evolving fast. The only way to grasp what’s real is to use it and then constantly update our thinking. Committing to any single tool long-term is a mistake, as the tech shifts weekly. But innovative behaviors, once established, endure. We’ve made mistakes (overcommitting, presenting results without context), but that’s how we learn. If it’s better in some way, we adopt and integrate. We’ve encountered naysayers, hype merchants, and resistance to change. Nothing cuts through this BS better than doing it live and being transparent. We’re already drowning in AI slop, but the real issue is sloppy thinking and assumptions. We’d rather forge ahead and build on what’s real.”

Kat Patterson, managing director, Art of the Possible: “I’m passionate about leaning into AI. But what about the teething problems, the glitches, the moments where the tech just didn’t deliver what we expected (or worse, what we promised clients)? Maybe it was an AI-generated image that looked incredible… until you spotted the extra fingers (or the haunting, slightly possessed-looking faces). Or an AI copy tool that nailed the grammar but stripped all the personality, turning a brand’s distinctive tone into a beige, lifeless blur. And then there’s the backlash. We’ve seen brands swap out real voices for AI narrators and get slammed for it. AI-generated ads that feel hollow. Chatbots that can’t quite grasp sarcasm, nuance, or basic human emotion. Instead of feeling cutting-edge, some executions have landed as cold, clunky, or just plain ‘off.’”