For the finale (well, it was after lunch) of two packed days at IIEX EU in Amsterdam, Colonel Mustard Lucy Davison brought together our expertise in insights communication with the hottest topic of the moment – AI. We teamed up with Hennieke Potman, Global Consumer and Market Insight Manager from Flora Food Group to share our research into a common question we hear when advocating for thoughtful, structured insights communication: “Why not just use AI?”
In true research fashion, we put AI to the test in a real-world experiment: could it help us communicate insights faster, better, cheaper, or more effectively? The answer: not yet. While tools like ChatGPT, Piktochart, and Synthesia showed promise, we learned that human input is still essential – for structure, storytelling, data safety – and sanity.
Our brief was to create materials that went beyond Powerpoint reports which could be shared in Flora and which should make a measurable impact.
Here’s what we found out…
Lesson 1: The short-list only gets shorter
We began by exploring the landscape of generative AI tools, testing a wide range from LLMs like ChatGPT, Gemini, and Copilot to tools like Claude and NotebookLM. We wanted to see if they could meet key criteria: data secure, low-cost, easy-to-use, time-saving, and genuinely useful. We also explored visual content generators like Custom GPTs, Vyond, Pictory, and our current favourite, Gamma, to see how they handled video and infographic production.
To ensure a fair comparison, we used the same input each time – a dense, 48-slide insights deck filled with data, charts, and infographics, on gender equality. We asked them to create videos, infographics and summaries – the materials for Flora to share.
The results? Very mixed. The tools varied widely in output, usefulness, and quality – and some offered more headaches than help.
Yes, generative AI can produce content based on your input – but that’s where the optimism ends. When we fed our insights deck into popular tools like ChatGPT, the results were underwhelming (and occasionally laughable).
Even custom GPTs designed specifically for infographic creation for example, could barely manage a bar chart. The outputs were incorrect, cluttered, or just plain bizarre – a classic case of hallucination.
And animated videos? Let’s just say the chaos on screen earned a solid laugh from the IIEX audience – but not for the right reasons.
Most of the tools we tried did not make it through the first round of testing.
Lesson 2: Trust still matters – read that data security clause!
Amongst the dozens of tools we tested, a few stood out for producing semi-decent outputs – notably Gamma, Piktochart, and Synthesia. These made our initial shortlist thanks to their ease of use and relatively polished results.
But when we looked closer at their data policies, things got murky. Scrutinising legal clauses and privacy terms quickly became a task in itself. And while many in the insights world love Gamma, its data protection terms are vague at best – offering no clear commitment that your input or output won’t be used to train their models. For this experiment, where we were working with confidential Flora research, that was a dealbreaker. Gamma had to go.
Lesson 3: It’ll take (your) time… A lot of it
AI tools love to promise the same things: faster, better, more efficient outcomes. And yes, in some cases, they delivered. For example, ChatGPT-4o (with its latest image generation update) helped us create custom visuals for our infographic. But other tools? Across the board, we found that the tools required a lot of human editing, creative judgement, and iteration – more than without using them.
Take Synthesia. It’s a platform that lets you create talking head videos using stock avatars – or even one in your own likeness, complete with your voice. Sounds futuristic, right? And on the surface, it is. The editing interface is user-friendly, and generating an avatar version of Hennieke was eerily realistic.
But things got manual, fast. We still had to write the storyboard, script the content, and manually adjust everything from tone of voice to timing and animation. After seven iterations and over five days of tweaking tone, timing, and visuals, we finally got a version we liked. Ironically, doing it the ‘old-fashioned’ way and filming Hennieke presenting, would have taken us just three. So, while Synthesia automated the avatar, the rest still relied heavily on human creative direction.
The same was true with Piktochart. It gave us a decent starting point for our infographic, based on a storyboard we fed in – but missed key insights, context, and tone. We had to fill in the gaps ourselves, then redesign fonts, colours, and layout elements to align with Flora’s brand. What began as a “smart” AI quickly became a clunky design tool. In the end, using good ol’ Adobe design platforms – with a real human at the helm – proved quicker and better for the job.
In total, the Flora and Mustard team spent over 60 hours producing all the content.
What does this mean for insights communications?
The bottom line? These tools not plug-and-play. AI can support the process – if you’re prepared to invest more time editing, refining, and steering output than you would without them. Above all they don’t replace the creative thinking and sharp eye that only humans bring.
What is important to know is that when we shared the final materials and measured impact at Flora, the content captured attention and sparked engagement across the board – proving again the value of thinking beyond the PowerPoint report and creating insight communication campaigns that resonate.
Final lesson: AI is promising but people make it meaningful
Of course, AI isn’t the problem – it’s the expectation that it can do it all. The creativity, context, and quality control still need to come from people – AI just gives us a new set of crayons to colour with.
Experimenting with these tools reminded us just how valuable we are.