AI doesn’t fix broken workflows — it accelerates them. Publicis Sapient’s Chief Marketing and Communications Officer Teresa Barreira explains why getting AI-ready starts with fixing your processes, not just your data.
Company leaders have often referred to AI as an ‘amplifier’. If you feed AI models bad data, it will produce bad outcomes, and if your workflows are based on flawed or inefficient knowledge, AI can make those weak processes more evident, and simply speed them up.
Of course, marketers use AI in varied ways – including large language models (LLMs) like Claude and ChatGPT, automation functionality within a platform they use, company AI agents, and more. Across all of these tools, giving AI bad quality information, or using it as part of an already flawed workflow, doesn’t magically fix that issue in your organisation. Instead, it likely speeds it up, enabling you to do the same thing more, potentially exacerbating a once minor problem.
“AI accelerates what is already there,” explains Teresa Barreira, Chief Marketing & Communications Officer at Publicis Sapient, the consulting arm of marketing holding company Publicis Groupe, . “If you have efficient ways of working and good judgment, it can make them even more effective. But if you have broken workflows, too many approvals or poor decision-making, AI can simply help you do the wrong things faster.”
Most organisations rushed to implement AI tools in a bid to remain competitive, bolting them onto existing processes and workflows, which made it easy to overlook its ability to propagate flaws. However, getting your workflows, processes, and data in check provides a necessary blueprint to ensure successful AI implementation.
From Bad to Worse Insights
AI systems learn from and act on the information they are given – whether that is enterprise data, or customer data. By now, it’s no secret that having poorly organised or incorrect data only ever leads to poor results when using AI. This is because AI systems treat that information as absolute truths, and wouldn’t be able to differentiate between ‘good’ and ‘bad’ information.
As such, AI tools may confidently provide marketers with inaccurate information or predictions or biased suggestions.
Lost in Translation
When the underlying information or processes are flawed, the insights that AI gives marketers can act as an optical illusion – identifying irrelevant or useless patterns and insights. According to Barriera, this is where the importance of human judgement comes into play, which includes intention. This could be for defining an audience, applying creativity and understanding cultural context and nuance.
She continues: “The answer isn’t to have a human approve every single step. That just recreates the friction you’re trying to remove. You need to be deliberate about where human involvement really matters. AI can give you more information and more options. People still need to decide what matters.”
Getting Organised
So what does organising your data and reimagining your workflows actually mean in practice? For Publicis Sapient, this involved breaking down work into more than 1200 individual tasks and speeding up every one they could with AI. “We didn’t waste time agonising over which tasks to stop, start, or continue,” she says. “We brought our marketing and communications teams together…and opportunistically accelerated every task that we could.”
The next step for the company was then to reimagine its workflows. One campaign workflow that used to take 80 manual tasks, 50 handoffs and 20 days was cut down to 40 manual tasks, nine handoffs and three to five days.
For Barriera, the handoffs mattered more than the days saved. “The most interesting number was the reduction in handoffs. We removed 41 points where work could stop, wait or lose context.” This anecdote questions what ‘efficiency’ means – not simply completing work faster, but creating more cohesion and logic in workflows. And ultimately more opportunities to provide human context.
“This is about much more than cleaning up your data. You need to look at workflows, decision-making, approvals, roles and where human judgment is genuinely needed. Otherwise, you’re putting new technology on top of an old operating model.”
