From chaos to competence: 4 lessons on AI adoption challenges

The reality is that AI is here to stay. But with mounting questions around risk, governance and making a return on investment (ROI), how can leaders move from chaos to competence with AI? And how can they overcome AI adoption challenges?

We spoke to Rudy Lai, CEO at Tactic, and Jason Smith, AI Strategy Lead at Publicis Media, to tackle these questions and to share guidance on how best to get started.

What’s their best advice for starting AI transformation?

“The technology is the least of your worries,” says Jason. “People and process are two of the most difficult things to get right.”

Jason recommends assessing the ‘day in the life’ of your workers to understand how generative AI (GenAI) can help, while also encouraging people to be hands-on with the technology. He adds:

“Everybody understands that AI is the next big thing, the next business opportunity, the next tool to create impact,” says Rudy, agreeing on the need to focus on people and processes.

He argues many businesses struggle to find the right place to start, and suggests a three-tiered approach when thinking about AI adoption:

As you move through each tier, you’ll shift from internally focused AI use cases to external ones. How far you’ve progressed depends on a range of factors, including your level of data skills maturity.

However, Rudy argues: “No matter how you slice and dice the use cases of AI in your organisation, you always need to go back to the business impact.”

Upskilling should be treated as a priority – giving your people the foundational data skills to unleash the potential of AI.

When asked which teams are most receptive to AI, Rudy notes that it depends on task suitability, digital maturity, and staff readiness:

“One of the first signs of a department being AI-ready is you can imagine putting the work they do into a large language model (LLM) and automating it.”

He adds that “well-defined problems” such as repeating tasks and processes are good candidates for automation. And not to forget: teams with an existing strong data culture will be better targets for initial AI adoption.

AI champions – people who are curious, proactive, and willing to experiment with new tools – is one way to encourage adoption.

So what else makes a great AI champion?

“[They have] the right mindset, a willingness to give it a go,” says Jason. AI optimism goes a long way – and these people may already be building rough and ready prototypes to show what can be achieved – even if these prototypes aren’t perfect.

After getting AI champions on a consistent skill level, they can act as mentors to others around them – as well as being focal points for spotting new use cases.

“Those use cases are when you can start to get some traction with the help of these champions, who can then hopefully bring other team members along,” he adds.

One initial step you can take to identify your AI champions is reviewing your skills inventory – helping you spot AI capability and strengths that may already exist in your workforce.

Significant risks are emerging from unmanaged AI use – also known as ‘shadow AI’.

Without proper AI tool oversight, risks can include data leakage, compliance issues, and misinformation. It’s a challenge that adds to the barriers to AI adoption.

Given the availability of free-to-access tools, as well as new players entering the market such as DeepSeek, businesses can run into difficulty when they have no rules or protections in place around tool use.

“There's not a huge amount of visibility on how and what people are using, and it’s fairly challenging to detect,” says Rudy.

As well as the need to have solid AI governance frameworks in place, Jason argues responsible AI usage should also be considered, with people asking the questions:

“Should I do this? Is it ethically correct? That's much more difficult to get right, but you have to factor in both [governance and responsible use]. It's important to recognise what might be the unintended consequences of deployment and adoption,” he says.

So how can these risks be contained?

Creating a ‘sandbox’ environment where your people can experiment with AI tools safely is one way to protect leaders from risk, data security and compliance challenges.

For it to work, Jason suggests: “It’s a combination of leadership setting the tone and the policies. Make sure agreements are in place so you can use the tools in the sandbox. And then training so that people are aware of the risks.”

As AI projects shift from proof-of-concepts to full adoption, showing the ROI of AI is a recurring theme for leaders making a business case for AI.

And while there is no one-size-fits-all approach or a ‘magic measurement tool’ to share all the answers, the panel recommends going back to basics on measurement.

Rudy advises against using vanity metrics, and instead for people to look at the 'business as usual' KPIs they're already tracking.

It’s a sentiment echoed by Jason, who says how you measure can vary between use cases. He recommends establishing ‘baselines’ that you measure in the business:

“If you've got that in place it's going to make it much easier for you to measure your return on investment.”

With agentic AI influencing trends in 2025, you are likely looking closely at how to overcome your AI adoption challenges.

Rudy and Jason share three principles to help get you started: