4 ways to scale generative AI experiments into production services

The game-changing potential of generative AI (gen AI) is the talk of the boardroom. However, turning AI explorations into production-level services is proving challenging.

Recent research from Deloitte found that over two-thirds of executives believe fewer than one-third of their gen AI experiments will be fully scaled in the next three to six months.

The consultant said that while enterprises have seen “encouraging returns” on their initial AI investments, they often find that creating value with gen AI and deploying it at scale is hard work.

Also: 5 ways to turn AI’s time-saving magic into your productivity superpower

That sentiment resonated with Madoc Batters, head of cloud and IT security at Warner Leisure Hotels, when asked by ZDNET to contemplate the state of AI and the hype surrounding emerging technology.

“There’s a lot of talk about gen AI, and a lot of people saying they’re going to put the technology into certain areas of their business, but there aren’t many people doing it,” he said.

Batters has a long-standing interest in exploring AI and machine learning. Rather than sitting on the sidelines and waiting for other digital leaders to progress in AI, he’s helping Warner put emerging technology into production. Here are his four best-practice lessons.

1. Build from the bottom

Batters said digital and business leaders often feel under pressure to exploit AI as quickly as possible — and that’s a mistake.

“Many people focus on gen AI because it’s that burning sun in the sky,” he said. “They feel like they have to do work in this area. And I think, sometimes, you need to get all the other bits of the foundations in place first.”

Batters said essential underlying elements, including data, cloud, and networks, support Warner’s AI transformation efforts. Warner has a cloud-first strategy and uses technology specialist Alkira’s network infrastructure-as-a-service approach.

Also: Is your business AI-ready? 5 ways to avoid falling behind

A crucial element of the Warner approach is GitOps, an operational framework that extends software development best practices to infrastructure automation.  

Batters said these strong foundations are crucial for assessing how AI can boost operational processes.

“I go back to the whole ethos of what I believe is a proper cloud deployment, and that’s a deployment with a GitOps methodology and a pipeline in place,” he said.

“Once you get there, you can plug gen AI in and experiment with it.”

2. Experiment in new areas

Batters said a willingness to test is crucial for business leaders who want to push gen AI services into production.

“You need to experiment, make sure it works or doesn’t work, and be able to change things quickly,” he said, suggesting the importance of the oft-repeated mantra in IT development of “fail fast”.

“Having a pipeline that allows you to effect change is key. Then you’re ready to start experimenting with gen AI. See what works and what doesn’t. If it fails, you can fall back.”

While many companies struggle to turn AI explorations into production systems, research from consultant McKinsey suggests IT is the business function that has seen the largest increase in AI use during the past six months, with the share of respondents using AI increasing from 27% to 36%.

Also: 5 ways to boost your team’s productivity – without relying on generative AI

Warner has integrated gen AI into its FinOps pipeline. FinOps is a discipline that combines financial management with cloud operations to optimize spending. Batters said the company’s IT professionals are benefiting from the pioneering integration.

“It’s like having a FinOps person on their shoulder, just giving them suggestions as they do their work,” Batters said.

Warner has worked closely with AWS and its foundational models. The company also uses Infracost, a specialist solution that shows cost estimates and FinOps best practices for Terraform, the open-source infrastructure-as-code tool.

“Whenever we deploy any infrastructure as code, our gen AI tools will look at what we’re deploying, and the associated resources around that deployment, and it will make suggestions to optimize those resources, to cut down on costs or even right-size or scale up those resources,” he said.

3. Give workers a choice

Deploying gen AI into production often involves a new way of working. So, what do Warner’s IT and line-of-business professionals think of the technology?

Batters said they’re impressed, and that’s due to the company’s careful approach to implementation.

“We don’t enforce anything,” he said. “We can put guardrails on to stop people deploying things if we think it’s too much. But we believe in giving developers the autonomy of choice and being able to decide if it’s a good or bad thing.”

Also: The top 20 AI tools of 2025 – and the No. 1 thing to remember when you use them

Batters said giving people a choice to use or not use emerging technology is an important part of innovation.

“It’s like saying to your children, ‘Eat your vegetables,'” he said. “It’s down to them if they’ll eat them. But you can keep putting the vegetables on their plates and, in the end, it becomes the norm, and they’ll be more adjusted to do it, and you haven’t forced them into making a choice.”

Where workers have chosen to use gen AI, the results have been beneficial. 

“We can see where people have put their pull requests in, and once they’ve seen the recommendations come back, they will change them to meet those recommendations,” said Batters. 

“We’ve got some hard stats to say we’ve had developers save money over time by modifying their IT resources down.”

4. Keep exploring carefully

Batters said a challenge his business has found, and one that’s likely to be common across all enterprises, is ensuring data is ready for AI-led initiatives.

Once that hurdle is cleared, it’s easier to consider using gen AI across other use cases.

Also: 5 ways to manage your team more effectively in the AI-enabled enterprise

“This technology is cheap, especially when using it within your cloud deployment, rather than going externally to the third-party companies,” he said.

“You must embrace gen AI. If you don’t use it, your business could be left behind. However, you have to use gen AI responsibly, so that you’re not exposing any of your company’s data.”

Batters said the choice of models is crucial. Business leaders must ensure they know what’s happening with their data and how it’s used by a model, including for training purposes.

He also said prompting is critical to success — even more important, potentially, than the model your business chooses.

Also: AI agents aren’t just assistants: How they’re changing the future of work today

“You could pay for a much larger, more expensive model, and feed a basic prompt into it. Or you could use a cheaper, much smaller model and feed a good prompt into it, and you could get way better results out of that smaller model,” said Batters.

“Success isn’t all about the model’s size. It’s about how good your prompting and workflows are. You may ask your model a question and say, ‘Hey, based on the output you’ve just given me, I’m going to ask another question.’ So, it’s asking multiple levels of questions within your prompting and establishing a workflow for the query.”

Want more stories about AI? Sign up for Innovation, our weekly newsletter.