The last two years have seen a hive of activity around artificial intelligence. And enterprises’ interest in AI projects shows few signs of decline.
In fact, 48% of organisations are now investing in technologies to develop AI capabilities internally. A further 40% are considering doing so, according to Foundry’s 2025 AI Priorities Study.[1]
Adopting new AI technologies is not risk free. But CIOs have learned a lot about how to deploy AI over the last few years. Here are some of the most valuable lessons.
Lesson 1: Define the scope
As with any IT project, lacking a clear scope leads to mission creep, higher costs, and ultimately, frustrated stakeholders.
A clearly defined project is more likely to make it from pilot to production.
“From my own learnings, we made a few mistakes in the beginning in terms of defining the scope of what we wanted to do,” says Hasmukh Ranjan, CIO and senior vice president at AMD.
“Ask what you want to achieve, what objectives you have for your organisation, and what investments are you making.”
Lesson 2: Take the long view
Experience with deploying older technologies such as machine learning or natural language processing (NLP) can serve as a guide for generative AI projects.
“AI is not just about chatbots and copilots,” says Ranjan. “You must have a longer-term view, and a view of the full spectrum of what AI can bring to your enterprise.”
Ranjan envisions AI developing from simple “assistance” tools to fully autonomous systems. Exploiting this means building in-house AI capabilities, he suggests, and not just relying on the cloud.
Lesson 3: Do the prep
Poorly planned AI projects can “leak” confidential data or create back doors into enterprise systems. Data governance and IT security teams should be involved from the outset.
And good quality datasets are essential for AI. If they are not part of the equation, the results will disappoint, and AI systems will not be trusted.
This is all part of what Ranjan describes as responsible AI. “From a responsible AI perspective, it is about how we use the data. If you are playing with data, it is not a trivial thing,” says Ranjan. This should be a board-level discussion.
Organisations should be prepared to use vastly greater volumes of data. In the past, enterprises might have used just 5-10% of their data. “Now, with AI, you probably have to analyse north of 70% of your data,” says Ranjan.
Lesson 4: Drive an ROI from AI
Lastly, AI projects need to provide a return on investment. Organisations can control AI costs by investing more in AI-capable infrastructure and skills, and running, or even building, models in-house.
Some of this infrastructure – such as AI PCs – is relatively easy to acquire. Other elements, including data centres capable of handling high performance GPUs, takes longer. But careful vendor selection and aligning investments to an AI strategy will help.
“From a CIO perspective, the successful projects are where you have delivered that ROI,” says Ranjan. He points to an AMD project where AI delivered a 5% improvement in performance in compute grids.
There’s a value that’s been harnessed. If you do not have clarity of purpose and aligning to the business goals, then you are wasting time and resources for the company.”
For more information on how AMD can help your business achieve its AI goals, click below.
[1] Foundry’s AI Priorities Study, 2025 Q16