Although digital transformation has taken a leap forward in the past two years, many organizations still struggle to get value out of artificial intelligence (AI). The key, according to one expert, is strategic project planning.
“Getting AI off the ground is quite a complex task,” said Jonathan Druker, Product Marketing Manager with OVHcloud at a recent ITWC briefing. “But it bears tremendous fruit for any organization.”
Complexity and cost are among the biggest stumbling blocks. Getting the data out of data lakes and cleaning it is also a major obstacle, said Druker. A lack of skills and compliance requirements can make the process even more complicated.
Watch on demand: “The CIO Imperative: Secure the Future”
“It takes a team to execute AI projects,” said Druker. “Ultimately, it is the business that should drive what you want to explore and work from that point down. If you don’t have the business buy-in, you’re basically burning money.”
Infrastructure is important for managing AI
Once the purpose of the AI project is nailed down, the team should make decisions about infrastructure providers and platforms. For example, it must decide on a cloud architecture, including whether to use hybrid and/or multi-cloud services. As well, it must choose whether to manage the service on its own or opt for a managed service provider.
Druker said he’s seeing trends toward managed and multi-cloud services. Managed services allow companies to focus on their core business and help to address in-house skills shortages. Multi-cloud adoption by large enterprises has increased by 20 per cent in the past few years. It provides the ability to use different cloud providers to suit the use case and pricing requirements or for disaster recovery, he explained.
A public cloud AI platform offers numerous advantages. “The public cloud gives you more flexibility to do things by the project or by the job,” said Druker. “That is the way most AI projects are run. They want to try it out, fail fast and move on and not get too much into the costs.” Likewise, the public cloud allows organizations to get going quickly, rather than having to set up servers.
Another benefit is that public cloud providers can offer high compute power which produces better performance on large data sets. A public AI platform may also include storage options and data processing to clean the data. “Choosing the right type of storage and getting it into the cloud and into your lake is an important first consideration when you’re building your AI project,” said Druker. Finally, there is software that can be installed locally or in the cloud to allow data scientists to set up and train the AI project model.
Practical advice for using AI cost-effectively
An AI initiative should be set up as a “proper project,” with a team, a budget, and a timeline, said Druker. “You need to be aware of all potential expenses.”
Organizations must ensure they know the cost of pulling data out of storage and whether there are charges for moving data inside the network. “You need unmetered traffic for predictable costs as your infrastructure scales up,” he advised. It’s also important to consider how often the AI models will be retrained as this will impact costs.
Another cost control method is to work with providers that use open standards so that systems can talk to each other easily via APIs and to avoid lock-in.
With thorough business, architecture, and project planning in place, Druker said he’s confident that organizations will be able to turn AI into a competitive advantage.
WATCH ON DEMAND: “THE CIO IMPERATIVE: SECURE THE FUTURE”