Despite the hype and promise, only a small fraction of organisations have deployed and used artificial intelligence (AI) at scale. The vast majority are still working to define their AI strategy – identifying uses and applicability; devising business and technology models; and piloting early initiatives.
Infrastructure and operations (I&O) leaders are challenged with creating agile infrastructure to enable productive AI strategies. Twenty-five percent will be investing in and deploying AI-based initiatives by 2022, up from 4 percent this year, according to Gartner.
Successful AI initiatives require the adoption of new technologies, processes and governance models. Many organisations, however, struggle to identify the right infrastructure strategies due to a diverse set of factors – lack of employees with relevant skill sets; high infrastructure growth rate and aligned management complexity; exponential increase in data volumes; increasing demand for insights to support decisions; and shrinking IT budgets. This results in adoption latency and increases potential for competitive disadvantage.
In addition to supporting AI initiatives with infrastructure, I&O teams will themselves be using AI technologies. Gartner predicts that 30 percent of data centres that fail to effectively apply AI to support enterprise business will not be operationally and economically viable by 2020.
To avoid becoming a statistic, start by analysing successful AI initiatives. They reveal a common pattern – devising successful strategies means aligning AI initiatives to business value. To achieve this, there are four essential actions you must take; however, some are more difficult and time consuming to implement than others.
• Use intelligent automation to free up skilled IT personnel and build dexterity
With new business drivers and digital business initiatives like AI and Internet of Things (IoT), I&O leaders will be expected to manage an extreme growth in complexity. As a result, I&O teams are often bogged down in low-value, repetitious tasks like analysing log files and trying to perform root-cause analysis when problems occur.
Optimise infrastructure management by embracing intelligent automation of I&O and free up skilled IT resources from low-value, repetitive tasks. Instead, focus on reskilling high-performing I&O teams with new AI and data analytics skills. Build digital dexterity by addressing the technical skills gap and investing in change-aware culture and more versatile roles.
• Engage business units to devise productive infrastructure strategies
To succeed in AI initiatives, lead the transformation of I&O teams from IT service centrism to a collaborative model with business units and chief data officer (CDO) organisations. By partnering with business units to curate strategies that are contextually relevant and aligned with revenue-critical results, you’ll be able to devise productive and optimised infrastructure strategies for AI.
Rather than trying to solve all organisational problems at once, start with small pilot projects; use centres of excellence to scale success; and leverage agile methodologies to quickly validate projects that are relevant to the business and eliminate projects that are counterproductive.
• Implement strategic data collect and data connect initiatives
Data management challenges – such as data silos, quantity and quality – are another major source of stalled AI initiatives. Trying to solve all of these challenges at once can easily overwhelm even the best I&O teams.
Instead, accelerate AI implementation by differentiating between "data collect" and "data connect" initiatives. By aligning data collection and curation (clean up and transformation) strategies around the right AI uses, you can devise infrastructure strategies that are aligned to revenue-critical outcomes. Whereas, devising data connection strategies can lead to new disruptive results using AI, when data elements are connected that enable the identification of new higher-value features.
• Use AI workload requirements to drive technology selection
Delivering infrastructure in support of AI initiatives can introduce new integration complexity, productivity and cost challenges. In particular, the complexity of integrating compute acceleration technologies like FPGAs, ASICs and GPUs, can result in infrastructure that is overprovisioned or misaligned to target uses. In many cases, I&O leaders err toward overprovisioning specialised infrastructures, resulting in underutilised infrastructure and project cost overruns.
Use AI workload requirements to guide your infrastructure selection strategies ranging from accelerated compute infrastructure to opportunistic investment in cloud and hybrid strategies. Select technologies that feature broad ecosystem support. If you need to deploy compute accelerators, actively minimise risk by selecting technologies that feature the broadest software framework support and mature software deployment environments.
Pankaj Prasad is a principal research analyst at Gartner, focusing on data centre performance analysis and infrastructure monitoring (ITIM). He will be presenting on AI in infrastructure and operations at Gartner IT Infrastructure, Operations Management and Data Centre Summit in Sydney, 30 April – 1 May 2018.