Companies are failing to turn data science efforts and artificial intelligence implementation into real economic value, a new report says. The survey of 2,500 senior technology leaders found that despite high expectations, only a quarter were highly satisfied with AI performance.
This missing value is worth some $460bn in incremental profits across all the companies surveyed, the report from ITSP Infosys claims, with those companies gaining the most from AI focused on ensuring data science is fully integrated into the business, not just a side project
“It is crucial that companies do not view data and AI separately from the business, but instead think differently about it,” Mohit Joshi, president of Infosys told Tech Monitor. The key findings from the report are that the solution is to focus on three areas – data sharing, trust in advanced AI and business focus.
Despite high expectations when first launching AI projects, most companies failed to act on one or more of these key areas, the report revealed. In total 63% of AI models function only at basic capability, are driven by humans, and often fall short on data verification, data practices, and data strategies.
Only 26% of those questioned said they were highly satisfied with their data and AI tools. “Despite the siren song of AI, something is clearly missing,” said Joshi.
The UK had the highest overall satisfaction level with AI despite having one of the lowest data-sharing rates and a strong preference for on-premises AI apps rather than turning to cloud solutions, which could cause problems down the line.
“The most effective and useful data for a business problem and AI system may sit outside the walls of an organisation,” he explained adding that trusting the AI was also important.
“Our research found that advanced AI requires trust in AI to perform optimally. If people working alongside AI do not trust it, the model risks going unused. Best practice in data ethics and bias management is central to advancing AI.”
AI implementation: businesses seek automated tools
Other findings from the survey included the fact three out of four companies want to operate AI across their business, but most are new to AI and face daunting challenges to scale, heavily driven by a lack of skills and struggles with recruitment.
The Data+AI Radar research was carried out by the Infosys Knowledge Institute which found that what it describes as "high-performing" companies think differently about AI and data, with those treating data like currency – sharing it and letting it circulate – seeing the highest return.
When treated like currency and circulated through hub-and-spoke data management models companies could see $105bn incremental value, and those that refresh the data with low latency generate even more profit, revenue and other measures of value, the team discovered.
Outside of revenue increases companies that were highly satisfied with their use of AI have consistently trustworthy, ethical, and responsible data practices that the “challenges of data verification and bias, build trust, and enable practitioners to use deep learning and other advanced algorithms,” the report claims.
Those businesses that apply data science to practical requirements also created additional value, with integration accelerating efficiencies worth an additional $45bn profit growth.
When asked whether the rapid growth of AI had left some companies struggling to catch up, Joshi said the issue was whether companies can achieve good results as they apply AI. “AI and machine learning requires a new way of thinking, and that is where businesses need to pivot. Despite its fast advancements, we see that it’s the companies who have reframed their approach to data which are gaining the most value from machine learning and AI.”
Part of this is getting the data that is used to feed AI tools into shape and preparing it in a way to works for the business, which includes recognising the need to combine that data with practices that encourage sharing via a hub-and-spoke data management system.
“We believe that data is a new currency, and just like a currency, it increases in value when it is circulated. Many companies recognise that the emerging data economy holds great potential and that establishing a data-sharing ecosystem with partners and peers can deliver greater benefits than keeping it in isolation,” said Joshi.
This deviates from the traditional thought process that calls for data to be centralised. Joshi says they found that a system that centralises and organises data but then relies on spokes to give teams freedom to operate and use it flexibility is the best approach.”For example, importing data from third parties and high levels of data sharing delivered the highest boost to the bottom line than any other data or AI action.”
'Model ops' can help scale AI systems
Joshi said if companies fail to act now and start thinking differently about AI and machine learning they are going to come up against limitations, a lack of satisfaction with AI and struggles in the new data economy. “Companies will need to adopt an AI deployment framework that not only allows for a level experimentation, but can scale AI in a predictable manner," he added.
“Concepts such as 'model ops' can provide an architectural perspective of the enterprise to build a scalable platform driving that can drive agility in the roll-out, ensure processes are made standard, and support as a measure for baseline model performance.”
The other aspect Joshi says is important is ensuring companies uphold ethical and legal practices, particularly during this interim period while governments create legislation to protect against data misuse and unethical practices.
“AI must be adopted in a sustainable and thoughtful manner, so that it can co-exist with our social fabric and bring greater good,” he said. "It is therefore important that the technology industry promotes discussion within and across industries, communities and regulatory bodies on the benefits, interests, costs and consequences of any large AI technology, before it is released in the public sphere.”
Read more: What is the future of generative AI?
