“Garbage, garbage output” is not the worst in the world, but putting trust in that garbage warned Gartner Vice Presidential Analyst Carly Aidone in a keynote address at the Gartner Data & Analysis Summit held in Sydney this month.
She was joined by Vice Presidential Analyst Gareth Herschel. He highlighted the 2024 Gartner survey, which found data availability and quality to be the number one obstacle to implementing artificial intelligence (AI).
Governance is key to reliable data, but Herschel said it is not practical to achieve fully managed data before providing AI-powered capabilities. The answer he suggests is to implement a trust model that evaluates the reliability of data based on its lineage and curation, which can significantly reduce the risk of those using incorrect data.
An important theme of the summit was the distinction between different forms of AI. Erick Brethenoux, chief of AI Research and well-known vice presidential analyst, noted that the generator AI and AI agents “have nothing to do with each other.”
While AI agents have been around for at least 30 years and are used for tasks such as predictive machine maintenance, agent AI is primarily a marketing term. Vendors tend to blend the two concepts because of the high potential for revenue generation AI, but Brethenoux said “it's important to name the right way.”
“AI agents can use models such as large language models. [LLMs]He said, combining the two could potentially produce interesting results. However, since the generation AI is non-deterministic, it is impossible to rely on traditional tests as the same prompt can lead to different responses. Instead, organizations need to place guardrails around the model and run the simulation to work as intended.
One of the main advantages of agents is that they simply consume resources when the agent is active, making the multi-step process a faster and more cost-effective way in parallel. For example, if your loan application includes multiple checks, you can assign each to a different agent. Then, instead of waiting for all checks to complete in order, you can terminate the process as soon as one check fails.
Brethenoux said that if humans are not necessarily working in the background without making final decisions, it may be appropriate to allow the system to respond automatically if reaction times are important or delaying the decision increases risk. This also applies when the risk is low. For example, agents can automatically arrange trips based on the user's past preferences. In other circumstances, it is recommended that you suggest human checks before they are implemented.
“Autonomy is one of the most sticky issues we have with software agents,” observed Bretenuu, noting that people are comfortable getting advice from the software, but they are still adapting to autonomous behavior.
Return on investment
However, according to Gartner's vice presidential analyst Luke Ellarie, the economics of AI are not as simple as they look. Citing Microsoft's numbers, he noted that Copilot saves the average employee only 14 minutes a day. Considering productivity leaks, this is worth around $800 a year, but the cost of the co-pilot is all about $1,150.
Ellery said the benefits lie elsewhere. People using Copilot recognize themselves as productive each day, with employees having an average Net Promoter Score (NPS) of 59 compared to a 21 average.
Elary said that return on investment is suitable for evaluating generative AI use cases that extend current capabilities, but that is not a whole story. Using generative AI to fundamentally change business models is a more complicated situation, often with millions of dollars of simultaneous investments over the long term horizon.
As a result, he recommended that use cases be categorized into the type of value to be created, expectations must be carefully set for stakeholders, and that organizations should build a portfolio of projects that are collectively consistent with the desired outcome.
Perceptual analysis
Gartner's Supervision – Analyst Georgia O'Callaghan talks about the future of analytics, explaining that Gartner uses the term “perceptive” to describe a constantly on-system where it can understand the environment, raise alerts or take action.
She emphasized that the amount of automation should be appropriate based on the specific task at hand. For example, the system may suggest a series of actions in a medical situation for human approval, but is trusted to autonomously generate maintenance tickets for factory machinery.
O'Callaghan predicted by 2027 that enhanced analytics capabilities will evolve into an autonomous analytics platform that fully manages and executes 20% of business processes. This shift has several implications as analytical work is led from experts to non-expert users to AI agents, and the analysis itself becomes more aggressive rather than reactive. This replaces standalone tools with integrated systems, replacing interactive dashboards with dynamic, embedded insights. “Perceptual analysis will be everywhere,” she said.
Gartner presented its top data and analytical forecasts to guide client strategies. By 2027, we predicted that 50% of business decisions will be augmented or automated by AI agents for decision-making. Furthermore, organizations that emphasize executive AI literacy achieve 20% higher financial performance than those that do not. Gartner also expects organizations to implement task-specific small AI models with at least three times more usage than general-purpose LLM.
Looking further, Idoine says, “AI Agent is a new UI [user interface]”We predict that by 2030 they will leave Software-as-asa-asasa-application to a rich and rich data source.
Idoine concluded with a specific warning to Chief Data and Analysts, saying that 75% of people who have no measurable impact across the organization and not have an impact on measurable one, their top priorities will be assimilated into technical capabilities.
“Your role may be relatively new, but as it becomes more established, tensions can arise in your authority and responsibility,” she said. “AI has become a juicy award that other senior leaders look to.”
“Trust yourself, push the frontier and show how you and your team are crucial to AI success for your organization,” she added.
