17 Oct The Gartner AI Hype Cycle Reality Check
Last year, AI was riding high on the Gartner Hype Cycle’s Peak of Inflated Expectations, with generative AI hyped as the ultimate game-changer. In 2025, the Gartner Symposium, held on the Gold Coast last month, and temporarily renamed the “Gartner AI Symposium” describes AI as now moving past its Peak and dipping into the Trough of Disillusionment. They didn’t actually say this, but aligning symptoms with the Hype Cycle, then this is where we now are.
For instance, only one in five AI initiatives delivers ROI, and 95% of generative AI pilots flop. It was said that 74% of CFOs see time savings, but only 11% get financial wins—5% from cost cuts, 6% from revenue boosts—due to “productivity leakage.” So time saved, but no dollars unless processes are reengineered.
This move from hype to reality along the Hype cycle was actually welcomed this year by me. More practical insights and views, and a good reality check on navigating the way ahead, versus last year’s run for the hidden honeypot message.
Piloting versus scaling
ROI requires investment. And investment requires buy-in and alignment across the executive team.
AI has a lot of variability in investment options—from short-term pilots to long-term bets—all coming with wildly different payback periods and risk profiles. McKinsey’s 2025 report echoes this, noting only 15% of companies see significant AI returns, often due to misaligned strategies (McKinsey & Company, 2025). One Gartner session on “Healthcare AI Transformation” showed how AI’s driving diagnostics but not user satisfaction. Investment selection through the right high-impact use cases with clear metrics—and you may start to scale Gartner’s very own Slope of Enlightenment. Do this with a unified leadership and you may even stay there.
At IPG, our first objective for scaling AI is aligning on ROI and use cases.
From LLMs to Agents
Agentic AI—autonomous systems that act like your smartest teammate, took a lot of attention, and understandably so. We don’t just need agents, but multiple agents that can work together to seamlessly solve problems requiring multiple specialisations. Think Agent 86 and 99 working together, and Agent 13 pops in, hiding in the keyboard, just when you need them.
A simple example is an AI booking a restaurant table, syncing calendars, and mapping routes in seconds, followed by a well-timed pre-emptive Uber booking to get home (autonomous driving perhaps). This implies the need for multi-agent systems, agent-agent comms, and domain-specific models tailored to specific industries like healthcare’s “digital front doors” or government automation. Forrester predicts agentic AI will rule workflows by 2027 (Forrester, 2025).
One way of thinking about agents is pre-built, pre-informed, single-purpose LLM prompts, with pre-defined inputs and outputs.
Pre-defined inputs implies data is to be available. What data?
AI thrives on clean, accessible, reliable, and secure data. Without understanding your data holdings, quality, and protection needs, you cannot scale. It is that simple. Probabilistic models like LLMs will hallucinate if fed poor data or no data, just like a low-grade under-informed salesperson. A 2025 Deloitte report warns that 70% of organizations struggle with data quality, stalling AI projects (Deloitte, 2025). We would concur with this view, although we also see clients now addressing their data management so as to ramp up AI use. Depending on the data you have, its reliability, and availability, and protection requirements, then some Use Cases and ROI will be within reach and others may need delay. Hence, there is dependency between data and the available investment and Use Case options.
This is IPG’s second objective for scaling AI, data management.
ROI, Use Cases, Data…and more decisions to be made
Often a root cause for a very fragmented process and data environment is a fragmented technology environment. And the root cause for all of these is a fragmented decision-making environment.
In ensuring alignment of the investment profile and use cases, as previously mentioned, there are important decisions to be made. In improving data management there are also important decisions to be made. Progressing AI may also have decisions to be made in relation to Cyber and Information Security, new technology product decisions and potentially many other areas such as process redesign. Avoiding fragmented decision making and repeating past failures will again be critical across all areas of AI.
Plus, you do want to empower staff and team to use AI to the maximum possible, and so policies in relevant areas are critical to channel activity into appropriate use. Let’s face it, policies are effectively pre-emptive decisions. Gartner included some sessions on AI Governance to manage risks. This guides all staff use, and also guides the design, testing and deployment of new AI tools.
Singapore is ahead here as a country. It was covered at Gartner, and a discovery I made in my technology delegation to Singapore in March 2025. They have “guardrails as a service” and “testing as a service” tools to standardise AI adoption across agencies, blocking inappropriate content and ensuring compliance as models evolve. Australia’s 2025 AI Strategy Update takes a complementary tack, mandating risk assessments, and public consultation for high-stakes AI (such as in healthcare or law enforcement).
IPG’s third objective to support AI scaling is governance—not just for AI but for cyber, data, and investment decisions and any other related areas.
Geopolitics: What’s for AI?
Geopolitics has impacted information security (GDPR and the US Cloud Act) and thereby impacted procurement and contracting. This has some implications for AI. But AI is also driving other risks in this area.
Disclosure of AI use in products or by providers is an obvious risk but still dismissed by some legal advisors (from what we have seen).
Data and Capability sovereignty is a quickly emerging risk. Singapore is busy building its own LLM, to avoid its culture being reliant upon US, European or Chinese LLM, which may provide their own view of South East Asia. Trade blocks, IP restrictions, and specific and emerging country by country AI legislation are all part of the changing landscape. We have seen the recent debate on copyright holders’ rights in relation to the training of LLMs. These laws are progressing differently across the world.
Whether any of these have material consequence for you and your organisation will vary case by case. But it is another reason to have the Governance defined and agreed. As these areas emerge, they can be watched for impacts that can be dealt with in a co-ordinated way, with investment decisions and policies updated accordingly.
Capabilities: What’s needed where?
When many people think skills shortages, often thoughts turn to programming skills and other technical skills. While these are important, for most businesses, it is some combination of business, process, data, and technical skills that is most sought after and creating constraints on progress.
AI is no different. As referenced above, CFOs are seeing productivity but not financial savings without business process reengineering. To achieve this requires the blend of skills across business, process, data, and technical domains. And often when these skills do exist, they are stretched across multiple priorities.
AI brings new technology, new Use Cases and new ways of thinking, plus will require uplift and changes across business, process, data domains.
One Gartner session referenced a HBR 2025 paper noting that 60% of AI failures stem from poor change management. A 2025 BCG report highlights that 65% of organizations lack the talent to scale AI, making capability-building a priority.
Your current capabilities create your constraints, and your required capabilities create your objectives. And these are all subject to the Use Cases you select and the Investment profile you have chosen. And this Investment profile needs to include relevant capability uplift to scale and sustain any improvements.
Hence, capability development is IPG’s fourth objective to scaling AI.
Next Steps
Now that this year’s Gartner Symposium pared back the hype to reflect where AI now sits on its AI Hype Cycle, the information available and views discussed were much more grounded and practical. Coupled with improvements in tooling, improvements in policy direction and many more use cases of real-world experience, means that non-leading edge organisations (most organisations) have great opportunities to learn quickly and accelerate. And in summary that means getting focused on:
- Aligning ROI and Use Cases: To ensure the ROI profile for the investment is understood and agreed, and Use Cases are selected to generate the intended benefits.
- Uplifting Data Management, prioritising these improvements commensurate with target use cases.
- Establish Governance via new or existing forums to oversee decision making on the organisation’s AI use and all related areas.
- Understand existing, and position for uplift of capabilities, and include this as both a constraint in Use Case selection, and an objective in Use Case delivery.
Not sure where your organisation sits on the AI Hype Cycle, or what's needed to move past it?
Take the AI Readiness Assessment
Mark D Nicholls
CEO and Partner, Information Professionals Group (IPG)
IPG provides independent leadership support for government agencies, utilities, health services, and growing organisations across Australia on ICT and digital strategy, program delivery, and technology transformation.
Know more of our DIGITAL & ICT STRATEGY capabilities. Most organisations are continuously challenged, seeking to stay ahead through
Know more of our DIGITAL & ICT STRATEGY capabilities. Technology in the financial industry would normally bring about thinking
Know more of our DIGITAL & ICT STRATEGY capabilities. A new survey into the scope and frequency of legal
Know more of our DIGITAL & ICT STRATEGY capabilities. Gartner looks into the future, or tries to define it,
- When Governments Shape Digital Outcomes: What the Evidence Actually Shows - 20 May 2026
- The Gartner AI Hype Cycle Reality Check - 17 October 2025
- Digital Project Delivery – How Is It Going, and How to Tell Easily? - 6 September 2024