2023 was the year of AI hype. 2024 was supposed to be the year of AI reality. It was, but not in the way most people expected. The reality wasn't that AI failed to deliver. It was that the gap between organisations who approached AI strategically and those who didn't became impossible to ignore.
What You Need to Know
- 2024 was the year the "pilot graveyard" became visible. Enterprises that ran 5-10 AI pilots without a platform strategy are now sitting on a portfolio of disconnected experiments that delivered learning but not value.
- The organisations that got ahead did three things differently: they started with business problems (not technology), they built foundations (not projects), and they invested in governance early (not as an afterthought).
- Model capability stopped being the constraint. GPT-4, Claude 3.5, Gemini 1.5. The models are good enough. The constraint shifted to integration, data quality, change management, and organisational readiness.
- The compound advantage is now measurable. Organisations that built AI foundations in early 2024 are deploying new capabilities 3-4× faster than those still in project mode.
73%
of enterprise AI pilots initiated in 2023 had not reached production by end of 2024
Source: Gartner, AI in the Enterprise Survey, October 2024
3-4×
faster capability deployment for organisations with established AI foundations
Source: RIVER Group, enterprise engagement data, 2024
What Happened in 2024
The Hype Hangover (Q1-Q2)
The first half of 2024 brought a reckoning. After a year of breathless AI announcements, enterprise leaders started asking harder questions: Where's the ROI? Why hasn't the pilot scaled? How much are we actually spending on AI tools that individuals adopted without governance?
The "shadow AI" problem emerged as a serious governance concern. Employees across organisations were using ChatGPT, Claude, and a dozen other tools, uploading company data, making decisions based on AI outputs, with no oversight, no policy, and no audit trail.
Smart organisations treated this as a signal, not a threat. Employee demand for AI tools meant the appetite was there. The job wasn't to suppress it but to channel it through governed, enterprise-grade infrastructure.
The Foundation Builders Emerged (Q2-Q3)
By mid-2024, a clear pattern was visible. The organisations deploying AI successfully weren't the ones with the biggest budgets or the best technical teams. They were the ones who had invested in three things:
Shared infrastructure. Data pipelines, knowledge bases, and integration frameworks that served multiple use cases. When they built their second AI capability, they didn't start from scratch. They built on what already existed.
Governance frameworks. Not just policies in a document, but actual processes for use case approval, risk classification, monitoring, and incident response. These frameworks didn't slow them down. They sped them up by pre-answering the questions that stall every project at the approval stage.
Change management. They invested in helping people work differently. Training, workflow redesign, feedback loops. Technology deployed without change management is technology deployed without value.
Models Got Better, Faster (Q3-Q4)
The pace of model improvement in 2024 was staggering. Claude 3.5 Sonnet from Anthropic delivered a step-change in reasoning and reliability. Google's Gemini 1.5 Pro introduced million-token context windows. OpenAI's o1 model showed what structured reasoning could look like.
For enterprise leaders, the implication was clear: model capability is no longer the bottleneck. The models are good enough for the vast majority of enterprise use cases. The constraint has shifted entirely to how you build around the model: your data, your integration, your governance, your people.
This is actually good news. It means the hard part is organisational, not technical. And organisational problems are solvable with leadership, investment, and time.
The most important shift in 2024 wasn't a new model or a new tool. The organisations that internalised this are the ones pulling ahead.
Isaac Rolfe
Managing Director
What Worked
Starting with the Problem, Not the Technology
The enterprises that delivered AI value in 2024 started with a specific, measurable business problem. "We lose $2M annually to manual claims processing." "Our contract review takes 3 weeks per deal." "Customer response time is 48 hours for questions our knowledge base already answers."
The enterprises that stalled started with: "We need an AI strategy." Strategy is important, but it follows understanding, and understanding comes from concrete use cases with measurable outcomes.
Building for Reuse from Day One
The compound advantage showed up in the numbers. Organisations that designed their first AI capability with reuse in mind (model-agnostic architecture, shared data pipelines, abstracted integration layers) deployed their second capability in half the time and a third of the cost.
By year-end, the foundation builders had 3-4 capabilities live. The project builders had 1-2 pilots in various states of limbo.
Investing in AI-Adjacent Skills
The organisations that moved fastest invested in the skills that surround AI, not just the AI itself. Use case identification. Output evaluation. Workflow redesign. These "AI-adjacent" capabilities determined whether technology investments translated to business value.
What Worked vs What Didn't in Enterprise AI (2024)
Source: RIVER Group and McKinsey, enterprise engagement data, 2024
What Didn't Work
Pilot Proliferation Without a Platform
The most common failure pattern: launching 5-10 pilots across different departments, each with its own tools, vendors, and infrastructure. Some succeeded locally. None scaled. And the organisation ended up with a portfolio of disconnected experiments sharing nothing.
5-10
average number of disconnected AI pilots in enterprises that failed to scale AI in 2024
Source: McKinsey & Company, The State of AI in 2024
The solution isn't fewer pilots. It's a platform that pilots build on. Each pilot should extend the foundation, not create a new silo.
Buying AI Tools Instead of Building AI Capability
There's a meaningful difference between buying an AI tool and building AI capability. Tools are point solutions. They solve one problem. Capability is infrastructure. It enables solutions to many problems.
Many enterprises spent 2024 accumulating AI tool subscriptions: a writing assistant here, a document analyser there, a chatbot for the website. Each tool creates a new vendor relationship, a new data silo, and a new governance gap. None of them compound.
Ignoring Governance Until It Became Urgent
The EU AI Act timeline crystallised in 2024. Australia accelerated its AI governance framework. NZ's Algorithm Charter evolved. Organisations that had treated governance as "something we'll sort out later" found themselves scrambling to retrofit compliance into systems designed without it.
Governance is cheaper to build in than to bolt on. The organisations that learned this lesson early had a structural advantage by year-end.
The Gap Heading Into 2025
The most significant outcome of 2024 is the widening gap between AI leaders and AI laggards. This gap is primarily about organisational capability, not technology adoption.
AI leaders have: governed AI infrastructure, internal teams who understand AI, 3-4 production capabilities, measurable ROI, and a platform for compounding.
AI laggards have: scattered pilots, tool subscriptions, no governance framework, no internal capability, and a growing sense that AI isn't delivering.
The gap will widen in 2025. Foundation builders will add capabilities at increasing speed. Pilot experimenters will continue to start from scratch.
Looking Forward
Three dynamics will define enterprise AI in 2025:
Consolidation. Enterprises will rationalise their AI tool sprawl, replacing point solutions with platform capabilities. The winners will be AI partners who build foundations, not vendors who sell features.
Regulation. AI governance will shift from voluntary to expected, driven by international precedent and local regulatory evolution. Organisations with frameworks in place will move faster; those without will slow down.
Compound returns. The foundation builders will enter the exponential phase of compound value, where each new capability is faster, cheaper, and more powerful than the last. This is the most important dynamic in enterprise AI, and 2025 is when it becomes undeniable.
2024 was the year enterprise AI got real. Not because the technology matured (it was already mature). Because the gap between strategic adoption and experimental adoption became visible, measurable, and increasingly permanent.
The question for every enterprise entering 2025 is straightforward: are you building a foundation, or are you still running pilots?