A year ago, enterprise AI was a research topic. Today it's a board agenda item, a budget line, and a source of genuine organisational anxiety. 2023 was the year everything changed. Here's what we learned.
The Year in Context
January 2023: ChatGPT has just hit 100 million users. Enterprise leaders are asking "what is this?"
March 2023: GPT-4 launches. The capability jump from 3.5 to 4 compresses what we thought was a five-year timeline into five months. Enterprise viability suddenly feels close.
May-June 2023: Every vendor has an AI story. Every conference has an AI track. The hype is deafening. But underneath the noise, real work is starting - pilots, proof of concepts, governance conversations.
July-September 2023: The pilot phase. Enterprises are testing AI against real use cases. Some are working. Many are revealing gaps in data, processes, and organisational readiness. The "pilot graveyard" starts to fill.
October-December 2023: A more sober phase. The initial excitement has been tempered by reality. Organisations that rushed in are discovering that AI deployment requires more than an API key. Organisations that held back are discovering the compound effect of early movers who are now two or three pilots ahead.
What We Learned
1. Data Is Still the Bottleneck
We said this in January. We said it in March. We said it in June. And it remained true in December. The organisations that made the most progress with AI in 2023 were the ones that invested in data infrastructure early. Clean data, accessible data, governed data.
The AI model is the easy part. Getting your data into a state where AI can use it effectively is the hard part. This hasn't changed. I don't think it will change.
2. Change Management Is the Multiplier
The best AI system with poor adoption delivers less value than a mediocre AI system with broad adoption. We saw this pattern over and over. Technical teams built impressive AI capabilities. Users didn't adopt them. Not because the tools were bad, but because the change management was absent.
AI adoption is a human problem, not a technology problem. The organisations that invested in training, in workflow redesign, and in genuine engagement with users got better results than those that invested only in technology.
3. Governance Enables Speed (Seriously)
I've written about this three times this year because it keeps being misunderstood. Every organisation we worked with that had a governance framework - even a simple one - deployed AI faster than those without. The framework answered questions before they were asked: what data can we use? Who approves this? What happens when it's wrong?
Without governance, every decision is bespoke. With governance, teams operate within clear boundaries and move autonomously. Governance is infrastructure.
4. The Compound Effect Is Real
The organisations that started experimenting in Q1, even imperfectly, have a meaningful advantage over those that waited. Not because their first pilot was brilliant - it usually wasn't. But because each initiative built data pipelines, team capability, governance frameworks, and institutional knowledge that made the next one faster and better.
By Q4, the gap between early movers and wait-and-see organisations is visible. It will widen in 2024.
Enterprise AI Momentum Through 2023
Source: RIVER Group observation, December 2023
5. New Zealand Has Specific Advantages (and Specific Gaps)
NZ's small market, short decision chains, and pragmatic culture are advantages in the AI era. We can move faster than larger markets because there are fewer layers to navigate.
But the gaps are real: no sovereign AI infrastructure, a thin talent pool, limited NZ-specific AI solutions, and insufficient attention to indigenous data sovereignty in AI deployment. These are addressable, but they require deliberate investment.
12
months from ChatGPT launch to enterprise AI being a standard board agenda item across NZ and Australia
Source: RIVER Group observation, December 2023
What Surprised Us
The Pace of Model Improvement
We expected AI models to improve. We didn't expect the pace. GPT-3.5 to GPT-4 in four months. Claude's emergence as a credible alternative. Open-source models going from toy to viable. The technology is improving faster than enterprise processes can absorb it.
How Quickly "AI Company" Became Meaningless
By June, every software vendor was an "AI company." The term lost all meaning within months of becoming popular. The hype cycle compressed what normally takes years into weeks.
The Depth of the Data Problem
We knew data would be a challenge. We underestimated how deep the problem goes. It's not just data quality - it's data accessibility, data governance, data literacy, and organisational willingness to invest in infrastructure that isn't visible or exciting.
How Much Governance Matters
We expected governance to be important. We didn't expect it to be the primary differentiator between organisations that deployed AI successfully and those that didn't. But it was, consistently.
Where It's Heading
I'll be careful about predictions - we just published an honest assessment of our track record. But some trajectories feel clear:
AI agents. Systems that take actions, not just generate text. Early and fragile right now, but the enterprise implications are significant.
Multi-model architectures. No single model wins everything. Enterprises will route different tasks to different models based on capability, cost, and data sensitivity.
Sovereign AI. The conversation about where data gets processed will intensify, especially for health, government, and indigenous data.
Cultural intelligence. AI systems need to operate across cultural contexts, not just technical ones. Aotearoa has the opportunity to lead here.
Compounding. The organisations that invested in 2023 will compound in 2024. The gap between leaders and laggards will widen.
What We're Taking Into 2024
We entered 2023 as an enterprise technology company watching AI with interest. We're exiting 2023 with a conviction that AI is the defining platform shift of this era - not because of any single capability, but because of what it enables across every dimension of enterprise operations.
Our focus for 2024: helping NZ enterprises build the foundations - data, governance, capability, culture - that allow AI to compound. Not chasing the latest model. Building the infrastructure that makes every model more useful.
It was a hell of a year. The next one will be bigger.