AI is no longer “new”. Businesses have run pilots, new roles and career paths have emerged, and as of 2024, over $1.6 trillion was invested globally in the AI sector. With five months left in the year, if you’re a business still hedging its bets, waiting to see what its competitors are doing, now is the time to start.
Why successful AI adoption requires more than the technology
Gartner predicts that over 40% of agentic AI projects will be cancelled by 2027 and found that 50% of GenAI projects were abandoned at the end of last year. MIT research found a 95% failure rate for generative AI pilots at enterprises. But if you dig behind these headlines, the MIT research shows that 25% of those who have started piloting are seeing measurable PnL improvement from their Pilot alone, and pilots usually are not designed to change anything; they are just to pilot. The reality is that the technology works, what doesn’t work is how most companies are implementing it.
Companies are treating AI as something that can be bolted onto legacy operations, yet the challenge is that turning technology into bottom-line impact depends primarily on people and processes. This means redesigning workflows, redefining roles, and educating workforces to adapt to these transformational technologies. In practice, this means clear ownership and accountability for AI initiatives, consistent education and open communication with workforces about how their roles are changing, and an honest assessment of what problems AI is truly poised to solve. AI isn’t going anywhere, so ensuring the environment it operates in is optimum is critical to success.
Accelerating AI capabilities
We’ve seen this happen over the last few years, and 2026 will be the year AI advances beyond anything we’ve seen so far. Take the launch of Anthropic’s Mythos: which reportedly achieved in three months what the industry expected would take six. That may not sound dramatic at first, but when the baseline trajectory is already exponential, accelerating a power-law curve is anything but incremental.
This year, AI is proving to be about more than productivity gains, and instead, instrumental change that sees AI, and automated AI agents, working alongside human workforces. As such, the gap between adopters who have built the right foundations for AI, and everyone else, will widen dramatically. For organisations trying to catch up, this means improving understanding within the workforce, and committing to setting up these initiatives with contextual goals, giving them the time to prove their value. Expecting value instantaneously means believing the hype, without allocating the resources that allow AI to deliver.
What about sovereign AI models?
Almost every major tech company has released an AI offering, and burgeoning models from small but mighty competitors are being released constantly. But for organisations that aren’t seeing the value in off-the-shelf solutions, there’s an opportunity to develop context-specific, internally developed models. These AI tools can be tailored to specific workflows, efficiency needs, and compliance requirements – essential for regulated industries and the public sector, and McKinsey suggests that sovereign AI could represent a market of $600 billion by 2030. This path should be approached with caution.
The risk of building around a single model
As state-of-the-art models continue to accelerate, much of the value in tailoring a sovereign model risks becoming obsolete almost overnight. LLMs may well be the fastest-depreciating assets in human history. For many, arguably most, a more resilient strategy is to design AI systems with model interchangeability at their core, placing greater emphasis on agent evaluation and orchestration rather than on the underlying model itself.
This year so far has also seen an increase in context-specific tools. Enterprises have already adopted or implemented tools from Anthropic and Microsoft into their workflows; they now need AI tools that are built with their sectors and specific workflows and pain points in mind. So far this year, we’ve seen the launch of Claude for Healthcare, and for Financial Services, signifying the desire for these more specialised tools.
If you haven’t started, what should you do today?
If you’re behind, it’s time to move deliberately for the rest of 2026. The pilot mentality is one of the biggest problems that companies must overcome. It’s become too common for organisations to run a small test with promising results, only to be hindered by being unable to scale because organisational change hasn’t taken place.
Start with the work that has business impact
Start by identifying a high-impact use case in one of the few core processes that drive the majority of business value (often ~10 processes = ~70% of value). Focus on tangible outcomes, time saved or results improved, rather than the technology itself. Not all automation is equal; for example, fully automating F&A might only move the P&L by ~1%. Lead with real business impact, then measure, learn, scale, and repeat.
Control what matters, then introduce AI for the rest
Start with what won’t change as AI evolves, namely, the underlying processes and data flows. Understand them deeply. Then ask: Which decisions must remain with humans from a regulatory or enterprise control perspective? By definition, everything else is negotiable and can be delegated to AI and automation, even when existing workforce assumptions suggest otherwise.
Invest in people before platforms
Invest in your people before your platforms: The best AI tool in the world is simply useless if your team doesn’t understand it or know how to work with it. Expecting efficiencies without foundational training is a quick way to use up resources and halt your AI strategy before it has even begun.
Look beyond the initial costs of AI
Choose AI technologies that improve as underlying models advance. Strategically, you want to ride the tailwind of the hundreds of billions being invested in model development, not work against it.
When making build vs. buy decisions, recognise that up to 80% of total costs occur in the run phase. Don’t underestimate the value of contractually hedging downside risk, a lever you typically don’t have with internal teams.
The most sophisticated AI models in the world can’t make up for a resistance to organisational change, and the remainder of 2026 will be the test, in which we see the divide between the organisations that continue on this path and those that figure out how to change.
