Why It’s Hard to Build a Durable AI Strategy
Organizations are making decisions now about next year’s priorities and budgets. When it comes to AI, many of the assumptions behind those decisions may be outdated before 2027 begins.
Think about what has changed in the last year:
- ChatGPT has long been the dominant player with upwards of 60% market share this time last year. By May 2026, that fell below 50% for the first time.
- A year ago, Claude Cowork and Claude Design did not exist, and Anthropic felt like a startup. Claude ended 2025 with just 5% market share, by May it was 14%.
- Coding agents were starting to be useful to developers. But the idea that someone without an engineering background could use one to build a functional internal tool still felt novel. Now, it is increasingly normal.
It may feel impossible to plan a year out for AI. And yet, a year is a standard planning horizon for most companies. So why is planning for AI so uniquely challenging? Here are three reasons.
1. The assumptions behind the plan keep changing
Annual plans rely on assumptions about what teams can do, what it will cost, and what needs to be built. With AI, those assumptions can change in months.
A goal that looked too expensive or technically difficult during planning may become viable before the budget is approved. At the same time, something the organization planned to build may suddenly become a standard feature in a product it already uses.
Consider an organization that had planned to build its own internal knowledge assistant. They may suddenly find that much of the underlying functionality is available off the shelf when something like OpenAI’s company knowledge or Anthropic’s Enterprise Search launches.
Committing significant team capacity and budget feels increasingly like a gamble when it’s likely a more effective AI approach will emerge.
2. The evidence develops more slowly than investment decisions
AI produces very visible signs of progress: a polished demo, high adoption, hundreds of employee-created tools, billions of tokens used.
Those numbers are easy to share, but they don’t translate into business value.
In 2024, Klarna announced its AI assistant had taken two-thirds of all customer service chats, the equivalent work of 700 full-time agents, and that handle time dropped from 11 minutes to under two (Klarna). Speed, volume, and cost were quick metrics to measure and led to early decisions to cut staff. But those measures captured efficiency more clearly than the primary business objectives: supporting customers.
By May 2025 the CEO said the company had gone too far and would rehire humans because the AI-only experience was hurting the customer experience (Forbes). Klarna’s early signals were the basis for bigger investments, which means there wasn’t time budgeted to see the bigger picture.
Planning has always been an exercise in imperfect information. But the hyped grandeur of transformation leaves many leaders feeling like they have to make big changes, before understanding what is actually driving value.
3. AI moves faster than organizations can adapt
In a study of more than 1,250 firms, BCG reported that only 5% were getting substantial value from AI.
Value doesn’t come from access to the technology alone. In McKinsey’s analysis of 25 organizational practices, workflow redesign had the strongest relationship with bottom-line impact from generative AI. Adding a new tool to an existing process may make individual tasks faster; but creating value at scale requires rethinking the process itself.
Moderna is often cited as an example of rapid enterprise AI adoption. Within two months of introducing ChatGPT Enterprise, 40% of weekly users had created custom GPTs.
But that speed was built on years of preparation. Moderna had a dedicated team driving AI transformation. It conducted employee research, offered training, developed internal champions, and involved executive leadership. Before ChatGPT Enterprise arrived, its internal AI assistant had already been adopted by more than 80% of employees (OpenAI).
Moderna was able to adapt quickly because it had already built the technical infrastructure, organizational knowledge, and culture required to evaluate and absorb new technology as it emerged.
Planning for an uncertain future
A durable AI plan cannot depend on predicting exactly where the technology will be in 2027. It needs to make room for the assumptions, evidence, and available products to change.
In our planning work, we’ve found that the best strategies are grounded in business objectives, and build in optionality into the roadmap, so you can refine your AI priorities as the landscape shifts.
We’ve codified these best practices into our 2027 AI Planning Sprint. In this rapid engagement, we help leaders build an ambitious, executable AI strategy that balances business goals with the practical realities of today’s AI ecosystem. If you’re beginning your planning cycle, we’d love to connect and help pressure-test your approach.