# AI strategy in 2026: what's changed and what still matters
Most organisations that invested in AI over the past two years have something to show for it. A chatbot. A dashboard. A pilot that worked in the demo. What they often do not have is a business that operates meaningfully…
Published: 2025-01-06
Author: Rodan Analytics
Most organisations that invested in AI over the past two years have something to show for it. A chatbot. A dashboard. A pilot that worked in the demo. What they often do not have is a business that operates meaningfully differently because of AI.

That gap - between deployment and impact - is the defining strategic problem of this moment. Not capability. Not cost. Not access to models. The problem is that most AI strategies were built for 2023: optimistic, technology-led, and structured around the question "what can AI do?" rather than "what does our business need to do differently?"

The landscape has shifted. Models are cheaper, faster and more capable. Agentic systems are moving from research projects to production. The organisations pulling ahead are not the ones with the most AI tools - they are the ones that made deliberate decisions about where AI changes the economics of their business and built toward that.

This article sets out what has genuinely changed, what has not, and how to think about AI strategy if you are making decisions now rather than in 2023.

## The capability curve has moved further than most strategies have

Twelve months ago, the dominant use case for enterprise AI was augmentation: helping people do existing tasks faster. Document summarisation. Code generation. First-draft content. These use cases are real and they still matter, but they are no longer the frontier.

The frontier is autonomy. Agentic AI systems - where models plan, take actions, call tools and operate across multi-step workflows without human input at each stage - are now viable in production environments. This is not theoretical. Businesses are deploying agents that handle end-to-end procurement queries, run continuous competitive monitoring, generate and QA financial reports, and manage customer escalation workflows with human review only at defined exception points.

The strategic implication is significant. Augmentation improves individual productivity. Autonomy changes organisational structure. A team of five analysts supported by AI agents does not just produce the same output faster - it produces output that a team of twenty could not previously have produced at all, because the bottleneck was not talent, it was throughput.

If your AI strategy is still framed around productivity uplift, you are optimising for 2023 economics. The question your competitors are asking - and the one worth asking now - is which functions in your business change structurally when AI can act, not just advise.

## Data infrastructure is still the rate-limiting factor

This has not changed. It will not change. Every engagement Rodan runs into a new organisation surfaces the same underlying issue: the AI ambition is real, but the data foundation cannot support it.

The specific failure mode has evolved, though. In 2022 and 2023, the common problem was fragmented data - siloed systems, inconsistent definitions, no single source of truth. That problem persists. But a new one sits alongside it: organisations that have consolidated their data into a warehouse or lakehouse, but have not structured it in a way that makes it accessible to AI systems at runtime.

Consider a mid-market retail business with a clean Snowflake environment and a well-maintained product catalogue. They want to deploy an AI agent to handle supplier queries - checking stock positions, cross-referencing lead times, flagging pricing anomalies. The model capability is not the blocker. The blocker is that the data schemas were designed for human analysts running SQL queries, not for an AI agent making decisions in real time against ambiguous natural language inputs. The agent either hallucinates answers or returns nothing useful.

This is a solvable problem, but it requires deliberate work. Organisations that are serious about agentic AI in the next eighteen months need to audit their data environments not just for completeness, but for AI-readiness: is the data structured, labelled and accessible in a way that supports automated reasoning? That is a different bar from BI-readiness, and most organisations are not there yet.

## Strategy has to be vertical, not horizontal

The era of horizontal AI strategy - "we will use AI across the business" - is over, not because it was wrong, but because it is no longer differentiated. Everyone is doing it. The organisations building durable advantage are making specific bets in specific functions where AI changes the unit economics of value creation.

What this looks like in practice: a private equity-backed business services firm identifies that its margin is being compressed by the cost of manual reporting across its portfolio companies. It does not deploy AI broadly. It builds a focused capability around automated commercial reporting - pulling from its portfolio's ERP and CRM systems, generating board-ready insight packs at a cadence that was previously impossible without a team of analysts. The AI application is narrow. The commercial impact is structural.

The discipline required here is prioritisation, which is harder than it sounds when every function head has a use case and the board wants to see AI progress. A useful forcing question: where in your business does speed of insight or decision translate most directly into revenue or margin? Start there. Build depth before breadth.

For ecommerce and technology businesses, that often means customer and demand intelligence. For PE-backed businesses, it often means portfolio performance visibility. For mid-market operators, it frequently means automating the commercial reporting layer that currently consumes disproportionate senior time with low-value assembly work.

## Build versus buy has a different answer now

In 2023, the default advice for most organisations was: do not build, integrate. The foundation models were expensive to run, the tooling was immature, and the capability gap between a custom solution and a well-configured off-the-shelf product was large.

That calculus has shifted. Foundation model costs have fallen by an order of magnitude in two years. The open-source ecosystem - tooling for orchestration, memory management, tool use and evaluation - has matured rapidly. And the limitations of generic AI products are now visible: they are built for the median use case, not for the specific data environment, workflow logic and decision context of any particular business.

The current build-versus-buy decision should be structured around three questions:

1. Does this use case require access to proprietary data that a generic product cannot reach?
2. Is the workflow specific enough that a horizontal tool will require so much configuration it is effectively a build anyway?
3. Does the capability, if done well, create competitive advantage - or just operational parity?

If the answer to any of these is yes, a purpose-built or heavily customised solution will almost certainly outperform a packaged product over a two to three year horizon. The upfront cost is higher. The long-term economics are better, and the strategic optionality is preserved.

## What still matters, regardless of the year

Amid all the change, three things have not moved.

Clear problem ownership matters. AI initiatives that lack a senior business owner - someone accountable for the outcome, not the technology - fail at a rate that is not worth the investment. The failure mode is not technical. It is organisational.

Change management matters. Deploying AI into a function without engaging the people in that function creates resistance, workarounds and shelfware. The organisations getting value are the ones treating AI deployment as an organisational change programme, not an IT project.

Evaluation matters. You cannot manage what you do not measure. AI systems degrade, drift and produce inconsistent outputs. Production AI without a measurement framework is not a product - it is a liability.

These are not new principles. They are the same principles that determined whether data warehouse investments paid off in 2010, whether cloud migrations delivered value in 2015, and whether digital transformation programmes worked in 2019. The technology changes. The organisational conditions for success do not.

## The cost of waiting is no longer theoretical

A year ago, a reasonable position was cautious experimentation: run pilots, build understanding, wait for the technology to mature. That window has closed. The organisations that ran pilots in 2023 and 2024 and converted them into production systems now have compounding advantages - in capability, in data, and in organisational fluency with AI.

The risk in 2026 is not moving too fast. It is arriving late to a capability race where the leaders have already built the data infrastructure, the institutional knowledge and the deployed systems that take months to replicate.

If you are reviewing your AI strategy now - or building one for the first time - the place to start is not the technology. It is a clear-eyed assessment of where AI changes the economics of your specific business, what your data environment can currently support, and what the gap is between the two.

Rodan runs a focused diagnostic engagement designed to answer exactly those questions. It is a £1,500 fixed-fee piece of work, delivered in two to three weeks, that gives you a prioritised AI opportunity map and an honest assessment of what your infrastructure can currently support. If that is useful, [get in touch to book a diagnostic](https://rodan.io).
HTML: https://rodan.io/insights/ai-strategy-in-2026-whats-changed-and-what-still-matters