What to expect from AI in the enterprise in 2025

This year saw the initial hype and excitement over AI settle down with more realistic expectations taking hold. This is particularly true with enterprise deployments as the capabilities of existing models, coupled with the complexities of many business workflows, led to slower progress than many expected. Central to this is a realization among many corporate users that there’s no I in AI — so far anyway. Large language models (LLMs) are very good at spotting patterns in data of all types, and then creating artefacts in response to user prompts that match these patterns. But this isn’t intelligence in any human sense. The propensity of LLMs to make up plausible looking but inaccurate information is evidence of this.

Despite these limitations and concerns among CIOs over AI costs, real progress has been made this year and we can expect to see this grow further in 2025. I see this taking shape in 5 key areas.

Augmenting employees, not replacing them

Whether it’s through cutting costs, innovating new products and services or improving the customer experience, building a competitive advantage is at the core of most technology deployments, and AI is no different. However, the wide availability of LLMs, open and closed, and the tools to deploy them means AI is available to all organizations. Like the PC revolution of the 80s and 90s, and the rise of cloud computing and SaaS in the early 2000s, when everyone has access to the same tools, it’s the way they’re used that confers a competitive advantage.

With AI, this means augmenting your existing skills base and leveraging your human assets. Businesses that see AI as a replacement for skilled and experienced workers will go down the wrong path. Employee knowledge of their companys’ products, processes, and the markets they operate in and customers they sell to is often uncoded and tacit. Assuming a technology can capture these risks will fail like many knowledge management “solutions” did in the 90s by trying to achieve the impossible. Michael Hobbs, founder of the isAI trust and compliance platform, agrees. “You can get answers fast from gen AI systems,” he says. “But CIOs need to ask if these are good answers, and if I’m growing the skills base within my organization, augmenting it with tools, or fundamentally reducing it?”

If the concerns about LLMs reaching saturation are correct, we can expect diminishing returns from every additional GPU that’s used in creating new models. In this scenario, using AI to improve employee capabilities by building on the existing knowledge base will be key.

Focus on data assets

Building on the previous point, a company’s data assets as well as its employees will become increasingly valuable in 2025. Foundation models (FMs) by design are trained on a wide range of data scraped and sourced from multiple public sources. The scale of this training makes them capable of providing answers to general questions, but limits their value to the specific requirements of most businesses. Retrieval augmented generation (RAG) provides a pathway to combining proprietary data with the capabilities of an LLM for more focused and relevant results. Forrester predicts that RAG services will become a key offering for most cloud providers in 2025, giving enterprises wider choice in vendors and, possibly, competitively priced offerings.

To benefit from this wider range of RAG services, organizations need to ensure their data is AI-ready. This involves the prosaic but essential activities of good information management: data cleaning, deduplicating, validating, structuring, and checking ownership. AI governance software will also become increasingly important in this process, with Forrester predicting spending on off-the-shelf solutions will more than quadruple by 2030, reaching almost $16 billion.

The sooner enterprises identify data assets from across the business, adopt a creative approach to how they might be used, and then get it in an AI-ready state, the sooner they’ll be able to take advantage of new RAG services coming down the line in 2025.

Controlling costs

According to Gartner, more than 90% of CIOs surveyed in 2024 believed that managing costs limited their ability to get value for the enterprise from their AI investments. Part of the solution, Gartner argues, is calculating how costs will scale before any widespread deployments are made. Failure to do so could mean a 500% to 1,000% error increase in their cost calculations. In 2025, we can expect to see better frameworks for calculating these costs from firms such as Gartner, IDC, and Forrester that build on their growing knowledge bases from proofs of concept and early deployments.

As AI offerings from cloud providers such as Microsoft Azure, AWS, and Google Cloud develop in 2025, we can expect to see more competitive pricing that could help keep a check on costs for enterprises. However, this will depend on the speed at which new AI-ready data centers are built relative to demand. McKinsey has calculated that global demand for data center capacity could rise at an annual rate of 19% to 22% from 2023 to 2030. Sourcing sufficient electricity to power these new centers will continue to constrain demand in 2025 and beyond.

Measuring AI ROI

As the complexity of deploying AI within the enterprise becomes more apparent in 2025, concerns over ROI will also grow. However, we might expect to see a more nuanced approach to calculating ROI in the coming year. Measuring the impact of new technologies from a financial and productivity perspective has been a challenge for many years. In 1987, Nobel prize winning economist Robert Solow famously quipped, “You can see the computer age everywhere but in the productivity statistics.”

This will continue to be true in 2025 as managers struggle to quantify the benefits of their AI investments. Part of the problem is the lack of common standards for measuring returns. Costs are relatively easy to calculate as they can be reduced to a dollar amount and compared to previous years. However, putting a value on the qualitative improvements in workers’ outputs from AI presents greater challenges. As with calculating scaling AI deployment costs, new frameworks will emerge in 2025 to help managers measure value from their investments. These will move beyond traditional KPIs, and need to incorporate measures such as customer satisfaction levels, improved decision making, and accelerated innovation processes.

Avoiding irrelevance

The transformative nature of the current wave of AI products threatens the business models of many enterprises in the same way that the internet undermined and then displaced companies like Blockbuster, Borders, and HMV. Clayton Christensen’s concept of the innovator’s dilemma explained how well run and successful businesses can be made obsolete by new entrants that harness new technologies and business practices in innovative ways. This year we saw online education giant Chegg lose 99% of its market value, or $14.5 billion, after students switched to ChatGPT free help for homework, rather than pay $19.95 a month for a subscription service.

We can expect to see similar but perhaps not so dramatic examples appear in 2025. These will be across a number of sectors including marketing, publishing, entertainment, and education in both B2C and B2B environments. Chegg’s misfortune should be a wake-up call to all businesses, but can also be seen as an opportunity for many. Scenario planning should be a priority along the lines of a SWOT analysis, which is a good starting point: what strengths does your business have that can capitalize on the benefits of AI, and how might external opportunities and threats impact on these?

Next year is going to be challenging in many ways. From an enterprise perspective, AI-driven change is only going to accelerate, albeit slowly and steadily. The accessibility of so many models and their growing incorporation into existing applications means their availability to any business that wishes to adopt them. How they’re deployed and used to complement existing corporate strengths and data assets, as well as aligned to strategic objectives, will separate winners from the rest.