AI Adoption in SMEs

AI will have a non-trivial but modest impact on the economy in the next ten years

There have been many quite bombastic and headline grabbing claims made about the onset of AI, with thoughts ranging from its potential to be the single most important driver of economic growth in the next decade, to those that believe AI will develop a sense of human-like sentience, and to those that go further and predict a doomsday scenario of the ‘rise of the machines’ and the subjugation of humans to robots.

However, some academics are beginning to take a rather more modest view. Daren Acemoglu, a professor of Economics at MIT and famous author of “Why Nations Fail”, published an article (“The Simple Macroeconomics of AI”) in 2024 in which he cited that only 5% of tasks in the US economy will be able to be profitably performed by AI in the next 10 years, amounting to around 1% of GDP growth over that period. Growth from AI driven productivity in the US will only be 0.7%.

He also cites some concerns about AI’s impact on welfare, such as its impact on mental health. Others have referred to this as “AI psychosis”, where there is a growing trend of people perceiving AI to be conscious, and therefore beginning to develop addictions, worsening paranoia and delusions, including the development of romantic relationships with AI, the sudden withdrawal from social life, or believing that AI is deliberately trying to harm them. The modest economic gains from AI, though non-trivial, will be combined by potential welfare declines.

95% of companies have failed AI projects

Interestingly, Daren Acemoglu also cites the disparity between the concentration of AI investment in large firms, and the fact that the majority of tasks in the economy that could be replaced by AI reside in small and medium sized businesses, where access to capital and resource is very limited.

Another paper published recently in August 2025 by MIT’s NANDA initiative conducted 150 interviews with business leaders, surveyed 350 employees and analysed 300 public AI project deployments. Its conclusions were quite revealing, concluding that 95% of companies in the dataset experienced failed AI projects where they were not able to proceed beyond the prototype phase. They also noted, however, that this was not because of AI’s current capabilities, but rather down to the challenge of enterprise technical debt, and the ability to integrate with new technologies and to adapt to new workflows.

The 5% succeeding are predominantly startups that have identified a niche and accelerated revenue growth by partnering with the right companies. The majority of failed projects were those that tried to develop ‘in-house tools’ rather than partnering with vendors. Those that succeeded were also good at ensuring wide decentralised adoption of AI, rather than centralised AI hubs.

Where is AI spend going?

An article published by Andreessen Horowitz in their a16z enterprise substack earlier this month provided the results of a study they did with Mercury (a fintech) on 200,000 customers, looking to identify the top 50 AI -native application layer tools that companies were spending money on, excluding spend on infrastructure and cloud services.

Their most interesting finding was that 60% of applications were ‘horizontal’, focused on boosting productivity across a company, while 40% were vertical, targeting a specific role or function. Horizontal applications included general LLM use, such as tools like ChatGPT, Anthropic and Perplexity, or those relating to workspaces such as Notion or Manus, or notetaking such as Fyxer or Happyscribe. In the vertical categories, companies were using AI for customer services (e.g. tools like Customer.io or Ada), sales (with tools like Instantly and Clay), or HR (tools included Micro1 and Metaview).

So what does this all mean for SMEs

At EPIC, our internal Data & Automation team is advising our portfolio companies and clients on the following:

1. Pick your use cases wisely, and spend on AI carefully. Consider whether you prefer a ‘horizontal’ or ‘vertical’ approach to AI. The former is likely to cultivate AI adoption widely across your firm and serve as a stepping stone to empowering employees to upskill with AI. If you opt for the latter (vertical) approach, you may need to consider whether you need to reorient your entire process to integrate with new designed workflows.
2. Technical debt. The biggest barrier to successful adoption of AI has been legacy technology. In SMEs, this problem will be more pronounced, where resource constraints may have meant years of under-investment in upgrading systems and technology. This may finally be the time for change.
3. Data.  The underlying commodity that underpins the success of AI is data. The common adage is “garbage in, garbage out”. Data that has been inconsistently captured and KPIs that have been poorly defined will significantly hold you back from success in AI. You will need an accelerated journey to get this back on track.
4. Business partnering. If you have an internal tech or IT team, encourage them to upskill on AI first, and invite them to work closely with wider teams and staff to roll out AI use cases. Where you don’t have that resource, partner with vendors that have tangible use cases specific to your business context.
5. Privacy. Most AI tools now offer enterprise versions of their technology which ringfence your data in a private environment that is not made available for wider training purposes. Make use of these versions and caution against potential sharing of confidential information on the consumer versions of these platforms.
6. Capital: Whilst the majority of admin tasks and productivity opportunities lie in small and medium sized businesses, they do not always have the capital to unlock these gains. Where you believe these gains to be quite significant differentiators for you, consider your overall growth strategy and potential need for capital injection to roll out investment in technology and AI.