To deliver ROI on AI, start small before you scale

Over half of organisations today (65%) include AI in their corporate strategy – with the vast majority (88%) reporting regular AI use in at least one business function. Despite this, a recent MIT study found that 95% of enterprise AI initiatives fail. This is largely because AI models built for the masses aren’t easy to tailor, so the outputs are often generic and don’t meet specific business needs. This trend is creating fear in the industry that AI is overhyped, and it’s only a matter of time before the ‘bubble’ bursts.  

The truth is that AI has enormous potential but we’re misallocating it. Enterprises are giving all-knowing models the ‘keys to the kingdom’ and applying AI indiscriminately. Instead, they should start small, prove value, and then scale.  

To deliver ROI on AI now and into the future, organisations should look towards purpose-built use cases and help employees to use AI effectively in their day-to-day. Here’s how they can do that. 

Make sure AI doesn’t create more problems than it solves 

Leaders are investing heavily into broad AI models, which are costly to run, and require constant tuning and extensive integration work. They are also likely to make unpredictable leaps in logic because they link disparate data points together to provide answers. At worst, they may hallucinate, potentially creating nightmarish scenarios if integrated into broader business workflows. At best, these models produce polished spreadsheets, slide decks, and formatted reports that look like productivity but lack real substance – also known as AI ‘workslop’. 

To avoid these pitfalls, organisations need to go back to basics. They should ask themselves where AI is actually needed. Where are employees spending most of their time? Which time-consuming, rules-based tasks would benefit from automation? By answering these questions, organisations can ensure their AI deployments perform select functions exceptionally well rather than attempting to tackle everything at once.  

One way to look at it is to think of AI like a highly capable intern. It can quickly pull together information and get you most of the way there, but it still needs a careful review before anything goes out the door. Just as you wouldn’t trust an intern with your entire operations on day one, you shouldn’t deploy AI everywhere at once. The smartest approach is to start small and focused, prioritising low-risk tasks where the value is clear. These smaller deployments keep teams close to the work, where they can see what’s happening and step in with real-world expertise. 

AI’s success depends on how you use it  

Recent data has highlighted a persistent gap between AI use and AI literacy, with 59% of leaders believing there is an AI skills gap in their organisation.  

You can have the best AI tools in the world, but without enablement, they won’t deliver value for the business. Employees need to be able to identify the kind of repetitive, data-heavy tasks that are prime candidates for automation. In IT, this might mean walking technicians through how automation can simplify tasks like configurations, installations, monitoring, vulnerability management, patching, and endpoint management. In these use cases, automation can significantly reduce human error, improve consistency, and save time for more proactive work. 

Organisations can help each department to explore the same use-case specific approach. For instance, in HR, this could include payroll checks which identify unexpected salary changes, duplicate payments, or incorrect tax deductions. In finance, that could be tasks like cash flow forecasting, expense auditing, and categorisation. 

As well as understanding where AI is most useful, employees should also know where not to apply AI. Take the HR example. HR leads should understand at a basic level what types of employee data AI models are accessing. They don’t need to understand all the technical details, but they are responsible for working with IT teams to ensure sensitive or privileged information isn’t compromised. 

In order for employees to truly understand where they should or shouldn’t apply AI, organisations need to invest into prompt engineering training. The goal is for employees to think critically about AI-driven outputs and follow responsible-use protocols. Training initiatives won’t be static and will need continuous updating as AI evolves. 

From investment to enablement 

The question isn’t whether organisations should be investing in AI technologies. It’s whether leaders are doing enough to support employees to apply these tools in ways that are genuinely useful. The current market skepticism is an opportunity to take a pulse check and ask: how can we be more intentional about where to apply AI? What else can we do to enable employees to identify high-value use cases for these tools? The path forward won’t just be about keeping pace with AI innovation – it will require leaders to balance usability and security with wider business needs.  

If you decide AI is the solution before you define the problem, stop and think – because that’s hype.