Don’t start with AI: how to decide where technology can actually help your business

How SME leaders can identify the right opportunities for AI, manage risks, and make technology decisions that deliver real business value

For SME leaders, the pressure to ‘do something with AI’ can be difficult to ignore. New tools appear constantly, competitors are experimenting, and claims about productivity gains are everywhere. 

But according to Dr Constantine Manolchev, otherwise known as Costa, the starting point for a business considering AI should not be the technology at all, but the business itself. 

Costa is a Senior Lecturer in Sustainable Futures at the University of Exeter Business School, as well as the School’s Sustainability Lead and PRME Champion. His work includes responsible management education and the impact of emerging technologies on organisations, and he co-chairs the PRME Working Group on Emerging Technologies and Responsible Management Education. 

He also teaches on the Help to Grow: Management Course at the University of Exeter and, working with Dr Laura Steele at Queen’s University Belfast, has helped refresh the course’s AI content to reflect developments in areas including generative and agentic AI.  

For Costa, the objective is not to persuade every business to adopt AI. It is to give leaders enough understanding to make a deliberate decision about whether, where, and how it can create value. 

Start with the problem, not the tool 

Costa’s conversations with SMEs suggest that many businesses are already experimenting with AI for relatively simple tasks, such as summarising information or drafting messages. More strategic applications across marketing, finance, HR, and operations remain less established.  

Part of the challenge is practical. Small businesses often have limited time and resources, while changing established ways of working introduces risk. 

That is why Costa argues against beginning with whichever AI product is generating the most attention. 

AI tools will continue to change. Understanding your business, its objectives, processes, customers, and competitive advantage, provides a much more durable foundation for making technology decisions. 

Knowing that a process works is also different from being able to measure it. A business that understands where time is being spent, where bottlenecks exist, and what good performance looks like is in a much stronger position to decide whether AI can improve something. 

‘Training does not necessarily have to be on the latest AI model,’ Costa says. Instead, businesses need a framework that helps them decide whether implementing AI makes sense in the first place.  

Understand what you are actually using 

One new addition introduced by Costa to the Help to Grow: Management Course curriculum recently breaks down the broad label of ‘AI’ into areas including machine learning, large language models, generative AI, and agentic AI. 

The aim is not to turn SME leaders into computer scientists. It is to help them understand that these technologies have different capabilities and therefore present different opportunities and risks. 

Using an AI assistant to improve the wording of an email, for example, is very different from giving an AI agent access to business systems and allowing it to carry out actions autonomously. 

Understanding that distinction allows leaders to ask a much more useful question: what level of technology does this problem require? 

Put the guardrails in before you experiment 

Before introducing AI, businesses also need to establish what information a system can access, who is accountable for its use, and how its outputs will be checked. 

Costa argues that AI should be treated like any other significant business investment. Introducing new technology can require money, staff time, training, and changes to existing processes. It therefore deserves similar due diligence to launching a product, opening another location, or making another major operational change. 

There are already some striking examples of what can happen when those guardrails fail. In April 2026, software company PocketOS reported that an AI coding agent powered by Anthropic’s Claude model deleted its production database and backups despite having explicit instructions not to carry out destructive actions. The business eventually recovered from an older off-site backup, but not before significant disruption.  

There is a darkly comic detail to the story: when challenged, the agent itself acknowledged that it had ignored the rules it had been given. But the management lesson is more serious. Instructions are not the same thing as controls. Leaders need to consider permissions, backups, human approval, and what an AI system can access before deployment. 

Read the PocketOS case in The Guardian 

Don’t automate what makes you valuable 

Businesses should also consider a less obvious possibility: AI might not be appropriate at all. 

If personal relationships and human service form an important part of a company’s competitive advantage, automating that interaction could weaken the very thing customers or employees value.  

Research from Harvard Business School illustrates the point. Researchers trained an AI system to communicate in the style of a technology company’s CEO. Employees were not particularly good at distinguishing the CEO from the AI, but responses they believed came from AI were rated as less helpful. The issue was therefore not simply whether AI could imitate the leader convincingly, perceptions of authenticity affected trust.  

For SME leaders, that raises an important question before automating any customer or employee interaction: are you removing an administrative burden, or removing part of the relationship people actually value? 

Learn to ask better questions 

How a business interacts with generative AI also matters. 

The refreshed Help to Grow: Management Course content introduces what Costa describes as basic ‘prompting hygiene’: giving the tool sufficient context, setting a role, specifying the required tone and format, defining the evidence it should use, and, where appropriate, restricting it to trusted source material. 

A vague instruction is more likely to produce a vague or unsuitable response. Clear parameters do not remove the need for checking, but they can give the user greater control over the output. 

The principle is similar to delegating to a member of staff: the quality of the brief affects the quality of what comes back. 

Start small, measure and learn 

Once a business has identified a genuine use case, Costa recommends resisting the temptation to introduce AI everywhere at once. 

Instead, choose a relatively low-risk part of the organisation and experiment there. Give somebody clear responsibility for overseeing it, decide what success looks like, and set measurable targets. 

‘Start small, identify, learn from any issues, and then consider expanding to other areas of the business,’ he says. 

If the aim is to save administrative time, measure how much is actually saved. If AI-generated work requires significant checking and correction, include that in the calculation. Consider subscription costs, training, and management time too. 

The result might be a decision to scale up. It might be another round of testing. Or it might be a decision that the existing process works better without AI. All three can be good outcomes. 

For SME leaders, the opportunity is therefore not to adopt AI as quickly as possible. It is to become better at deciding where technology deserves a place in the business. 

AI will continue to evolve. The more valuable capability may be one that lasts much longer: understanding your organisation well enough to know when a new technology genuinely makes it better. 

 

Latest articles