It’s time for a serious conversation about AI

“We need to do more with AI, or we’ll be left behind.”

You have probably heard some version of that in a meeting, at an event or from a supplier. It captures the pressure surrounding AI: move quickly, demonstrate progress and avoid being the organisation that missed the opportunity.

But it leaves an important question unanswered. What, exactly, are you trying to improve?

If your starting point is adopting AI, you have already chosen a technology before defining the problem. You would question that approach in a finance system replacement or a major infrastructure investment. AI deserves the same scrutiny.

You can recognise its potential, explore it enthusiastically and still expect a clear explanation of how it will help your organisation.

Finding the value behind the noise

Your inbox and LinkedIn feed offer no shortage of AI experts, demonstrations and predictions. Alongside useful advice, you encounter sweeping claims about transformation and polished content that says surprisingly little. It can become difficult to separate expertise in using a tool from expertise in improving a business.

You should expect more than a demonstration of what a model can generate. Useful advice should connect the technology to your operating model, your priorities and the constraints your people work within. It should also explain where another approach might serve you better.

AI does work. There is credible evidence of it's value. An early study involving more than 5,000 customer-support agents found that an AI assistant increased issues resolved per hour by around 14%, with larger benefits for less experienced staff. That is a meaningful improvement in a defined operational activity. (nber.org)

So this does give you a reason to investigate similar opportunities. It does not however, give you a business case for deploying AI across every department.

That distinction matters. Your investment decision depends on whether the technology improves a particular activity under your conditions, at an acceptable cost and level of risk.

Faster work is only part of the equation

You can produce a document faster with AI. Whether that represents an improvement depends on the document.

If you spend twenty minutes generating a report and another hour correcting unsupported conclusions, the initial speed tells you very little. If the recipient then has to unravel vague recommendations or check invented details, you have transferred work elsewhere.

You may have encountered the result already: material that looks finished before the thinking is finished.

Even where the time saving is real, you still have a management decision to make. What will the released capacity allow your team to do?

You might respond to enquiries sooner, improve the quality of advice, reduce a backlog or avoid additional recruitment. Those are worthwhile outcomes. But “an hour saved” remains an incomplete benefit until you understand how that hour changes performance.

Your measures should reflect the intended outcome. If your objective is better service, measure service. If it is lower cost, include checking, implementation and ongoing support. Counting prompts or licences will tell you little about either.

Put AI within your wider strategy

Your business strategy should explain what the organisation is trying to achieve. Your IT strategy should identify how technology, information and capability can support those ambitions.

AI belongs within that conversation.

If growth is constrained by slow quotation turnaround, examine the quotation process. If staff cannot find reliable information, understand how that information is stored, maintained and accessed. If reporting takes too long, investigate the data and the manual work behind it.

You may find a valuable role for AI. You may also find that your existing software has unused functionality, that two systems need connecting or that nobody owns the process.

Without that understanding, you risk placing a new tool over an unresolved problem. You then inherit another subscription, another source of information and another technology dependency to manage.

Your priorities should determine where AI receives attention and investment, alongside other opportunities to improve the organisation.

Look beyond the price of the licence

A relatively inexpensive subscription can make an AI initiative appear straightforward. However, your actual commitment includes the effort required to make it useful: preparing information, configuring access, connecting systems, training staff, changing processes and checking outputs. Someone also has to own the service once the initial enthusiasm passes.

Before committing, establish who will maintain the information, resolve problems and assess whether the results remain acceptable. Consider how the tool fits your existing systems and what happens if the supplier changes its product or pricing.

You should apply the same ownership and procurement disciplines expected elsewhere, with additional attention to AI’s particular characteristics.

A convincing answer can still be wrong. Sensitive information can be mishandled. An automated action can have consequences beyond the screen. Your controls should reflect those consequences, with clear limits and appropriate human judgement.

Experiment with a purpose

You do not need a perfect strategy before allowing anyone to experiment. Small trials can help you understand capabilities, build confidence and discover useful applications.

Give each trial enough structure to produce a decision.

Choose a recognisable problem. Record how the work is done today. Agree what improvement would justify continuing, and include the effort required to check and correct the output. Then test it with the people who do the work, using realistic examples and appropriate information.

You might discover that AI performs well on routine cases but struggles with exceptions. You might find that staff value it for preparation but prefer to retain control of the final output. You might conclude that a simpler workflow delivers the same benefit.

Each finding is useful. Your trial has succeeded if it supports a better decision, including a decision to stop.

Expect a more serious conversation

You should feel able to challenge an AI proposal without being labelled resistant to change.

Ask what problem it addresses, how it compares with alternatives and what evidence would justify scaling it. Expect anyone advising you to understand your organisation as well as the technology.

You can be ambitious about AI while remaining selective about where you use it. Your strongest opportunities will come from connecting its capabilities to work that matters, with clear ownership and measurable outcomes.

The next serious conversation about AI should begin with your business priorities. The technology decision can follow next.