AI is rarely the starting point

Artificial intelligence is opening new possibilities for organisations. Tasks that used to take several hours can be done in minutes, scattered information can become more accessible, and activities that depended entirely on human intervention can now benefit from different levels of automation.

It is natural that these possibilities attract the interest of companies. But there is an important difference between recognising the potential of a technology and understanding where it can actually add value.

Many initiatives begin with an apparently simple question: where can we apply artificial intelligence? Legitimate as it is, this approach can lead organisations to look for problems that justify using the technology, rather than starting from the difficulties they need to solve.

The question should perhaps be a different one: which problems, decisions or activities do we want to improve, and what part might technology play in that change?

When the technology becomes the objective

The pressure to bring artificial intelligence into organisations is understandable. The technology is moving fast, new applications appear constantly, and there is a legitimate concern about missing opportunities to stay competitive.

In that context, starting an AI project can be read as a sign of innovation. But building a technically interesting demonstration does not necessarily create a meaningful improvement for the organisation.

A pilot can show that a given task can be automated, or that a model produces satisfactory results under controlled conditions. Yet when the time comes to fold that capability into day-to-day operations, difficulties can appear that were not considered at the outset.

Who is going to use the solution? Which process will have to change? Who will be accountable for the results? How will exceptions be handled? And what concrete benefit is expected?

When these questions go unanswered, the project can end with a successful demonstration and no operational life beyond it.

This does not mean exploratory projects have no value. Trying out a technology to understand what it can do is a perfectly legitimate decision. What matters is telling apart an experiment meant to produce learning from an investment meant to produce operational results.

They are different objectives and they should be judged differently.

Before experimenting, understanding what we want to change

An artificial intelligence project should begin with a sufficiently clear understanding of the situation it sets out to improve.

That understanding does not necessarily require an extensive diagnosis. Often it is enough to watch a process, talk to the people involved, and identify where the effort, the difficulty or the loss of value is concentrated.

There are three questions I find particularly useful at this stage.

The first is to work out what we want to start happening differently. It may be reducing the time needed to prepare a given piece of information, improving the consistency of an analysis, or freeing a team to concentrate on higher-value work.

The second is to understand what will have to change in the organisation if the experiment works. A technology solution does not exist in isolation. It needs people who use it, processes that absorb it, and responsibilities clear enough to keep it running.

The third is to decide how we will judge the result. Before starting, it matters to identify what evidence will tell us whether the experiment deserves to continue, needs adjusting, or should be abandoned.

These questions help keep enthusiasm about technological possibility from replacing the discussion about what actually matters to the business.

Where can artificial intelligence add value?

There are situations where artificial intelligence can represent a particularly interesting opportunity.

One is when people spend a significant part of their time gathering, organising and interpreting information scattered across different sources. Another arises when certain activities require recurring analysis, comparison of documents, or the consistent application of criteria.

There may also be opportunities in communication tasks, content production, research support, pattern identification, or preparing information for decisions.

In some of these cases AI can reduce operational effort. In others it can make information more accessible, or make it possible to explore options that would previously have taken too long.

But the existence of a technological opportunity does not mean that implementing it is automatically worthwhile.

The quality and availability of the information, the frequency of the activity, the costs of implementation and upkeep, the risks attached to the results, and the organisation’s capacity to absorb the change all have to be weighed.

There is also an important distinction between supporting a decision and automating it. A tool that organises information and presents different perspectives can create value without replacing the judgement of the person deciding. In some situations that is a more suitable approach than trying to automate the process end to end.

The aim should be to find the combination of technology, human knowledge and work organisation that produces the best results.

The weight of information and processes

One of the obstacles to implementing artificial intelligence may lie not in the technology but in how the organisation manages its information.

The data needed may well exist, but be spread across different systems, used with different meanings, or depend on informal procedures to be interpreted at all.

In those circumstances, introducing a new tool can make problems that already existed more visible without necessarily solving them.

Imagine a company that wants to use AI to produce automatic analyses of client profitability. If different teams use different criteria to allocate costs or classify revenue, the main difficulty is not only the model’s ability to interpret the data. It is also the consistency of the information it is given.

The answer may be to clarify concepts, revise processes or improve how systems talk to one another before moving on to a more sophisticated solution.

This does not mean waiting for perfect data. Many experiments can start with limited information, provided its limitations are known and compatible with the intended purpose.

The essential question is whether the information available can produce results useful and reliable enough for the context in which they will be used.

Experimenting in proportion to the problem

The speed of experimentation is one of the most interesting opportunities today’s technologies offer.

In many cases it is no longer necessary to build a complete application to find out whether an idea has potential. You can start with existing tools, a limited set of data and a small group of users.

But experimenting quickly does not mean experimenting without judgement.

The scope should be small enough to allow learning on a limited investment, and representative enough to reveal the difficulties that would appear in real use.

If the aim is to support a monthly activity, at least one full cycle may need to be followed. If use is daily, a shorter period may be enough to gather what matters. The right duration will depend on the nature of the process, the risks involved and what is being validated.

It also matters to involve the people who may use the solution from early on. An experiment run entirely by a technical team can demonstrate interesting capabilities while revealing little about how the tool would fit into everyday work.

The success of a pilot should not be measured only by its ability to produce a technically correct result. It should also take in how useful that result is, how easy the tool is to use, the effort of integration, and what would be needed for it to continue.

And sometimes the most valuable result of an experiment is discovering that the solution first imagined is not the right one.

When the best solution needs no AI

Not every organisational problem calls for artificial intelligence. Some can be solved by defining responsibilities more clearly, simplifying a procedure, or making better use of the tools that already exist.

A team losing time reconciling information may need two systems to talk to each other. A process that depends on repeated sign-offs may benefit from a review of the approval rules. And a recurring difficulty in how information circulates may come from responsibilities that were never defined clearly enough.

In those cases, adding a more sophisticated layer of technology can raise cost and complexity without removing the source of the problem.

It is also possible that conventional automation is simpler, more predictable and cheaper than an AI-based solution.

The decision should come from comparing the alternatives, weighing the expected benefit, the investment required, the risks and the capacity to implement.

This requires a degree of independence in the analysis. Once an organisation has decided it wants an AI project, it can become harder to recognise that a simpler solution exists.

But identifying that alternative is not a failure of innovation. It may be precisely the more intelligent use of the resources available.

Starting from the problem, staying open to discovery

Arguing that artificial intelligence should rarely be the starting point does not mean dismissing technological exploration.

There are situations where learning about a new capability lets us see opportunities we could not previously imagine. Experimentation can therefore come before a specific problem has been identified, particularly when the aim is to learn and explore.

What matters is not to confuse that discovery with the justification for an operational investment.

When we set out to improve a concrete activity, it makes sense to start by understanding how it works, what difficulties it presents and what results we are after. Technology enters that thinking afterwards, as one of the possible options.

When we set out to explore the potential of a new technology, we should own that intention and define what we want to learn from the experiment.

Both approaches can create value, as long as their objectives are clear.

At Aetheris, what interests us most is understanding how organisations can use technological possibility to develop new capabilities, improve decisions and solve real difficulties.

Sometimes that will involve artificial intelligence. In other situations the answer lies in a change of process, a better organisation of information, or a simpler technological solution.

Technology should widen an organisation’s capacity to solve problems, not become one more problem the organisation has to manage.