Analysis Became Cheaper. Mistakes Became More Expensive.

The cost of making a decision can fall.

The cost of making the wrong decision can rise at the same time.

Two developments in recent days bring precisely this possibility to mind.

On one side, major technology companies are promoting AI agents that do more than provide answers in the workplace: they perform tasks. We are now talking about systems with their own access permissions that open files, search data, prepare reports, manage calendars and even trigger other tasks. In other words, AI is gradually moving from a corporate “tool” towards something resembling a “colleague”.

On the other side, a substantial European fine for providing incorrect and misleading information during a merger review has brought an old, unwelcome truth back to the table: speeding up a process is not the same as transferring responsibility.

The central issue is not technology.

It is this: when analysis becomes cheaper to produce, why does accuracy become more valuable?

Management teams once had a limited number of scenarios placed before them. An investment decision, an acquisition thesis, a budget revision, a financing alternative… Each required time, a team, the work of collecting data and significant coordination. Analysis was expensive. In a sense, that also served as a natural filter.

Now that filter is thinning.

The same team can examine far more scenarios in far less time. Sales forecasts, pricing assumptions, demand segments, inventory plans, loan repayment capacity and target-company synergies can all be calculated almost instantly. At first glance, this looks like a productivity story. It is. But that is not the whole story.

Because cheaper analysis increases two things at once:

first, the capacity to make decisions;

second, the risk of generating false confidence.

This is where the problem begins.

Building a model faster does not make its input data more accurate. Nor does a traceable process make its results economically sound. Knowing which file was used is one thing; knowing whether a number in that file represents commercial reality is another.

In finance, errors rarely look dramatic at first. They start small. A misclassified customer, an optimistic collection assumption, an outdated cost curve, an idle factory modelled as though it were operating close to full capacity… In Excel, these are cells. In real life, they can become a credit limit, an investment decision or an acquisition price.

That is why the challenge in the AI era is not simply producing “more analysis”.

It is distinguishing which analysis deserves our trust.

Many management teams today are interested in how much work AI can perform within the company. I think a better question would be:

Which AI-generated outputs can the company use directly, and where must human judgement remain mandatory?

Automation built without this distinction may save time while also expanding the scale of errors. A system does not only accelerate sound work; it can scale broken processes too.

This is where the issue becomes particularly critical in Türkiye.

Globally, the question is often: “How far will AI transform the way we work?”

For many companies in Türkiye, a more fundamental question remains: what balance sheet will we use to finance that transformation?

Recent data show that the economy is not moving uniformly. There are signs of strength in exports and a recovery in capacity utilisation in some sectors. At the same time, manufacturing remains weak, confidence cautious, financing costs high and cash pressure unevenly distributed across companies. Not every company faces the same pressure, and not every sector is equally constrained. But borrowing costs, working-capital needs and the demand outlook are distinguishing viable investment decisions from unviable ones far more ruthlessly than before.

In this environment, AI investment is too complex to be justified by “we will need it in the future”.

An AI project may create a real competitive advantage for one company. For another, the same project may simply add another burden to an already expensive balance sheet. Returns depend on more than how intelligent the software is. How clean is your data? How standardised are your processes? Who will make the decision? How often will outputs be validated? What will an incorrect result cost financially? More importantly, would the same capital generate a higher return today if allocated to inventory optimisation, faster collections, an export channel or debt reduction?

That is the real capital-allocation question.

Globally, AI investment is often described in terms of growth, productivity and new business models. For many Turkish companies, the question translates differently:

Will this investment actually release cash, or merely create an expense that looks more sophisticated?

Seen through the lens of Türkiye, AI is therefore more a financial question than a technical one.

Limited AI adoption in Türkiye cannot be explained simply by reluctance towards technology. A shortage of expertise, legal uncertainty, data-security concerns and integration difficulties all matter. But a deeper problem remains: many companies have not yet brought sufficient discipline to the way they produce data and make decisions. Expecting AI to work miracles on that foundation is like speeding up a broken clock. There is more movement, but no more accuracy.

Now consider M&A.

I expect merger and acquisition processes to scrutinise not only financial statements but also the data architecture that produces them more closely. Particularly in companies using AI-supported operations, pricing, risk assessment and reporting, buyers will ask more than “What is the number?” Questions such as “How was it produced?”, “What data supports it?”, “Who validated it?” and “Where did human approval enter the process?” will become increasingly important.

This could affect due diligence not only technically but also in terms of valuation. Sound data governance and well-designed validation mechanisms can increase confidence in a company. Conversely, AI adoption can become an additional risk rather than a sign of modernity.

In other words, AI may support a higher valuation multiple for some companies and justify a discount for others.

The difference will lie in decision architecture, not the level of technology.

I do not think the winners in the period ahead will be the companies that use the most AI.

The companies that understand how much AI to use in each decision will stand out.

Not every decision suits automation. Not every data point is reliable. Not every analysis should lead to action.

In some areas, speed is a major advantage.

In others, the ability to brake is more valuable.

Finance understands this well: liquidity is not the same as value; revenue is not the same as profitability; abundant data is not the same as good decisions.

Today, AI enables us to generate more options.

But a company’s real strength emerges not in how many options it produces, but in knowing which to eliminate.

Ultimately, the defining question of this new era may not be “How much can AI do?”

The more critical question may be:

Which truths can a company identify within analysis that has become cheaper to produce?

Decisions have become faster.

The challenge now is to ensure judgement does not become cheaper.