When a company needs a report, a feature or a small application, does it still need to buy another piece of software? AI is making it faster and cheaper to build custom applications, challenging the economics of traditional enterprise software. Companies that once relied on expensive subscriptions and off-the-shelf solutions can increasingly create tools tailored to their own needs, using the data and systems they already have.

The shift became apparent during my recent conversations with executives from three companies in Lithuania. I sat down with leaders from Lithuanian Railways Group, international payments company TransferGo and software developer Helmes to understand how they are putting AI to work in their businesses.

Their experiences revealed something more interesting than another round of productivity gains. AI is changing how companies extract value from their data, identify opportunities and decide whether they need to purchase new software at all.

At Lithuanian Railways, an executive who once searched financial reports line by line now asks AI to identify the changes worth investigating. At TransferGo, an AI system uncovered an unexpected surge in money transfers from Iceland, a market nobody had been assigned to monitor. Executives at Helmes described how companies are using AI to build applications around their own business requirements rather than relying exclusively on commercial software. Beyond Lithuania, GoDaddy is testing whether applications built with its own AI tools can replace smaller software subscriptions in its corporate offices.

Together, these conversations raise a question for the enterprise software industry. If companies can build their own AI applications and extract more value from their existing systems, how much of the software they currently pay for will they still need?

A Better Question Than Another Dashboard

Irmantas Beržauskas, chief legal and organizational development officer at Lithuanian Railways Group, or LTG, has used business intelligence dashboards for years, and he is not dismissive of them. The company’s Power BI dashboards gather information from different systems and make it easier to see what is happening. His frustration comes from what comes after the dashboard.

An executive reviewing financial figures rarely wants to admire a chart. The questions are more pointed. Which expenses changed? Where did performance depart from the previous period? Is a discrepancy worth investigating, or does it reflect a routine accounting adjustment?

Beržauskas recalls examining the figures himself, working through the lines and comparing reporting periods. Now he gives the data to AI and uses simple prompts to direct the analysis. He uses AI for this work so often that he scarcely thinks of it as an AI application anymore. It has become part of how he reviews the numbers.

“What are the changes, what are the trends,” he asks.

Dashboards aren’t dead. LTG still finds its dashboards useful, but the change is that an executive can ask a question the dashboard was not designed to answer, without waiting for someone to construct another chart.

Software vendors have spotted the same opening. In June, Google introduced conversational agents in a preview version of Looker that let users ask questions from inside a dashboard and follow up on the answers. Google is putting conversation into an existing reporting product, rather than asking customers to abandon the product.

These approaches are putting pressure on traditional reporting software. A dashboard that shows monthly revenue must now compete with a system that can help investigate an unexpected change in revenue. The first displays the figure. The second may help a manager decide what to do about it.

When Building Becomes A Purchasing Decision

In the past, companies faced an expensive choice when an application lacked a feature. They could ask the vendor to add it, hire developers to build it or make do with a spreadsheet.

AI is making a fourth option more practical. A team can build a narrowly focused tool for its own needs, sometimes using an AI coding assistant to do much of the initial work.

GoDaddy described that approach to investors in April . The company said it was testing internally built replacements for smaller third party SaaS tools, with corporate functions among its initial targets. Its stated goals were lower costs and less operational complexity. It did not say that it had replaced its entire software portfolio.

A February survey from enterprise software company Retool offers some more perspective. Among 817 surveyed builders, including Retool customers, 35% reported that their teams had replaced at least one SaaS tool with a custom application. That is evidence of activity among people already engaged in building software, not a measurement of how many companies worldwide have canceled subscriptions.

For Šarūnas Putrius, CEO of Vilnius-based Helmes Lithuania, the attraction of custom development is familiar. Helmes works with companies whose purchased applications do almost everything they need, but fail at a particular task or cannot connect cleanly to another system.

Helmes executives see customers try to close those gaps with AI generated scripts and spreadsheets. The fixes can work for a time, but then a source system changes, a new employee inherits the process or the script encounters data its creator never anticipated. The result is not a cheaper, durable application. It is another system someone must look after.

Helmes is responding by concentrating on the business process behind the request. Its engineers examine where the data comes from, who needs access, what the tool must do and how the company will maintain it. The code may be easier to write now. The work of making it dependable has not vanished.

That approach appears in Helmes’s customer projects. In one, Helmes executive Bart Kappel shared that the company built an AI assistant that searches more than 50,000 service manuals and answers questions using the organization’s technical documentation. In another, an assistant helps prepare structured audit findings from documents, with people reviewing the output. Neither task calls for a general purpose application that tries to do everything. Each calls for a tool closely fitted to work the customer already performs.

The choice is not necessarily between a large SaaS subscription and a homegrown application. A company might keep the database, security controls and core business system it already trusts, then build a smaller AI tool on top.

You Can Build Faster, But Data or Systems Might Not Move As Fast

TransferGo illustrates what happens when the cost of developing software falls faster than the cost of running a regulated business.

The payments company serves people moving money internationally, including customers who prefer different languages and need services tailored to their circumstances. Company CEO Daumantas Dvilinskas says AI now handles 80% of inbound customer service requests. He reports a 94% satisfaction rating for customers served by the AI system.

TransferGo has since brought AI into engineering, marketing and other functions. Dvilinskas says its software development output doubled over the past year with roughly the same number of people.

“You can throw a lot of things into the market, test if they work,” Dvilinskas says.

One of Dvilinskas’s most interesting experiments involves an AI application built using Anthropic’s Claude and connected to TransferGo’s data warehouse. Designed to act as his own chief revenue officer, the system examines transaction patterns across approximately 1,600 international payment corridors and produces a weekly decision log.

It recently identified an unexpected surge in money transfers originating in Iceland, a market with no dedicated country manager. Nobody had asked the AI to investigate Iceland. It identified the increase independently and uncovered a possible connection to seasonal employment and work permits, which the company subsequently investigated.

TransferGo already uses Tableau for business intelligence, but its custom AI application can investigate developments that conventional dashboards might miss. The underlying data was already available, but what changed was the company’s ability to examine it without requiring someone to decide in advance which markets deserved attention. For businesses paying for multiple reporting platforms, that raises a compelling question about whether custom AI applications can deliver more value from existing data without another software subscription.

Faster coding has not made every part of the business faster.

“Now the bottleneck is infrastructure and backend,” Dvilinskas told me.

Unremarkable Tasks With Significant, Measurable Returns

LTG offers another example of AI finding work far from the company’s most visible operations. Beržauskas says its freight business receives about 7,000 customer inquiries each month. The group uses AI to check responses against 17 quality criteria and identify replies that fall short of its standards.

LTG is testing a more direct way to reduce paperwork for train stewards, too. Staff must complete reports about incidents and operational matters after journeys. Under the trial, stewards can describe events aloud and have the system help fill in the forms. Beržauskas says LTG estimates the approach could reduce that work from roughly ten hours to three hours per steward each week. It is a projected saving, not a completed result.

The company is not handing every decision to an AI agent. Beržauskas stresses that people must remain involved in decisions where errors could affect railway safety. A system that drafts a report and a system that controls train movements carry very different risks. But even then in the internal systems, the practical spending question is whether each purchased tool still earns its place.

A manager who can investigate a financial discrepancy by asking a question may need fewer custom dashboards. A customer service team that uses AI to review thousands of responses may spend less time checking them manually. A developer who can produce a small internal application in days may have less reason to buy a subscription solely for one missing feature.

one of these companies has confirmed cutting SaaS subscriptions, but their experiences suggest that the economics of buying software are changing. As AI makes custom applications cheaper to build and more capable of extracting value from existing data, businesses may find themselves paying for features they can now create themselves.

For software vendors, the challenge is no longer simply proving that their products work, but demonstrating that they deliver value customers cannot readily reproduce. The next time a renewal invoice arrives, the question may no longer be whether the software is worth the price, but whether the company needs to buy it at all.