In a post, an entrepreneur described how he used AI to build his own CRM and cancelled his subscription to a third-party service. He said several customers of his own company later left its service for tools they built themselves. In one case, he saved on a subscription; in the other, he lost some subscription revenue.
That, to me, is where the conversation about the future of software as a service gets interesting. AI gives businesses more freedom to choose between buying and building. Vendors will have to show more clearly what customers get for their monthly fee.
Who will look after the CRM once it goes live?
Take a hypothetical CRM priced at $100 a month: the subscription costs $1,200 a year. Comparing that figure with the cost of one evening at a computer would be misleading. Look at the same year and add up hosting, data migration, testing updates, maintaining integrations and the time people spend away from other work.
The difference becomes clear on an ordinary Monday: the accounting team updates its system, and the CRM stops sending invoices across. Someone has to spot the problem, restore the connection and reconcile the missing records. AI can help diagnose the issue. The person responsible for the system still has to decide when it is safe to resume operations.
Building in-house can pay off when off-the-shelf products force the team to work around limitations every day and the company has someone to maintain its system. A subscription makes sense when the vendor takes on those duties and can do the work for less by sharing the cost across many customers. Before choosing, I would ask both sides to show how they would recover data after a mistake and how long it would take.
Cheaper code can expand the market
The first products to feel pressure will be those customers use for a few forms, a task list and a standard report. It is a different decision when accounting, payments or production continuity depend on the system. Buyers then consider industry knowledge, migration costs, support and the consequences of a mistake. Recreating the interface quickly does little to remove those costs.
Cheaper development also opens up demand that did not exist before. A small bakery that used to track stock in a spreadsheet may now be able to afford a tool that connects orders, production and inventory. It is a possible rebound effect: as something becomes cheaper to use, people find more uses for it. There can be more software even as revenue falls for some vendors.
Some spending will go to model and compute providers. I would not assume they will capture all the gains: competition between models gives buyers options if an application can move to another service. More spending may go to integration and maintenance. Companies that build their own tools quickly today may need help tomorrow with data migration, access reviews and recovery after an outage.
What will customers keep paying for?
I see an opportunity for Ukrainian software companies to build a service around a specific workflow. In lead management, for instance, every enquiry from agreed channels could reach the right salesperson, while their manager can see which leads still need a response. The vendor can take responsibility for keeping integrations working and agree on an incident response time. That gives the customer a concrete basis for judging the subscription price.
By 2031, I expect a mix of three changes: standard features will get cheaper, more businesses will build narrow tools of their own, and some SaaS vendors will charge for running a workflow at a measurable level of service. The balance will depend on how easily customers can switch and what an outage costs them.
What happens to demand for labour?
The hardest question for the industry is whether demand for development will grow as quickly as productivity. An ILO review of generative AI and work points to uneven productivity gains and limited evidence of large-scale job displacement. It is too soon to say how many people the industry will need in five years. Managers can already think about how junior staff will gain experience if AI takes over some of the simpler tasks they used to learn on.
I would start with two questions for every product: what part of the work are customers willing to own, and what would they rather leave to us? The answers will show where subscriptions lose their appeal and where a vendor has a chance to offer a service customers will pay for year after year.
Related work at the Association: Software Ukraine’s AI Commission.