Enterprise SaaS spending is entering a new phase as artificial intelligence becomes deeply embedded across business operations. Rather than accelerating traditional SaaS growth on its own, AI is changing where incremental technology dollars flow. Increasingly, organizations are blending commercial SaaS platforms with custom-built AI systems, reshaping buying behavior and shifting value toward cloud infrastructure providers.
Insights from recent enterprise discussions and survey data show that while SaaS remains foundational, companies are prioritizing flexibility, customization, and control as they integrate AI into core workflows. This shift is redefining the economics of SaaS, particularly as generative AI and agent-based automation move from experimentation to production.
The Head of Data Science at a staffing firm suggests that while core SaaS spending may remain stable, opportunities for vendors could narrow in certain areas. They expect that many “out-of-the-box” AI components will fit naturally within their established Salesforce or ServiceNow ecosystems. “However, there are a lot of customizations that don’t work out-of-the-box, and for those you might want to build them out yourselves. That revenue stream most likely won’t be captured by the provider.” The remainder of the panel agree, similarly planning to anchor to major SaaS and cloud platforms for compliance and speed of development, diverting add-on spend as needed toward homegrown tools tailored to proprietary datasets and workflows. “Our focus has become more of an integration rather than building everything from scratch,” says the Chief Software Architect for a large educational institution. “Much of the compliance, if you want to facilitate by building, you have to recreate the wheel that you already have. Most of the time, 90%, 95%, we start from commercially off-the-shelf software and SaaS providers, and then we build integration to facilitate our needs.”
On internal needs, one executive described building an internal risk-scoring engine to monitor environmental changes; another uses AWS Bedrock to extract and structure data from vast collections of documents; a third has built in-house AI agents for compliance-sensitive research projects (also on Bedrock) for data privacy and institutional control. “We have 16 Nobel winners in our faculty base, and many of these are very, very secretive projects.” Initiatives aim to automate data extraction, making information “more machine readable and database ready,” with humans in the loop for oversight. “For the last few months, we’ve been doing some pilots, and it’s now at the position where we can actually take this to the next level and put a proper application solution around it.”
ETR Data: In ETR’s October 2025 AI Impact on SaaS Drill Down survey, only 3% of all respondents stated that AI systems required no level of customization, versus 94% citing some level of customization. The majority (53%) required moderate customization, with an additional 36% categorized the customization efforts as “significant”. When isolated to Large organizations (1,200+ employees), 98% stated that customization was required, without a single response indicating that zero customization was needed.
A fourth executive detailed how their company is integrating AI to streamline back-office operations; their firm’s initial model interpreted data from manually filled timesheets, but the latest evolution adds a conversational interface that instantly provides employees with expected pay details based on hours worked and declared deductions. “What’s improved now is that we allow people to query and understand, rather than having to call the call center.”
ETR’s most recent Macro Views Survey data shows that overall SaaS licenses are expanding by only 1.4% within Global 2000 organizations over the next six months, whereas DevOps spend is on a relative tear at 2.6% within that same cohort. “We’re trying to consolidate and go to a central one vendor, something like GitLab, where all the developers can have a streamlined experience with AI development assistance.” Total spending with the chosen vendor will likely rise as the company standardizes around it. Another describes a similar rationalization occurring in HR, though warns that once software providers begin monetizing agentic AI integrations (charging per digital agent as they do for human users) seat-based costs could rebound. “Right now, what we are doing is integrations that use your data or build on top of it, but it’s not purely agentic, so it’s not counted as a seat-based usage, but if you want an agentic integration, that starts becoming a seat-based cost. You might see a drop in spending now, which might change substantially in the future, the way pricing moves along.” A third echoes that uncertainty, predicting a reduction in some areas but an increase in others. “For example, Microsoft Copilot. I think it was all very exciting to start with, but we’ve actually reduced our license count in that probably by 50% in the last six months, because there are better ways and other ways of doing things.” In some cases, APIs enable access to applications without traditional seat licenses, which may portend an eventual shift from per-user pricing to models based on compute power or API usage.
One Chief Software Architect makes an interesting regulatory point: “You can’t just run an agent that’s anonymous, because of audits and compliance. Even if we have an agent, that agent actually has an identity and access management footprint—which, by the way, now adds to my seats.” In ETR’s October 2025 AI Impact on SaaS Drill Down survey, a question on AI agent autonomy was posed, and only 2% of respondents stated that their AI agents currently in use were fully autonomous and capable of operating workflows without oversight.
As to specific vendors, ServiceNow is solidly entrenched. “Within our organization, it’s [ServiceNow] our bread and butter. It’s tied into so many processes, so many workloads, CMDB, all the changes in production, and has so much history rich in data of the things that happen in our environment. I don’t see that going anywhere, honestly.” A second panelist added: “ServiceNow is just going to quadruple in our organization.” All are pleased with ServiceNow’s focus on automation and agent-driven innovation, though one warns of potential interoperability issues. “If you’re within the ecosystem of ServiceNow, it’s perfectly fine, you can use their agentic services. It’s usually a challenge when you have to move things out of their systems and then interact with other systems.” Most of our panelists are either piloting or preparing to roll out ServiceNow’s AI features in the coming year. “We’re sort of pushing ourselves more and more towards ServiceNow, because we want to start taking advantage of these.” Another IT team has already fully deployed. “We’re connecting chat, or using natural language to interact with ServiceNow, to query, ask questions, or to automate or do some simple tasks, repeated tasks. We have our own built-in internal ChatGPT portal, and we are tying more into some of the ServiceNow services it offers.”
However, Workday’s sustained growth in HR is less certain as automation spreads. “The problem is, even if you have Workday, it won’t be able to capitalize, perhaps, on the additional revenue on the agentic side initially, because of other tools that take more precedence in the ecosystem. However, you need the integrations and the APIs coming through it, because if that is your HR tool, you’re not going to simply stand up tomorrow and change the entire ecosystem just to have an agentic solution.” One panelist imagines HR as a balancing act between automation and human connection. “You’d like to think that quite a lot of what an HR team does is relatively repetitive, rife for AI to sort of come in and start to actually disrupt a bit.” They warn, however, that HCM must preserve a personal touch. “You’ve got the other piece of HR that you’re never going to ignore, which is that human-centric nature of it. I think there will be some impact to some players if they don’t get it right.”
Collectively, the panel is less enthusiastic about Salesforce Agentforce. Certain promised capabilities, such as seamless integration and the zero-copy feature, have yet to debut, while complex licensing and restrictive pricing structures make large-scale deployment impractical for some. “This is the vendor we were talking about where they were nickeling and diming us on cores.” Ultimately, nominally automated sales tasks do not justify the cost. “We are actually going to reduce Salesforce SaaS licenses, because not only it didn’t work for us, but also we actually ramped up with their promise and then now don’t know what to do with it.”
For these panelists, there is a consensus that companies are turning en masse to Azure and AWS for ready-made AI infrastructure and scalable compute powers. As one panelist succinctly stated, “No matter which software wins, it’s the cloud providers that are going to win the most because they are underneath everything.”