Is SaaS dead? The reports of SaaS’ death are greatly exaggerated.

The $64,000 question is: Will AI replace SaaS? No. Does it have the potential to replace parts of SaaS? Potentially. …read on.

The broad party line is that AI will replace SaaS. That statement has been repeated by a number of people from technology leaders to major business news outlets. Even the financial markets have leaned into the hype that AI will somehow replace SaaS. 

Unfortunately, the hype has largely ignored reality suggesting that one could simple vibe-code an enterprise solution like CRM or ERP by simply using AI. As if making a statement like this is bad enough, it has taken off as many vendors of SaaS-based solutions watched their stock get pummeled.

Unfortunately, the statement that ‘AI will replace SaaS’ is both wrong and misleading. Will AI replace SaaS? No. Will it replace parts of SaaS? Potentially, yes.

Let’s break down the reasons why AI will not replace SaaS…and a few potential ways it might replace components of SaaS offerings. But even those might be seen more as augmenting SaaS as opposed to replacing it.

Lets break it down…

Building enterprise applications

Building enterprise applications is not for the faint of heart. It takes a lot of experience and ingenuity to understand how a business operates and then programmatically create an application you can entrust with running your business.

Enterprise applications like SAP for ERP or Salesforce for CRM were not built overnight and have gone through many evolutions to get where they are today. In SAP’s case, encapsulating 50 years of enterprise experience is not written in a way to easily rebuild their suite of applications.

The degree of complexity and sophistication in applications such as these…and even (relatively) newer applications like Workday or ServiceNow is great. To replicate any of those would be a monumental feat.

Could someone vibe-code a sophisticated enterprise application like these through prompts? In a word…No. Not only would it be highly inefficient, but AI would create a level of undeterministic or non-deterministic output that would leave even the most extremely flexible enterprises wildly uncomfortable running their business on these solutions.

One might argue that ‘building’ an enterprise application this way is antiquated thinking in favor of leveraging agentic workflows. While that may be true for some processes, enterprises still need core architecture and deterministic outcomes to ensure stability in processing. Neither are core to today’s AI frameworks.

Is AI ready for building enterprise applications?

The short answer is no. AI can do many amazing things, and we are learning about new opportunities every day. Even so, looking beyond deterministic systems, enterprise applications need several components that AI is just not ready to address.

The first one is a topic that few vendors are openly discussing when it comes to broader AI systems: Governance. There is good reason for this. Solving the governance problem is hard. Really hard! Governance is hard, complicated and often conflicting. There are many layers and aspects that must be considered.

To counter those challenges, solutions using AI today often take one of two paths: 1) Their governance model is confined to data within their solution. By focusing on their own solution, the scope is specific creates a more controllable space. 2) Solutions have very specific, often smaller scope, well-defined functions that have little need for governance. Or if governance is required, the scope is limited making governance requirements more straightforward. Unfortunately, over time, both approaches are very limiting as data is increasingly coming from other systems of record. And… those systems have their own governance models to contend with.

To show how limiting this can be, I asked one vendor how they manage data and governance models from other systems. They told me “Our system manages the data within the system. We don’t leverage other governance models. If you don’t want to use our governance system, don’t bring the data into our system.” 

For the sake of argument, let’s say that one could use AI to build an enterprise application. There are many additional reasons why this approach would struggle to succeed.

Enterprise strategy: Buy vs build

CIOs constantly debate the buy vs build decision. Does it make sense to build a bespoke solution? Or is it more appropriate to buy a commercially viable solution?

It is a constant, ongoing debate and as the variables change, so does the outcome. It is not a binary decision to buy or build everything. There is often a thesis behind the strategy as to whether to buy vs build something. The thesis takes into account many aspects from organizational capacity, strategy, priorities and resources just to mention a few.

Enterprises have limits and finite resources in both funding and manpower. I’ll come back to these points.

By moving from commercial enterprise software to AI-built software, it requires an enterprise to fully understand, articulate and build the entire enterprise application. Even though most enterprises only use a portion of any given enterprise application, it is still a large lift.

Building the application

Building the architecture for an enterprise application, even with AI, would require a phenomenal amount of planning, architecture, work, testing and validation. Enterprise applications would require cross-functional understanding, which often does not exist in depth, within any given organization or department. Building an enterprise application would require cross-functional engagement throughout the process. In addition, it requires that each department has the priority of this application aligned across departments. From experience, that is challenging for even simpler bespoke applications let alone an enterprise application.

Operate/ Run the application

Once the application is built, enterprises would need to serve as the operator and run the application. By shifting from SaaS to AI-built, it shifts responsibility of running and operating the application to the enterprise. This adds both responsibility and scalability to ensure the application is running optimally. Because this is essentially a bespoke application versus a commercial application, there are nuances that operational teams would need to watch for.

The underlying resources would need to be built for scale at peak to avoid performance issues. With SaaS, vendors can normalize and smooth utilization across many customers, not just one. For the enterprise, this could be a costly venture.

Maintain/ Grow the application

As needs change and bugs are identified, unlike SaaS, the enterprise is squarely responsible for the upkeep and maintenance of the enterprise application. Updates would need to be assessed, priortized and then scheduled into a development team’s workload alongside everything else they are responsible for. It’s likely there would be conflicts in priorities and limitations on resources. Even using AI, there is still validation of code and testing required.

Ensuring security, trust and governance of the application

Enterprises only know what they know. Broadly speaking, enterprise IT organizations are not highly efficient software factories that specialize on one thing. They are complex organizations that understand the complexity of their business and how different applications interact.

By moving from SaaS to AI, it shifts the risk in building, operating and securing an application solely to the enterprise. Any security related issues are the responsibility of the enterprise, not a software vendor. The enterprise’s security and development teams would have a hefty workload ensuring that an AI-generated enterprise application is both secure and protected.

Holes in this protection would quickly lead to trust issues with both customers, leaders and stakeholders. This would ultimately lead to a significant impact to their business. That is a huge risk. Add to this the non-deterministic nature of AI and the problem grows even larger.

Today, the governance models for many AI solutions are focused on very specific data sets and functions. Why? Because enterprise applications is a complicated space that needs careful consideration. The thinking is: Limit the scope, limit the complexity. To date, I have not seen any solution that can truly address the complex governance that enterprises need today from AI. This is why we are seeing limited application of governance models in AI.

Supporting the application

One benefit of SaaS based solutions is a configure and use it approach. Once the application is deployed, the organization shifts into an exception-based mindset for supporting the solution.

With AI, the organization would be fully engaged to provide constant support from operations to functional changes to security issues. This also shifts the responsibility and risk for support solely to the enterprise. Unlike SaaS, there would be no backstop or other organizational to call for backup. The enterprise would accept that responsibility in earnest.

Cost/ scale to build, operate, support, maintain an enterprise application

The economics for SaaS-based enterprise applications are amortized across all customers. Vendors build specialized teams that are wholly focused on that application. Those costs to build, operate, support and maintain the application are shared across all customers…not just one.

When the enterprise builds the application using AI, it bears the entire cost of building, operating, supporting and maintaining the enterprise application. There is also a reality to operating an AI-based solution that could lead to further inefficiencies that the enterprise would have to solely absorb.

With SaaS-based solutions, costs are shared across all customers. Each customer only pays a portion of the costs. AI-based application costs are completely born by the enterprise. In addition, we are still not seeing the true costs for AI-based workloads passed to the enterprise consumer. Over time, as these costs get pushed down the value chain, it dramatically changes the calculus for ROI calculations. Today, we are only seeing a portion of those AI costs. Once the value chain starts to normalize, we can expect to see many AI efforts drop off due to cost. Hopefully efficiencies in AI operations over time will offset some of those costs, but likely only to a small degree.

Priorities and organizational impact

CIOs do not have unlimited resources (people and funding) and must prioritize the limited resources they do have toward efforts that align with the business outcomes. Just because a project has a positive ROI does not mean it is green lit to proceed. Any effort also needs to pass many hurdles beyond people and funding. For example, process reviews, organizational alignment in priorities, security assessments…and the list goes on.

These are not just hurdles that one might see as preventing innovation, but a reality that CIOs need to contend with to ensure the best use of resources and protection of assets.

There is a reality to the time involved by all parties involved, even with AI, that building any application, especially one of this magnitude, requires time and effort. It takes time to build, test, deploy and optimize. 

The prioritization also considers that any given enterprise will already have an incumbent solution installed and working. So, the challenge is not just to prove out the ROI of an AI-based solution over a SaaS-based solution…but also the switching costs to move from an existing solution to a new solution. Many times, the risk of change can not just tip the scales, but at times outweigh the benefits.

Change management

Change is hard. The more critical the business function, the more challenging it is for change. This is where AI might be able to help due to its dynamic nature. However, there are tradeoffs that change management would be equally concerning from a trust perspective.

Where could AI play a role?

All that to say that using AI to directly replace SaaS-based enterprise applications would be challenging today. This is not to say AI is not valuable to enterprise applications. AI is an incredibly powerful tool that enterprise applications need to leverage in meaningful ways.

More to the point…if enterprise systems are not already deeply leveraging AI, they are largely behind the curve. Even the most complicated and sophisticated systems such as ERP and CRM can benefit from AI today.

So where do we think AI will show up in enterprise applications? There are likely two swim lanes that AI development will impact enterprise systems: Augmentation and interfaces.

It is likely that we see AI augmenting existing systems by accelerating insights to business workflows. From anomaly detection in manufacturing to identifying ways to optimize business processes, AI is already being introduced deeply into core systems. AI is not just for internally focused processes. AI can help better understand customer signals, drive deeper relationships and ultimately provide new revenue growth opportunities. This work elevates insights that, in the past, were hard to reach if not impossible without AI.

The second opportunity is from the way customers and employees engage with your business and enterprise systems. Instead of a series of screens, forms or questions, a user would simply ask their question into a prompt and AI would do the work behind the scenes to find the correct responses or actions to take. 

There is a wide spectrum of classifications of applications. While I have focused on the largest examples of SaaS-based applications to make the point, there are smaller and less complicated applications with similar challenges. Many of the points made around shifting responsibility still apply such as ownership, responsibility, risk and costs apply regardless of the size and complexity of the enterprise application.

CIO Perspective

On the surface, it may sound very enticing to (theoretically) think that AI could replace existing enterprise applications. It might be even more attractive if your current solution feels very antiquated and lacking innovation.

It is an important and good thought experiment to consider how to open our minds to wholly new approaches. A core question we need to ask is: What if we do use AI to build new applications? We need to consider the answer holistically and honestly. 

Equally important is to ask the reverse question too: What if we don’t use AI to build new applications? What are the implications?

The bottom line is: Don’t let change itself be a stumbling block to considering alternatives.

The next logical question is: Are there other alternatives in-between where we can leverage the innovation while limiting the risk? The short answer to the last question is: Yes! AI is opening the door to so many opportunities. It is important to understand what problems it can solve and ensure that it is used appropriately. AI is not like peanut butter and spread evenly. Implementations of AI will be very uneven for good reason.

So, back to the $64,000 question: Will AI replace SaaS? No. Will it augment and change aspects of how we leverage SaaS? Absolutely…and it has already started to do so.

Should we continue to ask provocative questions like ‘could AI replace SaaS’? Yes. But we also need to ensure we are doing it in an honest and holistic way to ensure we are not misguided and creating collateral damage in the process.