AI's Impact on SaaS will be Uneven — an HBR Study
AI & Technology
AI doesn't make all SaaS less valuable — it makes weakly differentiated SaaS easy to replace. Harvard's Christopher Stanton offers a simple framework for telling the vendors worth keeping from the spreadsheet-wrappers worth cutting.
If you've been following the tech news lately, you've probably heard the term "SaaSpocalypse" being thrown around. Earlier this year, software stocks took a beating, and a lot of executives drew the same conclusion: AI can build software now, so why are we still paying all these SaaS vendors?
It's a tempting argument. But according to Harvard Business School professor Christopher Stanton, it's also a pretty dangerous oversimplification.
The problem with painting SaaS with one brush
Not all software does the same thing, and that distinction matters enormously right now. Stanton analyzed 45 publicly traded SaaS companies and found that the market is already starting to price in a key difference — even if most executive teams haven't caught up yet.
His framework comes down to two simple questions: What is the software actually doing, and where does its data come from?
Think of it as a 2×2 matrix
On one axis, you've got the nature of the task. Some software is deterministic — it just looks things up. What did we invoice last month? Which part fits this model? These are questions with a fixed answer somewhere in a database. Other software is predictive — it's trying to figure out what's most likely wrong, what should happen next, or which action is most promising when the answer isn't obvious.
On the other axis, you've got the data source. Some tools run purely on your company's internal data. Others are powered by pooled context — patterns learned across thousands of customers, environments, and edge cases.
Where the real disruption is happening
If your SaaS sits in the top-left — deterministic tasks, internal data — you should probably be worried. These tools mostly just surface your own information through a set of known rules. Scheduling software, basic workflow tools, internal ops platforms. AI coding tools have made it genuinely cheap to build these in-house now, and plenty of companies already are.
The pessimism around SaaS largely reflects this quadrant, and fairly so.
Where the pessimism goes too far
The bottom-right quadrant is a completely different story. This is what Stanton calls operational intelligence — predictive tools powered by pooled data from across an entire industry or ecosystem. And this is where blanket SaaS-cutting gets companies into trouble.
His go-to example is Bluon, an AI assistant for HVAC technicians. When a technician faces a failing unit in the field, they're not just doing a lookup — they're diagnosing a problem that could stem from a dozen different causes, many of them rare. Bluon is trained on 135,000 real tech support calls handled by veteran technicians, combined with documentation across 3 million models and 60 million parts cross-references.
Here's the kicker: 37% of the real-world queries in that dataset are genuine outliers — edge cases that don't fit neatly into any category. Those are exactly the situations where a single shop's records are useless, where mistakes are costly, and where a tool built purely on internal data would fail. That's the moat.
The market is already figuring this out
Since ChatGPT launched in late 2022, a portfolio of predictive-pooled SaaS companies has outperformed a portfolio of deterministic-internal ones by about two percentage points per month. During the early-2026 selloff, the deterministic-internal group dropped roughly 9% while the predictive-pooled group dropped only around 4%.
The logic embedded in those valuations is worth paying attention to.
So what should you actually do?
For tools in the deterministic-internal quadrant: seriously consider building in-house or consolidating vendors. The economics have shifted.
For pooled, predictive tools: don't just take the vendor's word for it. A slick demo will always show you the easy cases — the common, well-documented problems any half-decent product can handle. You need to test the edge cases. Find the queries that used to require your most experienced people and throw those at the tool. If it handles them well, it's earning its keep. If it doesn't, you've learned something valuable.
The bottom line
AI doesn't make all SaaS less valuable. It makes weakly differentiated SaaS easier to replace. The vendors worth keeping are the ones sitting on rare, pooled, operationally verified data — the kind that applies hard-won expertise to costly, judgment-heavy work that no single company could replicate on their own.
Cut the spreadsheet-wrappers. Keep the intelligence.
Source
Based on AI's Impact on SaaS Will Be Uneven — Here's What Leaders Need to Know, Harvard Business Review, May 2026.