SaaS unbundling is coming for your plant-floor module stack

Bain's 2025 technology report makes a straightforward argument: agentic AI is going to collapse chunks of the enterprise SaaS stack into orchestration layers, and a lot of per-seat tools are going to look very different in two years. You can read the full piece here.

That's an analyst framing. On the plant floor, the practical version is simpler. The module stack you've accumulated around your MES, historian, and CMMS is about to get re-evaluated whether you want it to or not.

What the stack actually looks like

Walk through most mid-sized plants and you'll find a familiar accumulation. The MES vendor's quality module, licensed per user. A separate OEE dashboard product with its own seats. A downtime tracker from a different vendor. A scheduling add-on. A spreadsheet-plus-portal for CAPA. Maybe a standalone SPC tool the quality team bought because the MES module was too clunky.

Each one made sense when it was bought. Each one has a per-seat or per-site line item. Each one holds a slice of data the other tools can't easily see.

Why the math is changing

Richard Chu's essay on the great SaaS unbundling frames it well: AI compresses the cost of workflow-heavy software, and the per-seat model starts to leak value for anything that's really just "pull data, apply logic, notify a human."

A lot of plant-floor modules are exactly that. Pull cavity pressure traces from the historian. Join to the MES work order. Flag deviations against the recipe. Create a quality hold. Notify the shift lead. Five years ago, that workflow justified a licensed module. Today it's a few hundred lines of code, an agent, and a connection to tags you already own.

The build side of build-vs-buy got dramatically cheaper. The buy side didn't.

What this doesn't mean

It doesn't mean rip and replace. Your MES is probably fine. Your historian is doing its job. The stuff that handles transactions, compliance, and system-of-record obligations is not what gets unbundled first.

What gets unbundled first is the fifth seat-based add-on sitting on top of data you already pay to collect. The workflow tools. The dashboard-with-logic products. The modules you bought because the base system didn't quite do the thing you needed.

A useful exercise

List every piece of plant-floor software with a per-user or per-site fee. Next to each one, write what workflow it actually performs and what data it touches. Anything where the workflow is specific to how your plant runs, and the data already lives in your historian or MES, is a candidate to own instead of rent.

You probably won't act on all of them. But the list itself is worth having before renewal season. The build-vs-buy answer that was obvious in 2020 is not the same answer in 2025, and the vendors know it before you do.