Vertical SaaS

The idea in one minute#

Vertical software wins a small market completely rather than a large one partially, and that inversion changes every default. The market is countable, often in thousands of businesses rather than millions, so growth comes from share and from revenue per customer instead of from reach. You replace a workflow rather than a tool, which means you absorb the parts of the job that are not software, and that is where the services gravity comes from: the pull toward doing the work for the customer, which raises revenue and destroys margin. This is also the sector where AI has the strongest case, because narrow domains with repetitive document work are exactly what models are good at, and nobody else is going to build it for an industry of four thousand firms.

A picture#

flowchart LR
  IND[":i-factory: <b>One industry</b><br/><small>countable, often thousands</small>"] --> WF[":i-workflow: <b>The whole workflow</b><br/><small>not one tool in it</small>"]
  WF --> SOR[(":i-database: <b>System of record</b><br/><small>the business runs on you</small>")]
  SOR --> EXP[":i-trending-up: <b>Expansion</b><br/><small>payments, payroll, lending</small>"]
  EXP --> REV[(":i-coins: <b>Revenue per customer</b>")]
  WF -.->|"the parts that are not software"| SVC[":i-hand: <b>Services gravity</b><br/><small>revenue up, margin down</small>"]
  IND -.->|"market is finite"| CAP[":i-triangle-alert: <b>Share, not reach</b>"]
  class IND neutral
  class WF compute
  class SOR,REV memory
  class EXP queue
  class SVC,CAP warn

The two dotted paths are the sector’s characteristic traps: a ceiling you can actually hit, and a pull toward doing the work yourself.

How it really works#

The market is countable, which changes the arithmetic#

You can usually write down how many potential customers exist. That single fact reorganises everything. Addressable market is a multiplication you can defend rather than a slide; a 20% share is realistic rather than fantastical; and growth past a point must come from revenue per customer rather than from new logos, because there are no more logos.

It also means losing a customer is not noise. In a market of four thousand, every churned account is a quarter of a percent of your total opportunity and a reference your competitor now has.

You replace a workflow, not a tool#

Horizontal software does one thing for everyone. Vertical software does everything for someone, which means the product has to cover the unglamorous ends of the job: scheduling, compliance documents specific to the industry, the forms the local regulator wants, the way this trade actually quotes work. Those are the parts that create switching costs, and they are the parts a horizontal competitor will never build.

The design consequence is that domain vocabulary is not decoration. Using the industry’s own words, correctly, is most of what makes the product feel like it was built by someone who understands the business, and getting one wrong is read as evidence that you do not.

Services gravity#

Customers will ask you to do the work rather than give them the software to do it, and they will pay. Accepting raises revenue immediately and lowers gross margin permanently, and it is almost impossible to reverse because the customers who bought services bought a relationship.

The honest positions are to refuse and stay software, to offer services deliberately as a priced onboarding product with a defined end, or to become a services business with software attached and value the company accordingly. All three work. Drifting into the third while telling investors you are the first is what goes wrong.

Expansion is where the money is#

Once you are the system of record, adjacent revenue becomes available: payments, payroll, lending, insurance, marketplace take. These often exceed the software subscription, and they are only available because the workflow runs on you. Which is why the system of record position, not the feature set, is the real strategic goal.

Why AI fits this sector better than most#

Three reasons. The domain is narrow, so the set of tasks is small enough to evaluate properly. The work is repetitive and document heavy, which is the shape models handle well. And nobody else is coming: an industry of four thousand firms will never get a dedicated model from a frontier lab, so whatever you build is the only thing that exists.

The constraint is that contract values are often modest, so the cost per task arithmetic matters as much as in ecommerce. A feature costing a dollar per use against a two hundred dollar monthly subscription has a small budget.

What fills the box in 2026#

The stack is unremarkable and that is the point: a standard web application, a payments provider, and whatever industry specific integrations exist, which are often file based, elderly and undocumented. Time spent on those integrations is usually the real moat, because it is tedious and nobody wants to do it twice.

Run the numbers#

Vertical markets are finite, so the question is what revenue per customer has to be for the company to work at all. This computes the ceiling and what expansion revenue does to it.

// ceiling.go: what a finite market can actually be worth.
package main

import "fmt"

type market struct {
	name      string
	firms     float64 // total addressable businesses
	share     float64 // realistic share at maturity
	subscript float64 // annual software subscription per customer
	expansion float64 // annual adjacent revenue per customer
}

func main() {
	markets := []market{
		{"small, high value", 4_000, 0.30, 9_600, 14_000},
		{"medium", 26_000, 0.18, 4_200, 5_500},
		{"large, low value", 180_000, 0.08, 1_320, 900},
	}
	fmt.Printf("%-20s %9s %8s %14s %14s %14s\n",
		"market", "firms", "share", "software only", "with expansion", "share needed")
	for _, m := range markets {
		customers := m.firms * m.share
		soft := customers * m.subscript
		both := customers * (m.subscript + m.expansion)
		// Share you would need, on software alone, to reach fifty million.
		needed := 50_000_000 / (m.firms * m.subscript)
		note := fmt.Sprintf("%.0f%%", 100*needed)
		if needed > 1 {
			note = "impossible"
		}
		fmt.Printf("%-20s %9.0f %7.0f%% %14.0f %14.0f %14s\n",
			m.name, m.firms, 100*m.share, soft, both, note)
	}
	fmt.Println("\nShare needed is what software alone would require to reach fifty million of revenue.")
	fmt.Println("Expansion revenue is not an upsell here. For two of these it is the entire thesis.")
}

The last column is the argument. To reach fifty million on software subscriptions alone, the small high value market would need more than every firm in it, which is to say it cannot; the medium market needs 46%, which essentially nobody achieves; and even the large low value market needs 21%, which is a dominant position.

Now read the two columns before it. At the realistic shares in column three, none of these three businesses reaches fifty million on software, and adding expansion revenue roughly doubles every one of them, taking the medium market most of the way there. That is why vertical companies move into payments, payroll and lending, and why it is a strategic requirement rather than opportunism. In a finite market, once share stops growing, revenue per customer is the only axis left.

What it changes for each function#

Remember this#

• The market is countable, so addressable market is a defensible multiplication and share has a real ceiling.
• You replace a workflow, which means owning the unglamorous parts. That is the switching cost.
• Services gravity raises revenue and permanently lowers margin. Choose a position rather than drifting into one.
• Expansion revenue is the thesis, not an upsell, because revenue per customer is the only axis left once share saturates.
• AI fits unusually well here, and is constrained by contract value rather than by capability.

Try it#

1. Run ceiling.go with your own market size and pricing. What share would software alone require, and is that number achievable?
2. Count your actual addressable firms from a real source, an industry register or an association membership list. Compare it to the number in your last deck.
3. List the revenue you earn that is really services. Compute gross margin with and without it.

Check yourself#

1. Why does a finite market make revenue per customer a strategic question rather than a pricing one?
2. Services gravity is hard to reverse. What specifically makes it hard, and which of the three honest positions is the dangerous one?
3. Why is an industry of four thousand firms a better AI opportunity than a market of four million consumers, and what limits it?