Vertical AI Agents Are Eating Horizontal SaaS, and the Reason Is Pricing

Vertical AI agents, built for one industry, one workflow, one role, are growing faster than any previous software cohort. The usual explanation is that focused products work better than general ones, which is true and does not explain the speed.

The structural cause is a pricing change. Horizontal software sells seats. A vertical agent sells finished work. When a buyer moves from needing fifty CRM licences to needing five thousand tickets resolved this month, every part of the business model changes with them.

Why the unit matters more than the product

A seat is a proxy for value. It assumes that more people using the software means more benefit, which held while software made a person faster and stopped holding when software started completing the task.

Completed work is not a proxy. It is the thing the buyer wanted. That collapses the entire argument a vendor normally has to make. Nobody needs to be persuaded that a resolved ticket is worth something, and nobody needs a business case to compare the price of one against what it currently costs them.

What this does to the sale

“The unit has to be something the industry already counts, already argues about, and already assigns a cost to. Choose the wrong one and the pricing conversation never becomes arithmetic.”

It removes the hardest conversation in B2B software, which is establishing that the product produces value at all. A seat-priced tool spends its entire first year arguing about adoption, utilisation, and whether the licences are being used. An outcome-priced agent argues about volume and accuracy, which are both measurable.

It also changes who buys. Seats come out of a software budget with a procurement process attached. Completed work often comes out of an operational budget, where the comparison is to the cost of doing it another way rather than to other software.

The trap in going pure outcome

The unit has to be something the industry already counts, already argues about, and already assigns a cost to.

The temptation is to price entirely on results, and the current guidance from people running these businesses is not to, at least at first. The recommended structure is hybrid: a small floor plus a usage layer, held until accuracy is proven in production.

The reason is that pure outcome pricing transfers all accuracy risk to the vendor before the vendor knows their own accuracy. A small early customer with messy data can produce a month where the agent resolves very little and the vendor earns almost nothing, for reasons that are the customer's rather than the product's.

A floor covers cost to serve while the accuracy question is still open. Once the numbers are boring and predictable, the floor can shrink.

What a domain expert should take from this

The opportunity is not to build an agent. Agent capability is available to everyone and improves regardless of what you do. The opportunity is knowing which unit of completed work is worth buying in an industry you already understand.

That is a genuinely hard question and it is not answerable from outside. Is it a resolved ticket, a reviewed contract, a validated claim, a closed defect, a benchmarked lane? The unit has to be something the industry already counts, already argues about, and already assigns a cost to. Choose the wrong unit and the pricing conversation never becomes arithmetic.

Somebody who spent a decade in that industry knows which number gets reported upward every month. That knowledge is the product decision, and it precedes any technical one.

See why seat-based pricing is shrinking for the wider shift, and hybrid pricing for how the floor and usage layer fit together.