Vertical Agent Pricing: How to Capture Industry-Specific Value Without Leaving Money on the Table

Table of Contents

• Why Vertical Agents Break Horizontal Pricing Logic
• The Four Pricing Archetypes for Vertical Agents
• Anchoring Price to Industry-Specific Value
• How Pricing Shifts by Vertical
• The Margin Problem Nobody Wants to Discuss
• Attribution: The Hidden Tax on Outcome Pricing
• Packaging, Floors, and the Land-and-Expand Path
• Insights Most People Overlook
• References

Why Vertical Agents Break Horizontal Pricing Logic

For two decades, B2B software priced itself one way: per seat, per month. The logic was clean. More users meant more value, more value justified a higher price, and the marginal cost of another login was roughly zero. Salesforce, Slack, and a thousand imitators rode that curve to enormous valuations.

Vertical agents detonate that logic for a simple reason. An agent doesn't sit at a desk waiting for a human to click it. It does the work. When a legal-review agent processes a contract, there's no "user" in the traditional sense -- there's a task that got completed and a paralegal who didn't have to do it. Pricing per seat for something that eliminates the seat is, on its face, absurd. You'd be charging for chairs in a room you just emptied.

This is the central tension in the broader agentic AI-as-a-service market, and vertical agents feel it most acutely because they're closest to measurable, dollar-denominated work. A horizontal platform like a general-purpose assistant can hide behind vague "productivity" claims. A vertical agent built for medical coding or insurance adjudication is staring directly at a number the customer already knows: what that work costs today.

That visibility cuts both ways. It makes value-based pricing possible -- you can point to the exact line item you're replacing. But it also means the customer can do the math you're doing, and they'll resist any price that doesn't leave them an obvious win. The art of vertical agent pricing is capturing enough of the value you create to build a real business, while leaving enough on the table that the customer feels like the smart one for buying.

The Four Pricing Archetypes for Vertical Agents

Most vertical agent pricing collapses into four shapes, each with a different relationship between what you charge and what you deliver.

Seat-based survives mostly in agents that augment rather than replace -- a research agent a financial analyst uses ten times a day, where the human is still firmly in the loop. It's familiar to buyers and easy to forecast, but it caps your upside and invites the obvious objection: "if it's so autonomous, why am I paying per person?"

Per-task charges for each discrete unit of work: a contract reviewed, a support ticket resolved, an invoice reconciled. This aligns price with consumption and scales naturally with usage. The risk is that it turns your agent into a metered utility, and buyers start optimizing to use it less -- the opposite of what you want.

Per-outcome only charges when a defined result happens: a qualified lead booked, a claim correctly adjudicated, a collections call that recovers money. This is the model investors and founders are most excited about, and a16z has argued that outcome-based pricing reflects where agentic software is heading because it ties vendor revenue directly to customer value. It's also the hardest to operationalize, for reasons we'll get to.

Value-share takes a percentage of the economic gain -- a cut of recovered revenue, a slice of cost saved, a fee tied to deal size. It's the most aggressive form of value capture and the one that most resembles how agencies and contingency-fee professionals already get paid. It works best where the value is large, discrete, and cleanly attributable.

In practice, the strongest vertical agent companies blend these. A platform fee plus per-outcome charges. A base subscription that includes a task allotment, with overage billing above it. Pure single-model pricing is increasingly rare because each archetype solves a different problem -- predictability, alignment, or upside -- and mature buyers want a structure that addresses all three.

Anchoring Price to Industry-Specific Value

Here's the move that separates vertical agent pricing from generic SaaS: you anchor to a number that already exists inside the industry.

Every vertical has a "reference price" for the work an agent does. For a medical-coding agent, it's the fully loaded cost of a certified coder plus the denial rate their errors produce. For a sales-development agent, it's the cost-per-meeting an SDR or an outsourced agency charges. For an e-discovery agent, it's the per-document review rate that contract attorneys bill. These numbers aren't secret -- they're operational benchmarks the buyer's finance team already tracks.

When you anchor your price to that reference, three things happen. The conversation stops being about software budgets (which are tight and contested) and starts being about labor or vendor budgets (which are far larger). The ROI math becomes self-evident. And you gain pricing power, because you're not competing against other software -- you're competing against the status quo, which is usually expensive, slow, and error-prone.

The discipline is figuring out which reference number to anchor to, because most verticals have several. Anchor too low -- say, to a junior employee's hourly wage -- and you've left enormous value uncaptured. Anchor too high -- to the total revenue your agent influences -- and you'll trigger sticker shock and resistance. McKinsey's research on the economic potential of generative AI found that the largest value pools cluster in a handful of functions, which is exactly where the strongest reference anchors live: customer operations, software engineering, sales, and knowledge work.

The cleanest anchors share three traits. The work is repetitive and high-volume, so the agent can do a lot of it. The cost is legible and tracked, so the buyer can verify your ROI claim. And the error has a price -- a denied claim, a lost deal, a compliance fine -- so you can argue you're not just cheaper but better. Verticals that hit all three (claims, coding, collections, compliance) support the most aggressive value capture. Verticals where the work is fuzzy or the error is invisible push you back toward per-task or seat pricing whether you like it or not.

How Pricing Shifts by Vertical

The same agent architecture demands wildly different pricing depending on the industry it serves. A few illustrative contrasts:

Legal. Lawyers already work on billable hours and contingency fees, so they understand both per-task and value-share intuitively. A contract-review agent can credibly charge per contract or per document, anchored to the attorney rate it displaces. But regulated liability means the firm still wants a human signing off, which keeps some seat-based revenue alive. Pricing here leans hybrid: platform fee plus per-document.

Healthcare. Clinical documentation and prior-authorization agents face a liability wall -- a wrong output can harm a patient or trigger a denial. Per-outcome pricing is tempting (charge per approved authorization) but fraught, because the agent doesn't fully control the outcome; the payer does. The realistic model is per-task with a quality SLA, because the buyer needs predictability and the vendor can't fully own the result.

Sales and marketing. This is per-outcome's home turf. A meeting booked, a qualified opportunity created, a piece of content published -- all discrete, attributable, and tied to revenue the buyer already values. Sales-development agents pricing per-meeting compete directly against outsourced SDR shops charging the same way, which makes the comparison brutal and the value capture clean.

Finance and insurance. Claims adjudication, underwriting, and collections involve large dollar amounts per transaction, which makes value-share viable -- take a percentage of recovered or saved money. The catch is heavy regulation and audit requirements that demand explainability, which raises your cost to serve and compresses the margin you can capture.

Logistics and operations. Dispatch, scheduling, and supply-chain agents create value through efficiency gains that are real but diffuse. Attribution is hard ("did the agent save that route or did fuel prices drop?"), which pushes pricing back toward per-task or subscription. You can't charge for value you can't cleanly prove.

The pattern: verticals with discrete, high-value, attributable transactions (sales, claims, legal) support outcome and value-share pricing. Verticals with diffuse, hard-to-attribute, or vendor-uncontrolled outcomes (logistics, healthcare delivery) get pushed toward per-task and subscription, no matter how much the founder wants the sexier model.

The Margin Problem Nobody Wants to Discuss

There's an uncomfortable fact lurking under every outcome-priced agent: inference costs money, and it costs money whether or not the outcome happens.

Traditional SaaS enjoyed gross margins north of 80% because serving one more customer cost almost nothing. Vertical agents don't get that gift. Every task an agent attempts burns tokens -- sometimes a lot of them, because agentic workflows loop, retry, call tools, and reason in multiple passes. If you charge per outcome but pay per attempt, your margin depends entirely on your success rate. An agent that succeeds 90% of the time has a very different cost structure than one that succeeds 50% of the time and burns compute on every failure.

This is why agent reliability isn't just an engineering concern -- it's a pricing concern. The relationship between your success rate and your unit economics is direct and unforgiving. A vertical agent that improves from 70% to 85% task completion doesn't just delight customers; it can swing the business from negative to positive gross margin under outcome pricing.

The implication for pricing strategy is sharp. Pure outcome pricing transfers all the execution risk to you, the vendor. Every failed attempt is a cost you eat. That's fine if your reliability is high and your model costs are falling (which, broadly, they are -- frontier inference prices have dropped dramatically over the past two years). It's fatal if your reliability is shaky or your workflows are token-hungry. Many founders rushing to "outcome-based" pricing because it sounds investor-friendly haven't run the math on what happens to their margin during the messy early period when reliability is still climbing.

The pragmatic answer is a floor. Charge a base platform fee that covers your fixed costs and a chunk of inference, then layer outcome pricing on top for upside. The floor protects you from the reliability problem; the outcome layer captures the value. This is also simply how mature buyers prefer to purchase -- they want a predictable line item plus variable upside, not a slot machine.

Attribution: The Hidden Tax on Outcome Pricing

Outcome pricing assumes everyone agrees on what the outcome was and who caused it. In practice, both assumptions are fragile.

Take a collections agent that calls debtors. A payment comes in three days after the call. Did the agent cause it? Maybe. Or maybe the debtor was going to pay anyway, or a separate email did the work, or the human supervisor's earlier letter finally landed. Attribution in multi-touch processes is genuinely hard, and your customer has every financial incentive to attribute outcomes to anything other than your agent when the bill arrives.

This creates an adverse dynamic that pure outcome pricing tends to ignore. The customer disputes marginal outcomes. You spend cycles arguing over attribution instead of building product. Trust erodes precisely where you need it. The cleaner your outcome definition -- "a meeting that appeared on the calendar," "a claim that paid," "a document marked privileged that survived audit" -- the less this bites. The fuzzier it is, the more outcome pricing becomes a recurring negotiation rather than a clean transaction.

There's a second-order problem too: adverse selection. If you price per successful outcome, customers will route you the hardest cases -- the ones their humans couldn't crack -- while keeping the easy wins in-house where they don't pay you. Your success rate craters on a portfolio deliberately skewed toward difficulty, and your economics suffer. Sophisticated vertical agent companies counter this by pricing on the full population of tasks (you process everything, easy and hard) rather than only the successes, or by defining outcomes narrowly enough that gaming is hard.

The lesson isn't that outcome pricing is bad. It's that outcome pricing without a crisp, mutually verifiable, hard-to-game outcome definition is a contract dispute waiting to happen. The pricing model is only as good as the measurement underneath it, and in many verticals that measurement is the real product work.

Packaging, Floors, and the Land-and-Expand Path

How you package matters as much as the underlying model. A few principles that hold across verticals.

Lead with a floor. A base platform fee accomplishes three things at once: it covers your inference and support costs, it filters out tire-kickers who'll never convert, and it gives the buyer the predictability their finance team requires. Even outcome-heavy pricing should sit on top of a floor, not replace it.

Make the unit legible. Whatever you charge per -- task, outcome, or value-share point -- the buyer should be able to count it without a spreadsheet and a lawyer. Legibility builds trust and shortens sales cycles. Verticals where the unit is naturally clean (meetings, documents, claims) close faster than those where it's murky.

Price the expansion path in from day one. Vertical agents win land-and-expand: start with one workflow, prove ROI, then absorb adjacent work. Your pricing should make that expansion frictionless -- a customer paying per-task for contract review should be able to add clause negotiation or matter intake without renegotiating from scratch. The deepest moats in vertical agents come from owning more of a workflow over time, and pricing that punishes expansion strangles that motion.

Finally, revisit pricing as reliability improves. The price you can charge at 70% task completion is not the price you can charge at 92%. Many vertical agent companies underprice early (reasonably, to win logos while reliability is still climbing) and then fail to ratchet up as the product improves. The value you capture should track the value you create, and that value is moving -- usually upward -- faster than in any prior software category.

Insights Most People Overlook

The best vertical agent pricing competes against a budget that isn't the software budget. Founders instinctively benchmark against other software and price into the IT line item, which is small and fiercely defended. The winners price into the labor, vendor, or cost-of-error budget -- which is often 10x to 50x larger and controlled by the operator who feels the pain. Same agent, radically different ceiling, depending on which budget you aim at.

Outcome pricing quietly transfers your reliability risk onto your income statement, and most founders haven't modeled it. "Pay only when it works" sounds customer-friendly and investor-friendly, but it means you absorb the full cost of every failed attempt during the exact period -- early product -- when failures are most common. The sequencing is backwards: outcome pricing is a model you graduate into once reliability is high, not one you launch with.

Adverse selection is the silent killer of per-outcome agents. Charge only for wins and customers will feed you the cases their own people gave up on, skewing your portfolio toward the unwinnable. Your success rate -- and your margin -- collapses on a population that was rigged against you. Pricing on the full task volume, not just successes, is the unglamorous fix nobody advertises.

The "right" pricing model is dictated by attribution clarity, not by the founder's ambition. You can want value-share all you like; if your vertical can't cleanly prove the agent caused the outcome, you'll spend your life arguing invoices. Sales and claims support aggressive value capture because attribution is clean. Logistics and clinical delivery don't, no matter how much value the agent actually creates. Pick your model from the measurement reality, not the pitch deck.

Falling inference costs are silently improving your margins faster than your pricing reflects. Token prices have dropped sharply, which means an outcome-priced agent that was barely profitable last year may be comfortably profitable now at the same price -- or could cut price to win share. Few vertical agent companies actively re-underwrite their unit economics as model costs fall, leaving either margin or market share on the table.

References

• The Pricing of AI Agents -- Andreessen Horowitz (a16z)
• The Economic Potential of Generative AI: The Next Productivity Frontier -- McKinsey & Company