Cost-Plus vs. Value-Based: The GaaS Pricing Philosophy Debate

Table of Contents

• The Two Philosophies, Defined Without the Hand-Waving
• Why GaaS Breaks the Old SaaS Pricing Playbook
• The Case for Cost-Plus (And Where It Quietly Wins)
• The Case for Value-Based (And Where It Quietly Loses)
• The Volatile-Inference Problem Nobody Models Correctly
• How to Actually Choose: A Decision Framework
• The Blended Models Winning Right Now
• Insights Most People Overlook
• References

The Two Philosophies, Defined Without the Hand-Waving

Strip away the consulting jargon and the debate is simple. Cost-plus pricing starts at the bottom of your P&L: you tally what it costs to deliver the service, inference tokens, orchestration compute, third-party API calls, human review, support, then you stack a target margin on top. The price is a derivative of your costs. Value-based pricing starts at the top of the customer's P&L: you estimate the economic value the agent creates, hours saved, revenue captured, headcount avoided, errors prevented, and you price as a slice of that value, with little direct reference to what it cost you to produce.

The two are not just different math. They reflect different beliefs about who you are. A cost-plus vendor implicitly says, "I'm a utility, I deliver a unit of work efficiently and take a fair markup." A value-based vendor says, "I'm a partner, I create outcomes, and I deserve to share in the upside." Those are not interchangeable identities, and buyers can smell which one you actually believe.

The classic enterprise software answer was always "value-based, obviously." For two decades, the consulting class treated cost-plus as the amateur's mistake, leaving money on the table by anchoring to your costs instead of the buyer's willingness to pay. That orthodoxy is encoded in every pricing textbook, and it's where most GaaS founders start. The problem is that the orthodoxy was built for software with near-zero marginal cost. Agents are not that.

Why GaaS Breaks the Old SaaS Pricing Playbook

Here is the structural fact that makes this debate live again: agents have real, variable, and volatile marginal costs. A SaaS seat costs the vendor essentially nothing to provision, the 80%+ gross margins of classic software came from selling the same bytes a thousand times. An agent, by contrast, burns tokens on every task. A complex multi-step research agent can chew through hundreds of thousands of tokens, hit paid APIs, and occasionally route to a frontier model that costs 20x the cheap one. Your cost of goods sold is no longer a rounding error. It moves with usage, and it moves with the model provider's price list.

This is why the pure value-based purism that worked for SaaS becomes dangerous in GaaS. If you price purely on value and ignore cost, you can sell a deal that's wildly profitable on paper and quietly margin-negative on the heaviest users, the classic trap that metered-vs-flat decisions create (a tension explored in #88, the economics of "unlimited agent" plans). The unlimited-plan power user who runs the agent 400 times a day can turn your best logo into your worst unit-economics story.

It also reopens the case for cost-plus, which SaaS had buried. When your COGS is real and visible, anchoring to it isn't amateurish, it's risk management. The most sophisticated GaaS pricing I've seen treats cost not as the basis for price but as a floor that price must always clear. That's a meaningfully different stance than either pure philosophy.

The broader shift here, agents charging for work done rather than access granted, is the through-line of the entire pricing-and-monetization beat, and it changes the sales motion itself (see #68, how agent pricing changes the sales motion entirely). You cannot bolt a 2010s SaaS pricing page onto a 2026 agent product and expect the economics to hold.

The Case for Cost-Plus (And Where It Quietly Wins)

Cost-plus gets dismissed too fast. Its real virtues:

It's margin-safe by construction. If your price is always cost plus a target percentage, you mathematically cannot sell a money-losing deal. In a market where inference prices swing and usage is unpredictable, that's not nothing. A founder who has watched a single enterprise pilot blow through its compute budget learns to love a pricing model that can't go underwater.

It's transparent, and transparency is becoming a buyer demand. A growing cohort of buyers, especially FinOps and procurement teams, wants to see the cost stack. Showing token counts and pass-through compute (the transparency question taken up in the broader pricing-transparency discussion) builds trust precisely because it signals you're not gouging. Cost-plus pairs naturally with that posture.

It's defensible in commoditized verticals. If your agent does something where the value is genuinely hard to attribute, a transcription cleanup agent, say, or a routine data-entry bot, buyers will not believe your "we create $50k of value" story. They'll benchmark you against the raw model cost plus a reasonable markup, and you'll lose the argument if you pretend otherwise.

Where cost-plus quietly wins is the high-volume, low-differentiation middle of the market: agents that are useful but not magical, sold to cost-conscious buyers who can do the math. McKinsey's work on AI economics keeps returning to the point that as capability commoditizes, the durable margin shifts to workflow integration and proprietary data, not to the inference itself. In those zones, a clean cost-plus model with a real markup is honest and durable.

The downside is just as real: cost-plus caps your upside. You will never capture the deal where your agent saved a customer $2 million, because your price is tethered to your own costs. You've volunteered to be a utility.

The Case for Value-Based (And Where It Quietly Loses)

Value-based pricing is where the venture-scale outcomes live. When an agent genuinely replaces a function, a support agent resolving tickets, a sales-development agent booking meetings, a coding agent shipping features, the value created can dwarf the delivery cost. Pricing as a slice of that value lets you grow with the customer's success instead of capping out.

a16z and other investors have argued for years that the most valuable AI businesses will price on outcomes because outcome-based models align vendor incentives with customer results and unlock willingness-to-pay that per-seat or per-token models leave on the table. When it works, it's the strongest position in the market: the customer happily pays more because they're getting more, and your revenue compounds with their growth (the land-and-expand dynamic of #71).

But here's where the purists get quiet about the failure modes:

Attribution is brutal. To charge for value, you have to prove value, and outcomes are usually co-produced. Did the agent book the meeting, or did the marketing campaign warm the lead? Buyers contest attribution the moment the invoice arrives, and you end up litigating credit instead of collecting it.

Volatile costs can strand you. This is the failure cost-plus advocates point to. If you sold a fixed value-based price, "we charge $4 per resolved ticket", and the underlying model price spikes or a customer's tickets get systematically harder, your margin can evaporate while your price stays frozen. You took the cost risk without pricing for it.

It punishes you in commoditized segments. If a buyer can replicate 80% of your value with a raw model API and a weekend of prompting, your "share of value" story collapses. Value-based pricing only holds where the value is real, attributable, and hard to self-serve.

The honest summary: value-based pricing is the right philosophy when you can defend the value, and a liability when you can't.

The Volatile-Inference Problem Nobody Models Correctly

This deserves its own section because it's the single biggest reason the cost-plus-vs-value debate is harder in GaaS than anywhere else. Inference costs are not stable. Frontier-model prices have historically fallen fast, often dramatically year over year, which sounds like good news but creates a nasty repricing problem (the "grandfather problem" of #100). At the same time, within a given product, costs spike unpredictably: a hard task routes to an expensive model, a user triggers a long agentic loop, an upstream API raises rates.

Pure value-based pricing assumes your costs are roughly fixed and ignorable. They aren't. Pure cost-plus pricing assumes you can cleanly pass costs through. You often can't, because buyers hate invoices that move and FinOps teams budget annually. So both pure philosophies are quietly lying about the cost curve.

The vendors getting this right treat cost volatility as a first-class design input. They route to the cheapest model that clears the quality bar (margin expansion via model routing, #87), they cap usage to bound the downside (#80), and they price with a floor that protects margin even when value attribution is fuzzy. In other words, they refuse to be purists. The cost curve forces a blend.

How to Actually Choose: A Decision Framework

Skip the ideology. Ask four questions:

1. Can you defensibly attribute the value? If yes, lean value-based. If the buyer will contest attribution every cycle, lean cost-plus. The clarity of the value story matters more than the size of the value.

2. How volatile and material is your COGS? High and material (heavy multi-step agents, frontier-model dependence) pushes you toward cost-plus protection or a cost floor under any value price. Low and stable pushes you toward freedom to price on value.

3. How commoditized is the capability? If buyers can self-serve most of it with a raw API, value-based collapses, price near cost-plus and compete on integration and reliability. If you're genuinely differentiated, you've earned the right to price on value.

4. Who's buying, and how do they budget? FinOps and procurement want predictability and cost transparency (cost-plus-friendly). A line-of-business owner chasing an outcome will happily pay for value (value-based-friendly). The buyer's mental model should shape your pricing model, not just your pitch.

Notice that none of these questions returns "always value-based." The SaaS-era default is simply wrong as a starting assumption for agents. The right default is blended, with the mix set by your specific answers.

The Blended Models Winning Right Now

In practice, almost every durable GaaS pricing model I've examined is a hybrid that borrows the protection of cost-plus and the upside of value-based:

• Floor-and-ceiling pricing: a cost-derived floor that guarantees margin, plus value-linked pricing above it, capped so the customer's budget risk stays bounded (#67).
• Value-priced with a cost guardrail: you charge per outcome, but a usage cap or per-task cost ceiling protects you from the margin-negative power user.
• Hybrid base-plus-usage: a value-anchored base subscription that covers your fixed costs and signals partnership, plus metered usage that tracks your variable inference cost (#59).
• Tiered by autonomy or seniority: higher-autonomy or "senior" agents priced on value, routine "junior" agents priced closer to cost (#79 and #89).

The common thread: cost sets the floor, value sets the ceiling, and the model lives in between. Harvard Business Review's longstanding guidance that pricing should reflect customer-perceived value rather than internal cost still holds as the aspiration, but in GaaS, the cost floor is what keeps the aspiration from bankrupting you. The philosophy debate, properly understood, isn't a war. It's a negotiation between your P&L and your buyer's, and the winning vendors negotiate both sides explicitly.

Insights Most People Overlook

1. Your pricing philosophy is a churn predictor, not just a revenue lever. Cost-plus deals churn for boring reasons (a cheaper utility shows up). Value-based deals churn catastrophically (the customer disputes the value, stops believing the attribution, and rips you out in a budget review). If you go value-based, you're not just selling harder, you're signing up to prove value continuously, forever. Most founders underestimate that operational tax.

2. The "leave money on the table" critique of cost-plus is half wrong in GaaS. In zero-marginal-cost SaaS, cost-plus genuinely left money behind. In GaaS, a disciplined cost-plus markup on heavy usage can out-earn a value-based price that got renegotiated down after the first attribution fight. Capturing 100% of a defensible markup beats capturing 30% of a disputed value claim.

3. Falling inference prices secretly favor value-based pricing, but only if you don't pass the savings through. As model costs drop, a cost-plus vendor's price mechanically drops with them, surrendering the windfall to customers. A value-based vendor keeps the price tied to outcome value and pockets the widening margin. The catch: do this too visibly and you trigger the grandfather problem when customers notice. The play is to be value-based on price and quietly cost-optimized on delivery.

4. Buyers reverse-engineer your philosophy from your pricing page. Show token counts and you've told the market you're cost-plus, inviting commodity comparison. Show "$ per outcome" with no cost detail and you've told them you're value-based, inviting attribution scrutiny. The page itself is a strategic disclosure, not a neutral menu, which is exactly why writing one that doesn't scare buyers (#91) is its own discipline.

5. The cost floor is the most underused concept in agent pricing. Almost everyone debates the headline model (per-task vs. per-outcome vs. per-seat) and almost nobody designs the floor underneath it. A simple rule, "no deal prices below COGS + minimum viable margin" (#95), quietly prevents the discounting death spiral that kills early GaaS vendors, regardless of which philosophy sits on top.

References

• The economic potential of generative AI: The next productivity frontier, McKinsey & Company
• The Economics of Generative AI, Andreessen Horowitz (a16z)
• The Good-Better-Best Approach to Pricing, Harvard Business Review