Channel and Reseller Economics for Agent Products: Why the Old Margin Playbook Breaks

Table of Contents

• The SaaS Channel Assumption That No Longer Holds
• What a Reseller Actually Sells When the Product Is an Agent
• The Margin Math Problem: COGS Lives Inside the Product
• Four Channel Models for Agent Products
• Resale on a Wholesale Discount
• Referral and Influence Fees
• White-Label and Embedded Agents
• Build-On Partners and Marketplace Take Rates
• The Outcome-Pricing Wrinkle Nobody Warns Partners About
• How to Set a Discount That Survives a Model Price Cut
• Channel Conflict in the Agent Era
• Insights Most People Overlook
• References

The SaaS Channel Assumption That No Longer Holds

For two decades the software channel ran on a clean deal. A vendor set a list price for a seat or a tier, gave a reseller a 20 to 40 percent discount off that list, and the reseller's job was to find the customer, close the deal, handle the relationship, and pocket the spread. The margin was rich because software COGS was near zero, once the code existed, every additional seat cost the vendor almost nothing, so there was plenty of room to share. Microsoft, Salesforce, and the entire VAR ecosystem grew up on this arithmetic. A partner could build a real business on recurring 30 percent margin against a predictable subscription that renewed itself every year.

Agent products break that arithmetic in a specific and unforgiving way: the product has a real, variable cost of delivery that scales with usage. Every task an agent completes burns inference tokens, tool calls, and sometimes third-party API spend. The vendor isn't sitting on 85 percent gross margin with room to spare. Many GaaS companies are running gross margins in the 40 to 70 percent range early on, and a chunk of that is eaten by the model providers. When you ask that vendor to hand a reseller 30 points off the top, you're asking them to give away most of what's left. The channel discount that felt generous in SaaS can be mathematically impossible in GaaS.

This is the core tension, and most channel programs being stood up right now are quietly ignoring it. They copy the SaaS partner agreement, swap the logo, and discover six months later that their best-selling partners are signing deals that lose the vendor money on every high-usage account.

What a Reseller Actually Sells When the Product Is an Agent

Worth being precise here, because the word "reseller" hides several very different jobs. In traditional software a reseller mostly sold access and trust, they were a trusted local face who could vouch for a product and handle procurement. With agent products the value a partner adds shifts toward implementation, integration, and accountability for outcomes.

An agent that resolves support tickets or reconciles invoices doesn't drop in like a SaaS dashboard. It has to be wired into the customer's systems, given the right permissions, tuned against the customer's actual workflows, and monitored so it doesn't quietly go wrong. That work is real and a good partner can charge for it. So in practice the modern agent "reseller" is usually three roles wearing one badge: a seller who sources the deal, an integrator who makes the agent work in the customer's environment, and an operator who stays accountable when the agent's autonomy meets the customer's edge cases.

The implication for economics is large. A partner whose value is "I closed the deal" deserves a one-time or thin recurring cut. A partner whose value is "I keep this agent running correctly across your five back-office systems" is delivering ongoing labor and deserves a structure that pays for it. Lumping both into a flat resale discount is how channel programs end up either overpaying order-takers or underpaying the integrators who actually make the product stick. This is the same accountability question that runs through #57: Outcome-based pricing: who defines and audits the outcome?, when value is fuzzy, so is the fair split.

The Margin Math Problem: COGS Lives Inside the Product

Here is the number that should govern every channel decision a GaaS company makes, and almost nobody calculates it before launching a partner program: the gross margin remaining after inference cost, at the usage level your largest accounts actually run.

Run a simple example. An agent task costs the vendor $0.40 in model and tool spend to deliver. The vendor sells that task to end customers at $1.00, a 60 percent gross margin. Comfortable in a direct sale. Now route it through a reseller on a standard 30 percent off list. The reseller buys the task at $0.70 and sells at $1.00, keeping $0.30. The vendor now collects $0.70, pays $0.40 in COGS, and is left with $0.30, a 43 percent margin before any of the vendor's own sales, support, or R&D cost. On a power user who runs ten times the average volume, that thin margin is the whole account. One model price spike and the vendor is underwater on its biggest channel customers.

That's why the cleanest agent vendors are abandoning "percent off list" entirely for the channel and pricing partners off COGS instead. The discount is sized against the margin pool, not the sticker. The principle is the minimum viable margin every GaaS company has to protect on its own books before it shares anything, a theme covered directly in #95: The "minimum viable margin" every GaaS startup needs. If you don't know your loaded per-task cost down to a few cents, you can't responsibly sign a reseller, full stop. Analysts at firms like a16z have written extensively on how AI gross margins compress the economics compared to classic software, and the channel is where that compression bites first.

Four Channel Models for Agent Products

There isn't one right structure. There are roughly four, and they suit different partner types and different stages.

Resale on a Wholesale Discount

The partner buys at a wholesale rate and resells at whatever the market bears. This works when your COGS is genuinely low and stable, or when the partner is bundling the agent into a larger solution where your line item is a small fraction. The trap, as above, is offering a flat percentage when your margin pool is thin. The fix is a tiered wholesale rate keyed to volume and to the underlying cost structure, and, ideally, a floor that protects you when usage spikes. Some vendors set wholesale as "cost plus a fixed markup" rather than "list minus a discount," which keeps the vendor whole no matter how aggressively the partner prices downstream.

Referral and Influence Fees

The partner sources the deal and hands it to the vendor, who owns billing and the relationship. The partner gets a referral fee, usually a percentage of first-year revenue or a declining multi-year cut. This is the cleanest model for thin-margin agent products because the vendor keeps control of pricing and COGS exposure, and only pays for sourcing. It's underrated. A lot of GaaS companies would be better off running a generous referral program than a true resale channel, especially while their unit economics are still moving.

White-Label and Embedded Agents

The partner rebrands the agent as their own and embeds it in their platform. Here the economics flip toward a platform deal: a minimum commitment, a wholesale per-task or per-outcome rate, and often a revenue share above a threshold. White-label is its own discipline with its own pricing logic, covered in #102: Pricing white-labeled agents for platform partners. The key economic point is that white-label partners are effectively taking on your sales and support cost, so they earn a deeper share, but they also concentrate your COGS risk into a few large accounts you don't control. Set minimums and usage caps, or one runaway partner deployment becomes your largest unhedged cost line.

Build-On Partners and Marketplace Take Rates

The inverse of resale: third parties build agents on your platform, and you take a cut of what they sell. Now you're the one setting a take rate, and the marketplace economics of payment processors and app stores become the reference model. Take rates of 15 to 30 percent are common, but agent marketplaces face a wrinkle that app stores don't, the agents have live COGS, so a 30 percent take on top of inference cost can make the third party's business unviable. That dynamic, and how it shapes platform-versus-agent pricing, is the heart of #73: Marketplace take rates for third-party agents.

The Outcome-Pricing Wrinkle Nobody Warns Partners About

When a vendor prices on outcomes, per resolved ticket, per booked meeting, per collected invoice, the channel inherits a problem that doesn't exist in seat-based software: the partner's margin now depends on the agent actually succeeding. If the agent resolves 70 percent of tickets in one customer's messy environment and 95 percent in another's clean one, the revenue per deployment swings wildly, and so does the partner's cut. A reseller can't underwrite that variance the way they could underwrite a fixed per-seat license.

This pushes channel programs toward a few defensive moves. Some vendors guarantee partners a floor regardless of outcome volume in the first months, absorbing the variance themselves to keep partners willing to sell. Others pay partners on gross usage rather than net outcomes, so the partner is insulated from success-rate swings while the vendor carries the outcome risk against the end customer. Both are reasonable; what's not reasonable is letting a partner sign outcome-priced deals without telling them their commission is now a function of agent performance in environments they don't fully control. The outcome-definition fights described in the broader pricing beat land directly on the channel's lap, which is why partner enablement for outcome-priced agents has to include teaching partners how the outcome is measured and disputed.

How to Set a Discount That Survives a Model Price Cut

Model costs have been falling fast and unevenly. A discount you set today against today's COGS can become a loss-maker or a windfall the moment a provider cuts inference prices or you route to a cheaper model. The fix is to make the channel discount explicitly re-pricable, and to say so in the contract.

Three practical mechanisms. First, peg the wholesale rate to a margin band, not a fixed dollar figure, and reserve the right to adjust as COGS moves, the same grandfather-clause problem that direct customers face, examined in #100: The grandfather problem: repricing as model costs drop, applies doubly to partners who've built their own pricing on top of yours. Second, separate the platform fee from the consumption fee so you can cut consumption pricing (good for competitiveness) without gutting the partner's platform margin (good for loyalty). Third, when you save money through model routing, running the cheap model when the task allows, decide in advance whether that saving flows to the partner, the customer, or your margin, and write it down. Ambiguity here is how channel relationships sour: the partner assumes the savings are theirs, the vendor assumes they're the vendor's, and the first repricing turns into a fight.

Channel Conflict in the Agent Era

Classic channel conflict was about the vendor's direct sales team competing with partners for the same logo. That still happens. But agent products add a sharper version: the agent itself can expand usage automatically, without anyone selling anything. A support agent that handles more tickets as the customer grows generates more revenue with zero partner involvement. Who gets credit, and margin, for that automatic expansion?

If the partner sourced the account and gets a recurring cut of all usage, they earn margin forever on growth they didn't drive, which can feel unfair to the vendor on large accounts. If the partner only gets a one-time fee, they have no incentive to nurture the account toward the expansion that makes agents so lucrative. The land-and-expand dynamic that makes agent products attractive, explored in #71: Land-and-expand when expansion is automatic usage growth, is exactly what makes channel attribution messy. The cleanest answer most vendors land on is a recurring but declining partner share: full cut on sourced revenue, tapering over two or three years, so the partner is paid well for landing the account but the vendor reclaims margin on the automatic growth they're delivering through the product itself.

The honest summary is that the agent channel is still being invented. The vendors getting it right are the ones who start from their real loaded cost per task, decide what job each partner actually does, and build the economics up from there, rather than photocopying a SaaS partner agreement and hoping the margin works out. It usually doesn't.

Insights Most People Overlook

• The referral model beats the resale model for most early GaaS companies, and founders resist it for the wrong reason. They want the "stickiness" of a reseller owning the relationship. But with thin, volatile COGS, handing pricing control to a partner is the fastest way to lose money on your best accounts. Keep billing, pay generously for sourcing, and graduate to true resale only once your unit economics stop moving.

• Your largest channel deals are your highest-risk deals, which is the opposite of SaaS. In seat-based software, a big channel account was pure upside, near-zero marginal cost, fat margin. In GaaS, a big account is where thin per-task margin multiplied by huge volume can flip negative, especially under a flat partner discount. The deal your partner is proudest of may be the one quietly bleeding you. Cap usage or floor the margin on large channel accounts specifically.

• Partners will arbitrage your model-cost savings if you let them. When you route to a cheaper model and your COGS drops, a partner buying on a fixed wholesale rate captures the entire saving as extra margin, and has every incentive not to mention it. Margin-share clauses that automatically split COGS reductions are rare in channel contracts today and will become standard. Write one now.

• The integrator, not the seller, is the partner you actually need, and the channel comp plan is usually built for the seller. Agents fail in production from bad integration and unmonitored edge cases far more than from bad selling. Yet most partner programs pay for the close and treat implementation as an afterthought. Pay the integration and operations work explicitly, or your agents will churn and your "successful" channel will have a retention problem nobody can explain.

• Outcome pricing quietly transfers performance risk onto your partners, and unmanaged, it makes them stop selling. A reseller can't forecast commission when it depends on the agent's success rate in environments they don't control. The vendors who win the channel are the ones who absorb that variance, through floors or gross-usage commissions, rather than letting partners discover it the hard way after their first underperforming deployment.

References

• The Economic Case for Generative AI and Foundation Models, Andreessen Horowitz (a16z)
• Anthropic, Pricing and Model Routing Documentation