The financial AI operations & technology brief.

The numbers first, because they are the whole story. In FY26 a Cloud Solution Provider indirect reseller earned a flat Core incentive of 3.75% on every Microsoft 365 subscription it billed and 4.0% on every Dynamics 365 subscription, whether the account grew, shrank, or sat still. Effective July 1, 2026, that Core incentive is gone. In its place the Growth Accelerator rose from 7.5% to 12.5%, but it pays only on year-over-year growth in Microsoft’s strategic product tiers. The Tier 1 strategic accelerator on billed revenue (Business Premium, E3) fell from 4% to a flat 2.5%; Tier 2 (E5, Copilot) held at 7%. Azure kept its 3% consumption rate and replaced a single 7.5% growth rate with a tiered 7% / 10% / 12% ladder, the top rung reserved for databases and Fabric, the middle for Foundry, GitHub, Sentinel and Copilot Studio.

Then came October. On October 1, 2026, growth margins went live for CSP distributors and direct-bill partners: a discount earned transaction by transaction for new-to-offer sales, seat expansion and adoption of selected AI workloads, validated through an eligibility API that opened on September 23. On September 30 the Microsoft 365 E5, E7 and Copilot promotions ended “with the transition to growth margins.” A Copilot Cowork activation bonus launched September 1. And the Change of Channel Partner rule now means a partner who wins an account from a competitor earns incentives on none of the existing revenue for twelve months, only on what it adds.

The arithmetic on a $2 million book. Take an indirect reseller with $2 million of annual Microsoft 365 billings, all in Tier 1 products, and model only the published US rate card, nothing else. Flat book: FY26 paid $155,000 (Core 3.75% plus the 4% strategic accelerator); FY27 pays $50,000. That is a 68% cut for standing still. Grow the book 10%: FY26 $185,500 against FY27 $80,000. Grow it 20%: $216,000 against $110,000. The richer growth rate never catches the lost base, because 12.5% of a $400,000 delta is $50,000 and 3.75% of the whole $2 million was $75,000 before the Tier 1 cut is counted.

Move the same book into Tier 2 products, E5 and Copilot, and the shape softens but does not flip: flat, $215,000 becomes $140,000, a 35% cut; at 20% growth, $288,000 becomes $218,000. Run the model to its limit and there is no growth rate under 200% at which the FY27 schedule pays what the FY26 schedule paid. On the rate card alone, every partner lost money. The partners who lost least are the ones selling the AI SKUs and growing fastest.

Two honest caveats on that model, because the number is only as good as its edges. First, it excludes the reseller margin from the distributor, which does not change and remains the larger line for most partners; the incentive is a rebate layered on top of it. Second, it excludes everything Microsoft launched to replace the Core money: growth margins, the Cowork activation bonus, Frontier Accelerate. Those programs are real. They are also conditional. Each pays for a specific motion a partner must perform, with eligible SKUs, minimum seat counts and earning caps. The flat 3.75% paid for existing.

Why this is bigger than one vendor. Microsoft is not inventing a trend; it is the largest channel in software confirming one. DigitalRoute surveyed 631 executives across six markets in April and May 2026: 83.4% are exploring usage or outcome-based pricing, 49.1% are already live or optimising usage-metered pricing, 32.2% run hybrid subscription-plus-usage, and 24.6% have outcome-based pricing in place. The same survey found 67% manually fix more than 5% of their usage invoices before sending them, 75.4% see customer disputes on usage charges at least occasionally, and only 21.7% rate their billing and order-to-cash stack as very ready.

Hold those two facts together. A partner paid on a vendor’s revenue is paid on a number the vendor increasingly cannot bill cleanly. Microsoft’s own September notes describe a “continued shift toward consumption,” a flex spend plan now in the CSP sandbox as the construct for “several other consumption products in the future,” and a new Copilot commercial model in which user subscriptions cover everyday use “while usage-based billing gives customers flexibility to access advanced capabilities.” When the base is a meter and the payout is a rate on the growth of that meter, the partner’s income is a second derivative. It cannot be forecast from the contract. It can only be observed after the invoice, and a quarter of the time the invoice is in dispute.

The structure that does not need a July rewrite. Compare the alternative Issue No. 1 ran the math on: a 50% lifetime revenue share. It is indifferent to the seat-versus-usage question, because it is a share of whatever the customer actually pays, not a rebate schedule the vendor publishes each fiscal year. It pays on the book and on the growth at the same rate, because growth is simply more of the book. It carries no change-of-partner clock, no strategic tier list, no eligibility API. A $500-a-month account held for two years pays the partner $6,000 under a lifetime half-share; the same account under a typical 20%-for-twelve-months referral tier pays $1,200. Five times the money, and none of it depends on which products the vendor decided to call strategic that year.

The market benchmark for what partners are actually offered sits far below that. Across PartnerStack’s top 25 vendors in 2025, the best-performing commission offers were 20%, 25% and 30%; IT infrastructure vendors paid 30 to 35%. The same report put the network at $2.7 billion in all-time GMV after a 52% jump in transaction volume, with paid customer conversions up 112% and mid-market and enterprise accounts now 78% of volume. Partners are delivering buyers who pay. The question a lifetime share answers is how long the partner keeps getting paid for that.

The discipline it requires is the part vendors skip. A 50% share for life is only survivable with a quality gate on who gets it, because an uncapped liability attached to a partner who sends churn is a slow leak. That is why Issues 1 and 3 argued the gate is the product, and why the North Star model runs entry through an alignment interview instead of an open form.

The counterargument, and it is strong. Microsoft’s own October 1 note says the total partner investment commitment is unchanged; what changed is how it is targeted. Read generously, that is not a cut but a reallocation from partners who collected a rebate on renewals to partners who create net-new value, and the indirect reseller’s 12.5% growth rate is the richest in the channel, 2.5 points above direct bill. The old Core rebate had a real cost: it funded customer-swapping on price, which is exactly what the twelve-month change-of-partner rule exists to stop. A vendor that pays for growth gets growth. SoftwareOne’s channel director called it what it is: a filter, applied to roughly 500,000 partners.

The data has edges too. DigitalRoute sells usage-billing software and PartnerStack sells partner-management software; both surveys describe markets their authors profit from. And a lifetime revenue share is expensive for the vendor in a way a rebate schedule never is. It only closes at high gross margin, on a product with real retention, with the partner roster policed. For a vendor with thin margins or weak retention, Microsoft’s structure is the rational one, and that is most vendors.

The decision framework. If any part of your income is a vendor’s incentive schedule, three questions settle the next two quarters. First: what share of your partner income is paid on the book versus the delta, today, after July 1? If you have not re-run the number since the Core rebate ended, your FY27 forecast is wrong by roughly the figures above. Second: for each vendor you resell, is the base you are paid on a subscription or a meter, and who audits the meter? Where the answer is a meter and the auditor is the vendor, you are carrying their billing-quality risk on your margin line. Third: which of your revenue lines would survive a rate card you did not write? Lifetime shares, retained services, owned relationships survive. Rebates do not, and the largest channel in software just said so in writing.

Where this plugs in. The North Star Affiliate Network is built on the structure this issue argues for: a straight 50% lifetime recurring share across every asset in the ForgedOps.Ai suite, attributed by named relationship rather than by cookie or by fiscal-year rate card, with no upfront cost. Entry runs through a short Alignment Interview, which is the quality gate that makes an uncapped lifetime share survivable for the vendor and worth having for the partner.

Take the Alignment Interview — North Star Affiliate Network →

Sources: Microsoft Partner Center announcements, September 2026 (growth margins general availability October 1, 2026; eligibility API September 23; E5/E7/Copilot promotions ending September 30; Copilot Cowork CSP Activation Incentive September 1; flex spend plan sandbox; new Copilot commercial model), learn.microsoft.com; Microsoft partner news, October 2026, on the October 1 FY27 incentives portfolio update; FY26-to-FY27 CSP rate tables for US indirect resellers as published by Pax8 (July 28, 2026, updated August 4) and SoftwareOne (July 30, 2026), which summarise Microsoft’s incentive guide; DigitalRoute, From Usage to Revenue 2026 (631 executives, six markets, April–May 2026); PartnerStack Research Lab, network report (February 2026). The $2 million book model is the author’s arithmetic on the published rate card only; it excludes distributor margin, growth margins, activation bonuses, Frontier Accelerate and earning caps, and is illustrative rather than a forecast. Figures verified against the cited sources within the week of October 19, 2026.

Oracle reported its fiscal first quarter on September 10. Total revenue rose 30% to $19.3 billion. Inside that number is a split worth reading slowly: cloud infrastructure revenue grew 121% to $7.4 billion, while cloud applications grew 10% to $4.2 billion. Remaining performance obligations reached $664 billion, up $209 billion year over year, with more than $30 billion of additional AI cloud contracts booked in the quarter. Capital expenditure was roughly $28 billion; free cash flow was negative $5 billion.

A week earlier, on September 3, Nvidia agreed to acquire Hugging Face for $12.93 billion — the platform where more than 18 million developers share over 3 million models, and where more than 200,000 companies discover and deploy AI. The same day, Crusoe was reported to have raised more than $3 billion at a roughly $30 billion valuation, ten months after raising $1.38 billion at $10 billion.

Why the split matters more than the total. Oracle sells both the infrastructure and the applications, to overlapping customers, through the same company. That makes the 121%-versus-10% gap close to a controlled comparison rather than a story about one vendor out-executing another. When enterprises spent with Oracle this quarter, the infrastructure line grew roughly twelve times faster than the application line.

What infrastructure economics do to a channel. Application licences built the modern partner economy: a recurring subscription, a percentage to the reseller, services wrapped around the implementation. Infrastructure is bought differently — multi-year committed contracts, negotiated directly, constrained by capacity rather than by demand. Oracle reported 97.9% utilisation on its AI infrastructure and, per its earnings slides, a 20% average premium on GPU renewal and resale contracts. A scarce input that reprices upward at renewal does not need a channel to create demand for it.

The discovery layer changed owners too. Hugging Face is where a developer chooses a model before choosing where to run it. Nvidia says the platform “will remain an open platform for the entire AI ecosystem” and that developers will keep choosing “the clouds and inference service providers they want.” Take that at face value; the economic point holds either way. The default path from picking a model to paying for compute now starts on a hardware vendor’s property.

The counterargument, and it is strong. Infrastructure is bought ahead of workloads, and applications are where that spend eventually has to earn a return. Oracle’s own framing points there: “Virtually all of Oracle’s enterprise customers want to use AI to reason on their private data and to use AI agents to automate their business processes.” RPO is multi-year contracted backlog, not this year’s revenue; negative free cash flow says the build is being financed ahead of the income; and the Crusoe figures are reported, not confirmed by the company. Read the split as a timing signal about where money is going first, not as a verdict on the application layer.

The decision framework. If your income is a share of application revenue, the next two quarters turn on three questions. First: what does your client spend on AI in total, not just on the SKU you are paid on? If you only see your line, you are measuring the slow half of the budget. Second: can you be paid on, or at least advise on, the layer that is growing — where a workload runs, what capacity it commits to, what that commitment costs at renewal? That is judgment a model hub and a connector library do not supply. Third: when the application finally has to justify the infrastructure bill, will you be the one measuring the return? That is the seat that reopens when the build-out phase ends, and it goes to whoever already has the numbers.

Sources: Oracle fiscal first-quarter 2027 results (September 10, 2026), as reported and in Oracle’s earnings presentation; NVIDIA’s announcement of its agreement to acquire Hugging Face (September 3, 2026); Crusoe financing per Bloomberg and TechCrunch reporting (September 3, 2026), not confirmed by the company. RPO is multi-year contracted backlog, not current-period revenue. Growth-rate comparison is the author’s arithmetic on reported figures. Figures verified against the cited coverage within the week of September 7, 2026.

Workato released Otto for Everyone, putting its AI “superagents” in front of individuals and teams through Slack, the web and SMS, able to act across Gmail, Salesforce, GitLab and more than 1,400 business applications. Skyflow shipped a data-control layer for Glean, aimed at governing sensitive data as it moves through CRM systems, SaaS apps, data lakes and internal wikis. Runable raised $21 million. Logicalis reported that 60% of global CIOs intend to invest in AI agents within twelve months.

Read those together and the shape is not “more agents.” It is agents changing their point of entry.

Why the entry point is the whole economic story. An enterprise agent platform is bought the way platforms are always bought: a committee, a security review, an implementation partner, a twelve-month cycle. That cycle is where the channel lives. The partner runs the evaluation, does the integration work, and earns on both the licence and the services.

An agent that arrives in Slack is bought the way SaaS was bought in 2013 — by the person who needs it, on a card or a departmental budget, with no committee and no integration project. That motion has no natural place for a partner in it. It also has no natural place for a security review, which is precisely why Skyflow’s governance layer showed up in the same week: the moment agents reach individuals, someone has to answer for what data they touched.

The 1,400-connector number is the part to sit with. Integration breadth used to be the partner’s moat. Knowing how to wire Salesforce to GitLab to a data warehouse was billable expertise. A vendor shipping 1,400 pre-built connectors is not adding a feature; it is buying out that expertise and folding it into the licence.

That does not eliminate the partner. It relocates the value. What a connector cannot do is decide which processes should be agent-run, prove the outcome afterwards, or carry the governance answer to an auditor. Those are judgment, not plumbing, and they are the part of the work that a 1,400-connector library makes more valuable rather than less.

The counterargument, and it is strong. Sixty percent of CIOs stating an intention twelve months out is a survey, not a purchase order. Intent surveys in this category have historically overshot delivery by a wide margin, and “plan to invest” covers everything from a pilot to a line item. Treat it as a directional signal about attention, not as a forecast of spend. The Workato and Skyflow releases are the firmer evidence, because a shipped product is a company spending its own money on a thesis.

The decision framework. If your revenue depends on this category, the question for the next two quarters is not whether agents grow. It is which half of the work you are on.

Ask three things. Does your engagement bill for integration labour that a connector library now performs for free? If so, that line is being deprecated on someone else’s roadmap. Can you produce a defensible answer to “what data did the agent touch, and under what authority” without the vendor’s telemetry? That is the question governance layers exist to answer, and being able to answer it independently is a durable position. And finally: when the agent enters through Slack rather than through procurement, does anyone in your commercial motion ever meet the buyer? If the honest answer is no, the motion needs rebuilding before the volume arrives, not after.

Sources: product releases and funding per AI agent industry coverage for the week of August 24–30, 2026; CIO intent figure per Logicalis as reported in the same coverage. Survey-based figures are identified as such and are not treated as committed spend. Figures verified against the cited coverage within the week of August 31, 2026.

The finding, in Rippling’s own framing: for AI-native SaaS companies where inference is the largest variable cost, model selection is a gross-margin decision, not an engineering one. Fifteen models, live payroll data, quality flattening out across the top of the field while cost per run spanned $621 to $4,359 — a seven-fold spread for output a reviewer could not meaningfully separate.

Most software businesses have never had a variable cost that behaves like this. Hosting scales predictably. Support scales with headcount. Inference scales with usage and with a vendor decision you can change on a Tuesday.

What this does to the shape of a P&L. Median software gross margin sits near 80%, blended closer to 76% once services are counted. Those benchmarks were set by businesses whose cost of revenue was hosting and support. An AI-native product with inference as its largest variable cost is a different animal wearing the same label — and it can land anywhere from comfortably above that median to well below it, on the same revenue, depending on a routing choice.

That is why the same week produced Stigg acquiring Received.ai for usage-based billing infrastructure. Metering is becoming plumbing, because you cannot manage a variable cost you cannot attribute per customer. A business that bills a flat subscription while carrying a variable inference cost has written an unhedged option against its own margin, and the strike moves every time a customer gets more active.

Where this touches a partner or reseller directly. If you resell an AI-native product on a percentage of licence revenue, your economics are downstream of the vendor’s model choices and you have no visibility into them. A vendor under margin pressure has an obvious lever: route to a cheaper model. If quality genuinely plateaus, the customer never notices and nothing changes for you. If it does not, the degradation lands in your account, on your relationship, and you will hear about it before the vendor does.

The counterargument. One company, one workload. Payroll is unusually structured, unusually verifiable, and unusually tolerant of a smaller model — the answer is either arithmetically right or it is not. Quality is far more likely to plateau there than in open-ended generation, legal reasoning or anything requiring long-context synthesis. Do not read “quality plateaued” as a general law; read it as a test worth running on your workload, because the spread it revealed is large enough to be worth an afternoon.

The decision framework. Three questions, in order.

First, do you know your inference cost per customer, per month? Not in aggregate — per customer. If not, you cannot tell a profitable account from an unprofitable one, and the usage-based billing infrastructure being bought up this week exists precisely because that gap is common. Second, has anyone run the cheaper model against your actual workload with a quality bar written down in advance? Running it afterwards invites the result you want. Third, if you resell someone else’s AI product: does your agreement say anything at all about the model behind it? Almost none do. That silence is the vendor holding an option you are exposed to.

Sources: Rippling model-comparison figures and framing, plus funding and M&A activity, per This Week in SaaS, August 18–24, 2026; software gross-margin benchmarks per 2025 median software and blended total-revenue margin data. Figures verified against the cited coverage within the week of August 24, 2026.

The number to sit with this week is $120 per seat. That is the enterprise price attached to the Grok-based “AI coworker” offering — rented, per seat, per month, the same unit of account that has priced business software for twenty years. Not per resolved ticket. Not per hour of work displaced. Per seat.

Hold that against what else printed in the same seven days. Cognition was reported in talks at a roughly $40 billion valuation, with its coding agent Devin described as approaching $1 billion in annualized revenue. Stripe was reported to have acquired the model-routing platform OpenRouter for north of $7 billion. OpenAI put $150 million into a partner program and stood up DeployCo to embed forward-deployed engineers inside enterprise accounts. Microsoft moved to pipe verified business data from S&P Global and ZoomInfo into Copilot workflows. Synchrony and OpenAI announced a collaboration aimed squarely at agentic commerce.

That is a market assembling the plumbing for outcome-based pricing — routing, verified data, deployed engineers, payment rails — and then quoting the product in seats anyway.

Why this matters more to a partner than to a vendor. A per-seat product is resellable on the motion you already have. It has a countable unit, a renewal date, a per-unit margin, and an expansion path that looks like headcount. Every compensation plan, every quota, every partner tier in the channel is built on exactly that shape. If agents price in seats, the existing partner economy keeps working — discounted resale, co-sell, managed service on top.

An outcome-priced product is a different animal. If the vendor charges per resolved case or per dollar of cost removed, there is no seat to resell and no headcount to expand into. The margin moves to whoever can attribute the outcome, which is usually the vendor’s own telemetry, not the partner’s relationship. That is the version of this transition where the channel gets disintermediated, and it is the version that did not arrive this week.

The counterweight, and it is a real one. One number in the same week points the other way. Grab reported that AI took mechanical analytics work from 44% to 30% of the job — a fourteen-point reduction in labor content. That is precisely the statistic a CFO uses to argue that a seat is the wrong unit: if the tool removes a third of the work, why is it priced like a person?

Per-seat agent pricing is therefore a truce, not a settlement. It holds while buyers still measure adoption in logins. It breaks the first time a large buyer indexes renewal price to displaced labor and wins that negotiation publicly. Nothing this week suggests that has happened yet. Everything this week suggests the tooling to make it possible is being funded at multi-billion valuations.

Note also what the OpenAI move implies. A $150 million partner program with forward-deployed engineers is not a channel program in the classic sense — it is the vendor putting its own people inside the account to do the implementation work a systems integrator would normally sell. Read alongside per-seat pricing, the message is consistent: the vendor wants the seat revenue and the implementation surface, and is willing to fund both directly.

The decision framework. For anyone whose revenue depends on reselling or servicing this category, three questions are worth answering before renewal season:

First, does your agreement price in the same unit as your vendor? A partner earning a percentage of seat revenue on a product drifting toward consumption or outcome pricing is holding a contract that quietly stops paying. Second, can you attribute an outcome yourself, from your own data, without the vendor’s telemetry? If not, you cannot follow the pricing model when it moves. Third, what share of your book is implementation labor that the vendor is now funding itself? That is the line item OpenAI just put $150 million behind.

The pricing model is the whole negotiation. It was decided this week, provisionally, in your favour. Provisionally is the operative word.

Sources: reported enterprise pricing and funding figures per AI Agents News Brief, August 17, 2026; Grab efficiency figure as reported in the same brief. Valuations and acquisition amounts are reported figures and were not confirmed by the parties; where a deal was described as “in talks” or “reported,” it is characterised that way here. Figures verified against the cited coverage within the week of August 17, 2026.

Bending Spoons agreed to acquire Airtable for approximately $1.285 billion, all cash — a multiple of roughly 2.7x against about $480 million of ARR. Airtable is not a distressed asset. It is a well-known, widely deployed, category-defining product with real enterprise penetration.

In the same seven days: HappyRobot raised $150M at a $1.2 billion valuation. Sapiom raised $35M for agent infrastructure. Ordway took $20M in growth capital and said explicitly it was doubling R&D on AI agents. Ambrook raised $30M. Omilia raised $67M. Naïve raised $28.5M having grown ARR tenfold in six months. Klaviyo bought Agency for its AI customer-success team.

What the spread actually says. Two point seven times ARR for a mature, high-retention SaaS book, against double-digit forward multiples for pre-scale agent companies, is a market drawing a hard line between revenue that persists and revenue that compounds. Persistent revenue is now priced close to a cash-flow annuity. Compounding revenue is priced as an option.

That is a defensible position, not a bubble artifact. A high-retention seat book in a mature category has a knowable terminal value and very little optionality left. An agent product growing 10x in six months has almost no terminal value you can underwrite and enormous optionality. The buyer of the first is buying cash flows; the investor in the second is buying a distribution of outcomes.

Now apply it to a partner’s book. Most channel revenue is structurally the first kind: renewal-dependent, seat-denominated, high-retention, low-expansion. For years that was the good problem — predictable, bankable, easy to forecast. This week is a reminder that the market has started to discount exactly that profile, and that the discount is now visible in a headline transaction rather than buried in a private comp set.

The practical read is not “abandon renewals.” It is that a book made entirely of renewals is being valued as an annuity, and annuities do not command strategic premiums. If a partner business intends to be worth a multiple rather than a yield, some share of its revenue has to carry expansion that is attributable to the partner — not to the vendor’s product-led growth.

The counterargument, stated fairly. One deal is one deal, and this one has a specific buyer attached. Bending Spoons runs a buy-and-optimise playbook: acquire established software, cut cost, run it for cash. A 2.7x all-cash price reflects that playbook and that cost of capital as much as it reflects the asset. A strategic acquirer with a product adjacency might have paid materially more. Read this as a floor set by financial buyers, not as the market’s ceiling for durable SaaS.

It is also worth being precise about what 2.7x is measured against. ARR is not revenue quality. Median software gross margin sits near 80%, blended closer to 76% once services are included — and the gap between those two numbers is exactly where a services-heavy book gets repriced downward regardless of what the ARR line says.

The decision framework. Split your book into three buckets and size each one honestly. Bucket one: renewal revenue you would keep if you stopped selling tomorrow. Bucket two: expansion revenue inside existing accounts that you can attribute to your own motion. Bucket three: services revenue priced by the hour. Bucket one is an annuity and will be valued like one. Bucket three is labour and will be valued like that. Bucket two is the only part of the book being repriced upward this cycle, and it is the only one worth deliberately growing.

A useful sanity check while you are in there: healthy demo-to-close conversion runs 10–20%. Below 8–10% and the problem is upstream of pricing.

Sources: transaction, funding and valuation figures per This Week in SaaS, August 4–10, 2026; software gross-margin benchmarks per 2025 median software and blended total-revenue margin data. The Airtable multiple is derived from the reported ~$1.285B consideration against ~$480M ARR and is approximate. Deal terms not disclosed by the parties are characterised as reported. Figures verified against the cited coverage within the week of August 10, 2026.

The shortlist is being written before you know the deal exists

Start with the demand side, because the numbers have moved faster than most partner programs have. Forrester's 2026 buyer research finds 94% of B2B buyers used AI during their most recent purchase process, up from 89% a year earlier — with 55% using AI tools to compare vendors, 54% to research products, and 47% to build the internal business case before engaging any vendor. G2's data says 51% of B2B software buyers now start their research with an AI chatbot, and — the number worth pinning to the wall — generative-AI chatbots are now the single most influential source for vendor shortlists at 17.1%, ahead of software review sites (15.1%), vendor websites (12.8%), and even direct peer recommendations (8.9%).

Read that list again as a referral economist rather than a marketer. Review sites, vendor sites, and peer word-of-mouth are the three surfaces the affiliate and partner economy has monetized for twenty years. All three just got outranked by a surface none of them controls — and one that, today, pays no commission to anybody.

The click that never comes

The mechanism doing the damage is specific and measurable. When Google shows an AI Overview, Pew Research finds users click an organic result on just 8% of visits, versus 15% when no Overview is present — roughly half the clicks, gone. Clicks on the sources cited inside the Overview? About 1%. Similarweb's tracking shows zero-click rates in the news segment jumping from 56% to 69% in the twelve months after AI Overviews launched in the US, with publishers losing an estimated 600 million visits a month.

Now map that onto the affiliate economy. The classic content affiliate — the "best CRM for contractors" listicle, the comparison blog, the review roundup — monetizes by intercepting a search, earning the click, and passing it along with a tracking parameter attached. Every step of that chain assumes the click happens. When the model reads the listicle, synthesizes the answer, and the buyer never visits the page, the affiliate did the work, influenced the decision, and generated a commission for no one. That's not a traffic problem. It's the referral layer's billing system failing while the referrals themselves still occur.

The traffic that survives converts like a referral, not a click

Here's the twist that keeps this from being a simple obituary. The visitors AI assistants do send are extraordinary. Adobe Analytics has AI-referred traffic to US retail sites up 393% year-over-year in Q1 2026 — and the quality flipped even faster than the volume grew. In March 2025, AI-referred visitors converted 38% worse than other channels; by March 2026 they converted 42% better — an 80-point swing in twelve months. Over the 2025 holiday season, AI referrals converted 31% higher than non-AI sources with revenue per visit up 254%.

The explanation is the one every good partner already knows: the qualification happened upstream. A buyer who arrives after a long back-and-forth with an assistant has already compared, shortlisted, and objection-handled — the visit is the end of the funnel, not the top. Which is to say: AI-referred traffic behaves like a warm personal referral, because structurally that's what it is. The model played the role the trusted advisor used to play. The channel didn't lose its economics — the economics moved to whoever the model trusts.

Two referral channels the models can't eat

So the practical question for anyone building or joining a partner program: which referral assets appreciate under this regime? Two, and they sit at opposite ends of the stack.

• Machine-readable reputation. If assistants write the shortlist, being reliably in the answer is the new page-one ranking. That's an entity problem, not a keyword problem — structured data, consistent citations, verifiable reviews, third-party corroboration the model can retrieve and trust. Answer-engine optimization is to 2026 what SEO was to 2010, except the loser doesn't drop to page two; the loser was never mentioned.
• Human vouching. The model can synthesize every review on the internet, but it cannot look a Houston practice owner in the eye and say "I use this, bill me if it fails you." At the exact moment mass-produced recommendation content gets absorbed into answer engines, the recommendation that can't be scraped — a named human staking their reputation inside a real relationship — becomes the scarce input. Scarcity is pricing power for partners who have it.

And note what the 1%-cited-click statistic does to attribution. Last-click tracking — the cookie, the UTM, the affiliate link — undercounts influence more every quarter, because influence increasingly happens where no link is clicked. The compensation models that survive are the ones that never depended on the click: named-partner registration, personal codes, and relationship-attributed recurring splits. Which, not coincidentally, is how gated, vouching-based partner programs already work. The plumbing the affiliate industry considered legacy is turning out to be the durable part.

The honest counterargument

This brief's standing rule is to print the strongest case against its own thesis, so here it is — and it's substantial.

First, the affiliate channel is not dying by the only measure that settles arguments: money. US affiliate spend is running $13.81 billion in 2026, up 11.3% from $12.42 billion in 2025. Advertisers do not grow a channel double digits while it collapses. Second, the AI-referral growth rates are huge percentages on a still-small base — AI-referred visits remain a single-digit share of total retail traffic, and search plus direct still dwarf them in absolute volume. Third, the zero-click numbers are contested: Google disputes the third-party methodologies, and click-through varies enormously by query type — transactional and local queries still route clicks far more than informational ones. Fourth, in considered B2B purchases the model writes the shortlist, but humans still close — 94% using AI is not 94% deciding by AI.

The synthesis that survives all four objections: this is a re-weighting, not an extinction. Commodity, click-dependent, content-arbitrage affiliate motions are structurally impaired. Relationship-attributed, trust-based partner motions are structurally advantaged. The budget line isn't shrinking — it's migrating between those two buckets, and the 11.3% growth number is the migration being funded.

Where this plugs in

WCS operates on both sides of the map above, deliberately. The machine-readable layer is the service business — answer-engine optimization, Google Business Profile X-Ray, schema and entity work that makes Houston medical, dental, veterinary, and construction operators the answer the assistants retrieve. The human-vouching layer is the North Star Affiliate Network: a straight 50% lifetime recurring split across every asset in the ForgedOps.Ai suite, attributed by named relationship — not by cookie — with a 20% override on Skool community referrals, 10% on partners you bring into the network, and no upfront cost.

Entry runs through a short Alignment Interview rather than an open signup form — the partner-quality gate that Issues 1 and 3 argued is the condition making uncapped lifetime economics survivable. In a market where the click is dying and the vouch is appreciating, a roster of partners who actually vouch is the whole asset.

Take the Alignment Interview — North Star Affiliate Network

Buyer-behavior figures are drawn from Forrester 2026 B2B buyer research and G2 buyer-behavior data; click-through and zero-click figures from Pew Research Center and Similarweb tracking of the post-AI-Overviews period; AI-referred traffic and conversion figures from Adobe Analytics (Q1 2026 and 2025 holiday-season reporting); US affiliate spend from published 2026 industry estimates. All are third-party measurements current as of early August 2026, several are contested (see the counterargument section), and projections are not guarantees. North Star Affiliate Network commission figures are drawn from published tiers as of August 2026 and are illustrative — actual results depend on the assets promoted and the account mix.

Money that can only be spent one way

Enterprises are currently sitting on roughly $470 billion in committed cloud spend across AWS, Microsoft, and Google — and by some counts the three hyperscalers have booked north of $900 billion in total customer cloud commitments. Those commitments live inside AWS's Enterprise Discount Program, Microsoft's Azure Consumption Commitment, and Google's Committed Use Discounts. They are contractual. They have expiration dates. And a CFO who under-consumes against one is a CFO who negotiated badly.

That is the actual engine behind the marketplace boom, and it is worth being precise about it, because it gets described backwards constantly. Cloud marketplaces did not win because the buying experience is delightful. They won because they are the legally sanctioned mechanism by which a customer can point already-committed cloud dollars at third-party software. A $2M drawdown against a MACC is not a purchase decision in the ordinary sense. It's an allocation decision on money that is already spent.

The volume follows from there. Cloud marketplace GMV now exceeds an estimated $45 billion annually, roughly tripling in four years. Hyperscaler marketplace transactions are projected to run from about $16 billion in 2023 to $85 billion by 2028, with Omdia projecting $163 billion by 2030 at a 29.1% CAGR — of which agentic-AI transactions alone are forecast to account for $24.4 billion. As of mid-2025, Google closed the last structural gap by moving to 100% commit drawdown on qualifying Marketplace Channel Private Offers, reaching parity with AWS EDP and Azure MACC for channel-driven enterprise procurement. All three doors now open the same way.

The take-rate war nobody predicted

Here's the part that should reframe how you think about platform economics generally. Everyone spent a decade assuming platform intermediaries converge toward the app-store 30%. Cloud marketplaces went the other direction — hard.

AWS Marketplace private-offer listing fees are tiered down with deal size: roughly 3% on transactions under $1 million, falling to 1.5% above $10 million, with professional-services private offers cut to 0.5% as of June 2026. Google Cloud Marketplace has introduced a variable revenue share that can drop as low as 1.5%, scaled by deal size and by whether the transaction flows through a channel partner. That is not the behavior of a rent-extracting gatekeeper. It's the behavior of three companies competing for the privilege of hosting a transaction whose real value to them is the cloud consumption attached to it — not the software margin.

The strategic read: the hyperscalers are not monetizing the marketplace. They are monetizing the commitment. The marketplace exists to make the commitment stickier and easier to consume, which means listing fees are a loss leader and will likely keep compressing. If you sell six-figure-plus enterprise software and you are still treating a 3% marketplace fee as the reason not to list, you are optimizing the wrong variable by roughly an order of magnitude.

The channel isn't being disintermediated. It's being re-intermediated.

The obvious prediction when marketplaces took off was that they'd cut resellers out — direct vendor-to-buyer, no middle layer. The data says the opposite happened. Third-party gross transaction value resold through channel partners on Google Cloud Marketplace grew 170% from 2023 to 2024, and Omdia projects nearly 60% of all hyperscaler marketplace transactions will flow through the channel by 2030.

That makes sense once you stop thinking of the marketplace as a store. Drawdown mechanics, private-offer construction, multi-year term structuring, and the procurement-and-legal choreography around a MACC are genuinely specialized work. The marketplace removed the billing friction and, in doing so, made the remaining friction — deal engineering — more valuable, not less. The partner who used to add value by carrying paper now adds value by knowing which commitment vehicle a given buyer has, how much runway is left on it, and how to structure an offer that qualifies for full drawdown.

Which is a specific, teachable skill. It is also a moat that a 2021-vintage "we resell logos" partner program does not confer on anybody.

The gravity well: what this does to everyone under enterprise scale

Now the part that matters most to the readers of this brief, most of whom are not selling $10M contracts to Fortune 500 procurement.

Every advantage described above is conditional on one thing: the buyer has a committed cloud spend to draw down. Take that away and the entire model inverts. A dental group in Houston, a mid-market GC, a 40-person specialty clinic, a regional veterinary network — none of them have a MACC. None of them have an EDP. There is no committed pool for a marketplace to unlock, which means listing on one buys you a transaction rail you don't need and a discovery surface your buyer will never visit.

So for SMB- and mid-market-serving vendors, the partner economics run the opposite direction, and it's worth naming the inversion explicitly:

• Enterprise: the scarce resource is procurement access. The channel's job is to unlock money that already exists. Fees compress toward zero because the platform monetizes elsewhere.
• SMB / mid-market: the scarce resource is trust. There is no pre-committed budget to unlock — someone has to be persuaded to create new spend. The channel's job is vouching, and vouching cannot be discounted to 1.5%.

That's why the compensation structures look nothing alike, and why importing enterprise-channel thinking into an SMB motion produces programs that don't work. Recurring commission isn't the SMB version of a marketplace listing fee. It's the price of a fundamentally different and more expensive good.

The honest caveat on lifetime commissions

This brief has run two issues arguing for recurring, lifetime-style commission structures. Intellectual honesty requires printing the counterargument, because it's a real one.

The 2026 benchmarks: median B2B SaaS affiliate commission sits at 20% (15% for B2C), the standard band is 20–30% recurring, and most programs pay recurring for a fixed 6- or 12-month term rather than for life. A healthy affiliate CAC is generally cited at 20–40% of LTV — enough to be competitive, little enough that the LTV:CAC ratio still lands near 3:1 or better. And the specific warning worth taking seriously: uncapped lifetime commission is dangerous on sticky, low-churn products. A customer who stays seven years can earn an affiliate more in cumulative commission than that customer generated in margin, if the rate was set on first-year intuition rather than on a full-lifetime model.

So a 50%-lifetime split is not a generosity decision. It is a deliberate CAC decision, and it only survives contact with reality under three conditions:

• Gross margin has to carry it. Software-and-services delivery with thin variable cost can pay half of revenue away and still fund operations. A structure with heavy per-account delivery cost cannot, and shouldn't pretend otherwise.
• The split has to replace paid acquisition, not sit on top of it. If you're paying 50% to partners and running a paid-acquisition budget against the same segment, you've stacked two CACs on one LTV — the exact error Issue No. 2 flagged on the reseller side, just pointed the other way.
• Partner quality has to be gated at intake. Roughly 20–30% of recruited partners in a typical program ever produce a deal, and roughly 86% of SaaS affiliate programs now manually review every partner. Uncapped economics on an unfiltered roster is how a program becomes an accounts-payable problem instead of a growth channel.

Meet those three and lifetime recurring is simply accurate pricing for the value a partner delivers. Miss any one of them and the benchmark critics are right.

Where this plugs in

WCS sells into exactly the segment described above — Houston medical, dental, veterinary, and commercial construction operators who have no cloud commitment to draw down and every reason to want a vouched introduction instead of a procurement portal. That is why the North Star Affiliate Network is built the way it is: a straight 50% lifetime recurring split across every asset in the ForgedOps.Ai suite, a 20% override on Skool community referrals and 10% on partners you bring into the network, no upfront cost, no fee stack underneath the number, and no delivery obligation on the partner.

And per the third condition above, entry runs through a short Alignment Interview rather than an open signup form. That gate isn't gatekeeping for its own sake — it's the mechanism that makes uncapped lifetime economics survivable for both sides.

Take the Alignment Interview — North Star Affiliate Network

Committed-spend totals, marketplace GMV and growth projections, listing-fee and revenue-share schedules, channel-resold transaction growth, and affiliate commission / CAC benchmarks are drawn from public cloud-marketplace and SaaS-affiliate industry research and vendor documentation current as of July 2026; projections are third-party forecasts, not guarantees. North Star Affiliate Network commission figures are drawn from published tiers as of July 2026 and are illustrative — actual results depend on the assets promoted and the account mix.

The stack that compounds against you, not for you

White-label SaaS reselling is one of the two dominant partner monetization paths in 2026, and done right it is genuinely lucrative — agencies are routinely building $200K+ ARR service lines by licensing a platform, rebranding it, and folding it into an existing retainer. Straightforward reseller arrangements on $100–$700/month products typically run around 40% margin. Agencies willing to own sales, onboarding, and support themselves can push that to 65–85% gross margin.

The number that gets skipped in the pitch is what sits underneath that margin. A white-label stack routinely layers a base platform fee, per-product add-on fees, per-seat charges, and onboarding costs before the reseller's own markup ever gets applied. Each of those layers compresses the reseller's effective take — and critically, each layer compounds against margin without necessarily compounding the value the end client actually perceives. A client paying for "one platform" has no visibility into how many separate toll booths their monthly invoice is quietly funding on the vendor side. The reseller does the negotiating, absorbs the churn risk, and is the one who finds out — usually at renewal — that the stack grew faster than the retainer did.

The enablement model that keeps the stack honest

The white-label programs holding margin steady in 2026 aren't the ones with the flattest fee schedule — they're the ones with the clearest tiering. The pattern worth copying has three levels:

• Entry-level resellers sell standard, pre-packaged tiers with fixed margin and no negotiation overhead — fast to onboard, capped upside.
• Certified implementation partners take on onboarding and support directly, which is what unlocks the 65–85% margin band — the vendor is paying for labor the partner is now doing themselves.
• Strategic OEM partners embed the platform inside a broader solution entirely, with negotiated infrastructure pricing and joint roadmap input.

Each tier has a defined technical access level, a defined margin structure, and a defined service obligation — which is the actual point. Margin stacking hurts most when the fee schedule and the service obligation aren't clearly matched at every tier. A reseller who finds out mid-contract that their tier doesn't cover the support load they've already committed to is a reseller who took on liability without being told the price of it.

Why 2026 vendors are running both models at once

White-label reselling and recurring-commission affiliate partnership aren't competitors for the same partner — they're different risk/control trade-offs for two different kinds of partners, and this year's shift is vendors packaging both under one program instead of forcing a choice. "Hybrid Monetization" is the term showing up across 2026 SaaS strategy coverage: white-labeling turns the product into a high-margin, distributed asset for partners who want to own the client relationship and the invoice, while a parallel recurring-commission track serves partners who'd rather plug into an existing relationship than build their own delivery org. Same platform, same underlying economics, two very different partner profiles served without either one subsidizing the other.

That bifurcation is becoming necessary rather than optional because the partner ecosystem itself has gotten more specialized. The generalist partner — happy to resell anything with a decent margin — is losing share in 2026 to partners who understand a specific buyer's environment, workflows, and compliance exposure well enough to sell consultatively. A single flat program can't serve a specialist and a generalist with the same fee schedule without shortchanging one of them.

The signal underneath the stack

Three data points worth sitting with if you're deciding where to place your own partner bets this quarter:

• PRM (partner relationship management) platforms have crossed the adoption chasm in the $25M+ ARR cohort — 62% adoption, up from 39% in 2023 — meaning the vendors you're competing against for partner attention are increasingly running structured, tiered programs rather than a spreadsheet and a handshake.
• Ecosystem-led-growth motions are now reported to win roughly 3.6× more often than cold-direct deals, which is the actual reason vendor partner budgets keep moving away from paid acquisition and toward partner enablement.
• The modern partner ecosystem is no longer just resellers — it now routinely includes technology alliances, systems integrators, marketplace partners, affiliates, and developer communities, each requiring a different deal structure. A program built around one fee schedule for all of them is a program built for 2021, not 2026.

Where this plugs in

If margin stacking is the risk you're trying to underwrite around, it's worth knowing there's a structure that sidesteps it entirely rather than managing it: a straight 50%-lifetime recurring split, like the one running through WCS's own North Star Affiliate Network at DRYLAND.AI, across every asset in the ForgedOps.Ai suite. One number, no platform fee stacked underneath it, no onboarding cost to price in, no inventory or delivery obligation to carry — the partner earns on the relationship, not on the invoice. It's not a replacement for white-label reselling; it's the other side of the trade-off covered above, for partners who'd rather not carry the delivery org. Entry runs through a short Alignment Interview, not an open signup form — the data on partner activation above is exactly why.

Take the Alignment Interview — DRYLAND.AI

Figures on reseller margin bands, tiered enablement structures, PRM adoption, ecosystem-led-growth win rates, and partner ecosystem composition are drawn from public partner-ecosystem, channel-marketing, and white-label SaaS industry research current as of mid-2026. North Star Affiliate Network commission figures referenced above are illustrative and drawn from published tiers as of July 2026; actual results depend on the assets promoted and account mix.

The partner channel stopped being a side hustle

Ecosystem-Led Growth is no longer a nice-to-have overlay on top of direct sales — for a growing share of B2B software companies, it's becoming the primary go-to-market motion, and deals sourced through partner ecosystems reportedly close at multiples of the rate of cold-direct outbound. The scale of the shift shows up in two numbers worth sitting with: by 2026, an estimated 80% of B2B software buyers will use a marketplace to initiate or complete a purchase, up from just 35% in 2021, and healthy partner-sourced revenue now runs anywhere from roughly 24% of total revenue in horizontal SaaS to as high as 47-58% in cybersecurity and services-led businesses.

That's not a rounding error. It's a structural admission that the highest-trust, lowest-CAC customer acquisition channel most companies have isn't their ad account — it's the network of people who already vouch for them. The question isn't whether to build a partner channel anymore. It's whether the compensation structure you're offering actually gets partners to sell, or just gets them to sign up.

One-time bounty vs. recurring: the math that actually compounds

Most affiliate programs still default to a one-time bounty — a flat fee the moment a referral converts. It's simple to administer and it's the wrong instrument for software, because it pays the partner for the sale and nothing for the relationship. The standard for SaaS affiliate programs that have figured this out is a recurring commission in the 20-30% range, typically running 12-24 months per customer — long enough to reward partners for bringing in customers who stick, short enough that unit economics stay sane for the vendor. A handful of programs go further: Systeme's affiliate program pays a 60% lifetime commission, Sanebox pays 30% lifetime, and the common thread across every program that pays lifetime rather than a fixed window is retention. If your product keeps customers for three-plus years, a lifetime split isn't generosity — it's just accurate accounting for what the partner is actually worth to you.

Run the numbers on a $150/month product with a 24-month average customer lifetime. A $200 one-time bounty nets the affiliate $200, period. A 25% recurring commission over 24 months nets $900. A 50%-lifetime commission on the same account — assuming the customer stays past month 24, which lifetime deals are explicitly betting on — keeps paying for as long as the relationship exists. The vendor isn't giving away margin; they're renting customer acquisition cost against a revenue stream that would otherwise cost far more to generate through paid channels, and paying it out of gross margin they wouldn't have captured any other way.

Anatomy of a 50%-per-life deal, with real numbers

Here's what that structure looks like in practice, using the published tiers from our own North Star Affiliate Network (coming soon) at DRYLAND.AI, which pays 50% lifetime recurring on every asset across the ForgedOps.Ai suite:

The structural point matters more than any single number: because the commission is a straight 50/50 lifetime split rather than a stacked reseller markup, there's no margin-compression risk as the relationship ages. The partner's percentage doesn't decay in year two. That's the design difference between a recurring-commission affiliate model and a white-label reseller model — and it's worth understanding both before picking one.

White-label reselling: where the margin quietly disappears

White-label SaaS reselling is the other dominant monetization path in 2026, and it can be extremely lucrative — agencies are routinely building $200K+ ARR service lines by licensing a platform, rebranding it, and bundling it into an existing retainer. Reported reseller margins run wide: roughly 40% is typical for straightforward reseller arrangements on $100-700/month products, while true agency/reseller models that take on sales, onboarding, and support themselves can command 65-85% gross margins.

The risk resellers underweight is margin stacking. A white-label stack often layers a platform fee, per-product fees, per-seat charges, and onboarding costs on top of each other before the reseller ever adds their own markup — and each layer compounds against the reseller's effective margin without necessarily compounding the value the end customer perceives. A recurring-commission affiliate structure sidesteps this entirely: there's one number (the split), it doesn't stack, and it doesn't require the partner to hold inventory, manage billing, or absorb churn risk on infrastructure they don't own. White-label reselling and recurring-commission partnership aren't competitors — they're different risk/control trade-offs, and the right one depends on whether a partner wants to own the client relationship (reseller) or plug into someone else's (affiliate).

Recruitment isn't the bottleneck. Activation is.

The uncomfortable stat in most partner programs: only an estimated 20-30% of recruited partners ever actually produce a deal. Companies chase partner count because it's an easy number to report, when the number that actually moves revenue is the share of partners who are activated — onboarded, equipped, and actually selling. AI-assisted PRM tooling is starting to close that gap with churn prediction and next-best-action nudges, but the more durable fix is upstream: qualify partners before you recruit them, rather than recruiting broadly and hoping activation sorts itself out.

That's the actual argument for gating a program behind something like an alignment interview instead of an open signup form — it trades a larger top-of-funnel for a materially higher activation rate, which is the only number in this whole exercise that compounds.

Where this plugs in

If the math above is interesting rather than abstract: the North Star Affiliate Network is WCS's own 50%-lifetime-recurring partner program, covering every asset across the ForgedOps.Ai suite, with a 20% override on Skool community referrals and a 10% override on partners you bring into the network. There's no upfront cost — entry runs through a short alignment interview designed to do exactly what the activation data above argues for: place partners where they're actually positioned to sell, not just sign up.

Take the Alignment Interview — DRYLAND.AI (coming soon)

Earnings examples above are illustrative and drawn from published North Star Affiliate Network tiers as of July 2026; actual results depend on the assets promoted and account mix. Third-party figures (activation rates, marketplace adoption, partner-sourced revenue benchmarks, affiliate commission structures) are drawn from public partner-ecosystem and SaaS-affiliate industry research current as of mid-2026.