You are walking leadership through the quarter when the attribution report stops making sense. The largest self-serve deal of the period shows no campaign touch, no landing page visit, and no ad impression: the CRM lists the source as direct, and the account moved from first documentation read to paid plan in nine minutes. The same afternoon, a customer of three years fails to renew. Their procurement assistant re-ran the vendor comparison at the renewal date, read a competitor's machine-readable price list, and moved the workload without a conversation. Your marketing spend bought the quarter's traffic, but the revenue moved through a channel your dashboards cannot see, and loyalty never entered either decision.
The earlier chapters of this part covered the mechanics of agent traffic; this chapter covers the business math. A funnel is a model of human attention, and when the buyer is an agent, that attention never arrives, so the disciplines built on it, attribution, pricing, and brand, each have to be rebuilt.
The funnel converts attention, and agents bring none
Your funnel is a story about eyeballs: a person sees an ad, lands on a page built to persuade, reads a pricing table in the order you arranged it, and meets the upsell at the moment of highest intent. Agent traffic breaks every step. The read wave answers the person's question straight from your docs, so the visit never happens, and the shopper wave compares structured facts and completes the purchase, so nothing watched the ad, weighed the landing page, or paused at the upsell. Your reports already show the split, automated reads climbing steeply while referred human visits stay flat, a ratio that has moved by orders of magnitude in a couple of years, with current figures in the dated Interop Ledger.
That split is margin-relevant, not just interesting: you pay to produce and serve everything the read wave consumes, your acquisition costs assume human conversion rates, and the upsell page, the in-app prompt, and the renewal email sit on paths this buyer never renders.
Attribution, pricing, and brand were each built on a human watching the funnel, and each one has to be rebuilt on records a machine can read.
Attribution: assistant platforms are channels with owners
You cannot manage a channel your reports cannot see. Some of the identification is mechanical today: crawlers carry user-agent strings, and assistant referrals arrive with headers your analytics can already segment. The hard case is the purchase that completes inside the assistant, where no visit ever lands, and signed agents close that gap: verification schemes now shipping from infrastructure providers make "which assistant sent this buyer" an answerable question rather than a guess.
- Segment what already identifies itself. Split known agent user-agents and assistant referrals out of "direct" now, so the channel has its own line in the report before it has a strategy.
- Ask for identification where agents transact. Favor signed agents at signup and checkout, and record the sending platform on every agent-touched deal the way you record the campaign on a human one.
- Expect affiliate economics. In the commerce integrations shipping now, the assistant platform takes a fee on completed purchases, which makes it a paid channel with a negotiable rate rather than free traffic.
This is the app-store pattern repeating: mobile taught product teams to manage ranking, placement, and take rates inside a channel one company owned, and assistant platforms are assembling the same position over purchases, so give each one an owner and a line in your channel math.
Pricing: this buyer never skips the comparison
A meaningful slice of most subscription revenue is inertia revenue: seats a team of ten bought that a team of six still pays for, renewals that go through because nobody re-opened the evaluation, teaser prices that step up quietly after the first year. An agent handed a company's vendor spend re-runs the comparison at every cycle, trims a seat the day it goes idle, and treats the renewal-date price, not the teaser, as the price.
Inertia revenue erodes under agent procurement: unused seats, forgotten renewals, and teaser-then-raise all price the customer's inattention, and this buyer re-runs the evaluation at every cycle.
Machine-readable pricing changes the offer itself. A person reads a pricing page as positioning and skims the footnotes, while an agent is handed the whole document and prices on all of it: overage rates, auto-renewal terms, and cancellation windows carry the same weight as the headline number. To this buyer the fine print is part of the offer, and it will be compared.
Usage-based and per-action pricing hold up best here, because they bill in the unit agents consume, metered actions rather than occupied seats, and they leave the re-comparison nothing idle to cut. Pricing: charge in a currency your costs track works through the margin math of getting there, and the agent channel adds its own serving costs, the tool calls and retries itemized in The bill of materials: cost the task, not the call, so price the channel with those costs in front of you.
Brand: the record replaces the impression
Brand spend has bought impressions and recall: a logo seen often enough that the buyer reaches for it without deliberating. An agent renders no logo and starts every comparison cold, so what stands in for brand is the trail your product leaves in records.
- A reliability record. Uptime, error rates, and latency accumulate in the operator platform's logs, and a vendor that fails tasks gets deselected without a complaint ever reaching you.
- Docs that match behavior. Documentation is the only description of your product an agent acts on, so a doc that drifts from real behavior registers as a product that does not work.
- Clean disputes. Refunds, cancellations, and error handling become structured outcomes on record with the platform, and friction there follows you into the next comparison.
- Standing in registries. Presence and accurate metadata in the catalogs agents check decide whether you enter the comparison at all.
Record-brand replaces impression-brand: the agent carries no loyalty, but its operator platform carries defaults, and defaults are the new shelf space.
An individual agent starts cold, but the platform it runs on accumulates defaults: preferred merchant lists, cached tool choices, the vendors its commerce integration offers first. A default is the new eye-level shelf placement, because it decides which comparisons you are invited into.
The read wave: meter it or market with it
For a content business, the read wave is the product leaving the building, and a market now exists for charging at the door: pay-per-crawl schemes meter crawler access, and licensing deals put negotiated prices on archives. If your words are your revenue, study those instruments.
For a product business the calculus reverses: the crawler in your docs is the top of the new funnel, because what the read wave ingests is what the shopper wave will compare and the operator wave will act on, so blocking it saves serving cost but takes you out of those comparisons. Keep the read wave legible instead, with accurate docs, machine-readable pricing, and structured capability descriptions, maintained with the discipline you give the product.
Try it now
This drill takes twenty to thirty minutes and tells you whether your pricing survives this buyer.
Get your specimen. Pick one product with public pricing, your own if you have one, otherwise a tool your team pays for. Pull the current price list and a recent month of real usage: seats occupied, actions taken, whatever the meter shows.
Give the buyer its rules. The agent re-runs the full comparison at every renewal, never leaves a seat or an allowance idle, and reads every published term at face value. Compute a year of revenue from that account under the current scheme, cutting whatever those rules would cut at each renewal.
Re-price it usage-based. Sketch the nearest usage-based or per-action equivalent, a price on the metered thing the product actually does, and compute the same year against the same rules.
Write the verdict. In one paragraph, state which scheme survives, how much of today's revenue was inertia, and the one change to price, packaging, or terms that has to land before the shopper wave reaches your category.
Scale it down: one plan tier and one account is enough, since the point is the gap between the two numbers, not their precision.
Chapter Summary
- Your funnel assumes a human watching: crawlers answer the question so the visit never happens, and shoppers compare structured facts and buy, so nothing sees the ad, the landing page, or the upsell.
- Reads climbing while referred human visits stay flat is margin math, not trivia, because you pay for content and acquisition this buyer bypasses.
- Attribution starts with identification: segment agent user-agents and assistant referrals out of "direct" now, and favor signed agents where they transact.
- Treat assistant platforms as channels with owners, terms, and take rates, the way app stores became channels.
- Inertia revenue (idle seats, forgotten renewals, teaser-then-raise) erodes when the buyer re-compares at every cycle.
- An agent prices on every published term, so the fine print is part of the offer; usage-based and per-action schemes bill in the unit agents consume.
- Record-brand replaces impression-brand: reliability records, docs that match behavior, clean disputes, and registry standing do the work impressions used to.
- The agent carries no loyalty, but its platform carries defaults, and defaults are the new shelf space.
- Content businesses can meter and license the read wave; product businesses should keep it legible, because it is the top of the new funnel.
The next chapter, Write your Agent Access Policy and open the door deliberately, gathers this part's decisions into one document your team can ship.
Sources
- Cloudflare (2025). Pay Per Crawl announcement and Radar reporting on AI crawler crawl-to-referral ratios.
- Cloudflare (2025). Web Bot Auth and signed agents documentation for verifying automated traffic (last verified July 2026).
- OpenAI and Stripe (2025). Agentic Commerce Protocol specification and merchant documentation (last verified July 2026).
- Reddit (2024). Content licensing agreements with Google and OpenAI, company announcements and securities filings.
- News Corp and OpenAI (2024). Multi-year content licensing agreement announcement.