Email Earns from Outcomes
Who carries the risk
The first income line changes what an email can do. The second changes who is responsible when it does not work.
That is the bigger change and the more uncomfortable one, because every incumbent commercial model in marketing is built to avoid it. The ESP sells the send and is paid whether it converts. The martech platform sells the seat and is paid whether it is used. The agency sells the retainer and is paid whether the campaign lands. The ad platform sells the impression and is paid whether the customer was going to buy anyway. In every case the brand carries the entire outcome risk of a system it did not build and cannot fully see.
Progency inverts that. It is not a fifth product and not a services wrapper. It is an accountable operating layer that sits after the CRM and before the auction: it takes responsibility for a defined customer state, runs the interventions on the surface, measures against a control, and earns only on the lift it can prove.
Three mandates, three counterfactuals
Progency runs three mandates, one per attention state. Each is a bet against a different counterfactual — a different answer to the question what would this customer have done anyway?
Recover is for customers gone dark. The counterfactual is adtech: the money the brand would otherwise spend buying that person back through a platform which already holds their email address. This is the wedge, because it attacks the most expensive leak in the P&L and it is the one almost nobody else is fixing. Inside the email the sequence is connection first, then recovered attention, and only then conversion. A dark customer asked to buy on first contact simply stays dark.
Protect is for valuable customers whose attention is cooling. The counterfactual is drift: left alone they become lost, and a lost customer is later reacquired at several times the cost of having kept them. The email here is Digest and Relate, not Sell. The job is to arrest the slide while arresting it is still cheap.
Grow is for the attentive. The counterfactual is a slower next purchase and margin left on the table. The email is Sell and Notify made live — the right next thing, completed in the inbox, with the customer suppressed from paid retargeting because they are already reachable at no cost.
These are not marketing segments. They are attention states, and the point of measuring them is that a customer moves between them continuously and the correct intervention changes when they do.
One floor separation has to be held here, because it is exactly the kind of drift that quietly corrupts the vocabulary. The three mandates are not the four NeoMarketing zones — Retain, Finish, Recover, Acquire. Zones answer where the work sits in the operating architecture. Mandates answer what Progency is trying to change in the customer’s state. Finish, for example, is a zone: a KYC, a quote, a renewal or a purchase left incomplete. That unfinished job might sit inside Recover or inside Grow depending on whether the customer is dark or attentive. Recover appears on both floors and means different things on each. The vocabulary only works while the floors stay apart.
Figure 5 — Three mandates, one per attention state, each priced against a different counterfactual.
Beta, Alpha, Carry
The pricing follows from the mandates and is simple enough to survive a CFO meeting.
Beta is what would have happened anyway — the baseline, agreed in advance, measured rather than asserted. Alpha is the verified lift above that baseline. Carry is Progency’s share of the Alpha, and only the Alpha.
No lift, no fee. The old service-provider invoice was tied to usage. The old agency retainer was tied to activity. The adtech bill is tied to rented reach. This payout is tied to profit improvement and to nothing else.
The economic unit is worth stating precisely, because it is where outcome pricing usually cheats: the unit is not an email, an open, a click or an attributed conversion. It is the incremental completed outcome above the agreed baseline.
What Run leaves behind
There is a second thing Progency produces, and over time it is the more valuable one.
Every intervention writes a Decision Trace: the customer context, the eligible pool, the treatment chosen, the channel, the job the message was doing, the holdout status, the expected outcome, the cost, the actual outcome and the resulting state. Individually each is a row. Accumulated, they are something a competitor cannot buy or copy — a growing memory of decisions tied to their measured consequences.
This matters because the obvious moat is the wrong one. The intelligence that decides what to try — the models, the archetype priors, the agent confidence — will commoditise, and quickly. Everyone will have capable models. What does not commoditise is the corpus that says which decisions, on which customers, in which states, moved the outcome against a control. Progency earns revenue today and, in the same motion, builds the memory that makes tomorrow’s outcomes cheaper and more reliable to produce. Run is the point where the email business stops being a set of messages and starts being an underwritten system for moving customer states — one that gets better at it every time it is paid.
The holdout is not a detail. It is the apparatus.
Everything above is a claim about money, which means it stands or falls on the measurement. Three disciplines, none of them negotiable.
The holdout is concurrent, not historical. A comparison against last quarter measures the season, the pricing, the competitor’s campaign and the weather. A comparison against a randomly withheld group running at the same time measures the intervention. Historical baselines are how agencies have been claiming lift for thirty years, and they are why nobody believes the claims.
The holdout is enforced in the system, not in the contract. It is a hard gate in the automation layer: if the control group is not held, the campaign does not run. A discipline that depends on an operator remembering to apply it under quarter-end pressure is not a discipline.
Simulated judgement and measured Alpha never share a currency. The prediction layer decides what to send and to whom; it gets no vote on what is paid. The holdout decides what is paid; it has no opinion on what to send. Keeping the two separate is not conservatism. It is the difference between a measurement system and a marketing claim. A system must never promote itself using its own predictions as evidence.
Why this rung has to come before the next one
There is a structural argument for the ordering that is easy to miss, and it is the most important sentence in this essay.
The fourth income line — media — is the one that can destroy the asset. Ad load is a dial, and every dial in marketing is eventually turned up by somebody with a quarterly number to hit. The usual protections against this are policy documents and good intentions, and they fail.
An operator paid on carry cannot over-monetise the surface, because the carry depends on the attention surviving. If ad load rises to the point where opens fall, Recover misses, Protect misses, Grow misses, and Progency’s own income falls with them. The commercial model is the governor. It is not a promise not to spoil the surface; it is an arrangement in which spoiling the surface is immediately and directly expensive to the party holding the dial.
That is why Run comes before Network. The third income line is what makes the fourth one safe.
The honest constraint
One thing this model does not solve, named rather than buried: the throttle on outcome pricing is working capital, not demand.
An operator paid only on verified lift funds the interventions before it is paid for them, and the measurement window is counted in weeks or months. Appetite for a model where the brand pays only for improvement is not the scarce input. The scarce input is a balance sheet that can carry the gap between doing the work and proving it. Any plan that assumes otherwise will hit the wall at precisely the point where the model starts working.
Key points: (a) Capability revenue is paid for what an email can do. Outcome revenue is paid for what it caused. (b) Every incumbent model in marketing is arranged so the vendor is paid whether or not it worked. Correcting that is the whole of the second income line.
