Rajesh Jain

The Habit Engine

Why appointment mechanics are not enough

The best-known email-adjacent habit products are appointment mechanics: one thing, once a day, at a time you choose. They work, and they cap. A once-daily appointment produces one return per day — a respectable retention curve and a thin attention supply.

The media income line needs more than one open a day from an engaged base. So the design question is not how do we earn a daily return?, which is solved and insufficient, but how do we clear the appointment ceiling without reaching for mechanics we cannot use?

Four forces, working together.

Heartbeat — something is happening right now

Live events with a clock of their own: a match in progress, a market open, a forecast resolving, a challenge expiring. The heartbeat supplies the intraday cadence that no brand calendar contains, and it supplies it free, because the world is producing it anyway.

For India at mass scale the natural heartbeat is cricket. That is not a preference; it is an observation about where national attention already synchronises — for hours at a time, on a predictable calendar, across every demographic this venture would want to reach. A world with a cricket spine has a reason to be checked at eleven, at two and at six, and the checking is the entire point.

Around that spine sit the other clocked things: markets, weather, results, deadlines, resolutions.

Progression — I am further along than I was

Accumulated status, streaks, collections, asymmetric advancement. The oldest mechanic in the book and still the most reliable, because it converts past participation into a reason to participate again.

One design constraint, load-bearing rather than cosmetic: progression is earned, never bought. It records what a person did. It does not store value, it cannot be purchased, and it cannot be transferred. That constraint is partly regulatory and partly good sense — progression that can be bought stops being evidence of anything and becomes a leaderboard of wallets.

Social obligation — someone is waiting on me

Circles: small groups with shared challenges and live deadlines, where a person’s absence is noticed by named people who know them.

Two reasons this is the strongest of the four forces.

It is the most durable return mechanic that exists. Individual motivation decays; obligation to specific people does not, because the cost of not showing up is social and immediate rather than abstract and deferred.

And it solves cold start, which is otherwise the hardest problem in any consumer launch. Circles are seeded into the WhatsApp groups people are already in — the office group, the college group, the family group, the building group. The social graph does not have to be built. It has to be borrowed, with an invitation that makes sense inside a conversation already happening.

Standing — my record travels with me

The Predictor Score: a persistent, compounding record of how well-calibrated a person’s judgements turn out to be, built on Brier mechanics. It rises with accuracy over time and cannot be shortcut.

The distinction that keeps this clean is worth stating in its frozen form. Mu is the token — what flows. Standing is the reputation — what compounds. The Circle is the room; Standing is the passport. A score is a record of skill, not a holding of value. It cannot be bought, sold, transferred or cashed out, and it is more interesting for exactly that reason: it is the only thing in the system that money cannot acquire.

Figure 3 — The four forces of the habit engine. Two supply the reason to come back; two supply the reason to stay.

The regulatory position, stated directly

India’s Online Gaming Act 2025 rules out a set of mechanics an earlier version of this design would have used: purchasable currency, transferable currency, pooled stakes, and the entire vocabulary of wagering. That is a hard constraint, and it applies to the language as much as to the mechanics. A product that avoids staking but describes itself in betting terms has complied with nothing.

It is worth saying plainly that the constraint improved the design.

The mechanics it removed were loss-aversion mechanics. Loss aversion produces sharp early engagement, a particular kind of user, and an adversarial relationship between the product and the people who use it most. What replaced it — obligation to a named group, resolution against real-world events, progression that cannot be bought — produces slower early numbers and a considerably more durable habit. It is also, not incidentally, the version a brand advertiser is willing to place an ActionAd inside.

Where Magnets sit

Magnets — the interactive engagement units — do not disappear in this model. They change position. In the campaign model a Magnet is a diversion attached to a send. In the world model it is a component inside the world: the puzzle is how you advance, the poll is how your Circle decides, the micro-game is how a challenge is contested.

A game in email is a campaign. A game made of email is a habit.

The open question

Honesty about what remains unresolved. Whether these four forces together clear the appointment ceiling — whether obligation plus live resolution plus progression plus standing produces multiple meaningful returns a day, sustained across weeks — is not known. It is the central empirical bet of the consumer motion, and it is precisely what the first proof gate exists to test.

Loss aversion buys you a quarter. Obligation buys you a year.

One Persistent World, Viewed Through Email

The wrong idea, stated clearly so it can be discarded

The obvious move — the one that has been tried repeatedly, produces an excellent demonstration and a flat retention curve — is to put games inside email. A quiz in the Tuesday send. A scratch card in the Friday one. A puzzle in the newsletter.

Each of those is a campaign. Campaigns end. A campaign with a game in it is a campaign with a better open rate for three weeks.

Stacked together they become an arcade — a shelf of isolated experiences, each of which starts and ends. An arcade can produce an appointment: a person may return once a day for a daily word game, the way they return to any single-serving habit. But a shelf does not create a life that continues while the customer is away, and one appointment a day is exactly the ceiling the media income line cannot afford.

The right idea

Do not put a game inside an email. Build a world made of email.

The distinction is architectural rather than decorative. In the first model the email is a container that arrives with content inside it. In the second the email is a viewport — a window onto server-side state that is running continuously, whether or not anybody is looking.

Open it and you see what is happening now: the current score, the open challenge, the deadline that expires in forty minutes, what your Circle did while you were away, where your progression stands, what resolved overnight. None of that was decided when the message was sent. It is composed at the moment it is opened.

This is why the L3-to-L4 inflection — composed at send to composed at open — is the enabling condition for the entire consumer thesis and not merely a technical refinement. A static email is a prediction made at send. A Living Email is a decision made at open. Only the second can be a window onto something live.

The Infinite Email Thread

The delivery form follows from the architecture. Not a campaign, and not a series. One permanent thread from one sender, behaving the way a messaging conversation behaves: new episodes append to it, and each new episode returns the row to the top of the inbox. The thread is the container for the relationship; the episodes are its pulses.

Two properties of this are worth dwelling on, because they are what make the model unusual.

Reopening an old episode shows the current world. Because state is composed at open, episode 38 from last Thursday does not show last Thursday. It shows now. The archive is not an archive — every message in the thread is a live door into the same room. No app inbox and no notification stream has this property, and it means the thread accumulates entry points rather than accumulating dead weight.

The inbox is already the notification layer. No app install. No app store. No push-permission dialogue that most people decline. No home-screen real estate to win. The re-entry mechanism is a row moving to the top of a place the person already checks several times a day. For mass consumer reach in a market like India, that is a structurally cheaper distribution position than any application can occupy.

Figure 2 — The world runs continuously; the thread is the viewport; the pulse score decides what earns an interrupt.

The actual hard problem

If the world is running continuously and the thread can be updated at any time, something has to decide when a change is worth interrupting somebody for. That decision is the product.

Not every state change earns a place at the top of an inbox. A Circle member finishing a challenge might. A leaderboard shifting by one position probably does not. A deadline forty minutes out, for a person who has not yet acted, almost certainly does. The M-Agent layer weighs urgency, novelty, social consequence, the person’s stated preference and how recently they were last interrupted, and then decides which state change has earned the scarce right to interrupt. The job of the agent here is not to generate endless content. It is to say no to almost all of it.

This deserves to be blunt internally. The content is not the moat and the mechanics are not the moat. The pulse-scoring model is the moat, because it is the only part that improves with every open, every mute and every ignored interrupt, and it is the part a competitor cannot copy from the outside.

The condition on all of it

Interactive rendering is not universally supported, and a mass consumer product cannot be designed as though it were.

The fallback build is therefore not a courtesy path. For a large share of any real audience it is the primary experience, and it has to work on its own terms: a static render carrying the world’s state as of the moment of send, with a clear return path into the live view. Fallback-first is non-negotiable. A world that exists only in the interactive render is a world most of the market has never seen.

Key points: (a) The email is not the content. It is the window. (b) A campaign delivers something. A world is somewhere you go back to. Only the second one produces attention worth selling.

The previous essay put three income lines on one email statement and ended on a question it could not answer from inside the B2B model: who manufactures enough recurring attention to make the media line large? No individual brand can. This essay argues that the answer is a persistent consumer world made of email — and that this is not a second business but the supply side of the same one.

**

Why Individual Brands Cannot Create the Habit

The cadence arithmetic

Ask a plain question of any consumer brand: how many useful things does it have to say to a specific customer in a given week?

A bank has a statement, possibly a payment reminder, occasionally a rate change. An airline has a booking, a check-in and a delay. A retailer has an order, a dispatch and a return window. A coffee chain has, if we are honest, a loyalty balance. Add the seasonal peaks and the category-specific moments and the truthful answer for most brands is two or three a week — and for most customers in most weeks, zero.

Two or three moments a week is not a ritual. It is not even a rhythm. A daily inbox habit — the kind that produces the recurring, voluntary attention the media income line depends on — needs something far closer to a heartbeat than to a calendar.

The usual fix makes it worse

Every brand that notices the gap reaches for the same lever, and the lever is attached to the wrong thing.

The response to sparse cadence is increased frequency. The two or three useful things stay exactly as they were; twelve more sends are arranged around them. The list is now contacted daily, which was the goal, and the contacts are mostly worth nothing, which was not.

What follows is well documented and entirely predictable. Opens decline. Clicks decline faster. The engaged base — the ninety-day active portion of the list, which is the only part with commercial value — shrinks. Real Reach falls while list size stays flat, which is why list size is the most misleading number in marketing. Click Retention Rate, which measures the decay directly, moves the wrong way and keeps moving.

Frequency without value destroys attention faster than silence would have cost. A brand that sends nothing for a month is dormant. A brand that sends daily noise for a month has trained its customers against it.

Figure 1 — Three cadences: what a brand has, what raising frequency does to it, and what a live world produces.

Therefore

The conclusion follows without much room for argument. Recurring inbox attention at consumer scale cannot be manufactured brand by brand, because the raw material — worthwhile reasons to appear — does not exist in sufficient density inside any single commercial relationship.

It has to be manufactured by something that has its own reason to be interesting every day, and then made available to brands that do not.

Key points: (a) The B2C track is not a second, unrelated business. It is the supply side of the Email Venture. (b) B2B creates the economic demand for attention. B2C manufactures the attention. One venture, two motions.

Framing it any other way causes immediate and predictable damage. Treated as a third venture it competes for resource against the ventures it exists to supply, gets measured on consumer revenue it was never intended to produce in its first eighteen months, and is killed for missing a target that was never the point. Treated as the supply side, it is measured on the only thing that matters early: does the attention exist, and does it hold.

A brand has moments. A habit needs a heartbeat.

Email Earns from Attention

Changing seats

The third income line is the one that reverses the direction of the money.

In adtech the arrangement has five components: a brand, a platform, a publisher, a surface and a customer. The brand pays the platform, the platform pays the publisher, the publisher owns the surface, and the customer — very frequently the brand’s own existing customer — is reached through it. The brand experiences this as acquisition. The platform books it as rent.

NeoMails keep all five components and change exactly one thing: the brand takes the publisher’s seat.

It rebuilds attention with its own customer base through daily email worth opening. That attention, once rebuilt, has commercial value — not because the brand wants to advertise to its own customers, which it does not, but because other advertisers in adjacent categories will pay to reach people who are demonstrably attentive and verifiably identified. Through ActionAds those advertisers pay the brand. The money arrow has turned round.

Figure 6 — Same five components. The brand changes seats and the money arrow reverses.

Why this is not sponsored newsletters

The obvious objection is that ads in email are forty years old, and it is a fair objection to the shallow version. Three things make this structurally different.

The attention was rebuilt deliberately. Sponsored newsletters monetise attention that happens to exist. Here the attention is the output of a designed programme — Relate content, Magnets, daily cadence, composed-at-open — with its own instrumentation and its own decay metric. The inventory exists because it was manufactured, which means it can be manufactured again.

The relationship is the system, not the appendage. In the ordinary model the newsletter is a side channel and the sponsorship is a side revenue. Here the surface is the primary customer relationship and media is one of three income lines running through it. That changes how much abuse it can be allowed to absorb, which changes the governance.

The unit is an action, not an impression. An ActionAd is not a banner. It is a completable unit inside the message: subscribe with one tap, claim a sample, start a trial, book a slot, take the offer. The advertiser is buying a completed action from an identified person, not a probabilistic exposure.

The four things that make the inventory different

Sending economics that make daily rational. Under per-send pricing a daily Relate email is irrational: its return is real but too diffuse to justify the line. When media revenue offsets the send cost, the question changes from can we afford to send? to can we make something worth opening? — which is the only question that was ever worth asking.

One-tap subscribe. Acquiring a subscriber has always meant a landing page, a form, a confirmation and a drop-off at every stage. When the network operator is already processing the email, the prompt arrives pre-filled. One tap. Explicit, logged, in-context consent from somebody demonstrably engaged. No lead form on any platform matches that consent quality.

Authenticated identity. The advertising industry was built on the fact that brands could not reach their own customers directly at scale and so paid platforms to reach them probabilistically. This inventory is first-party and authenticated. Every identity is real, every opt-in is a real consent event, and the advertiser pays for certainty rather than for probability — while paying less, because there is no auction extracting margin from every impression.

A live-attention quality filter. Only live attention enters the pool. A dormant address sits outside the network until it proves itself with an open. Audience quality is therefore structurally higher than any purchased list, retargeting pool or lookalike — and it improves as the network grows, because more participating brands means more signals and more verified attention.

The progression is worth stating once, cleanly: NeoMails create the surface. ActionAds create the unit. NeoNet creates the marketplace.

The governor

Ad load must be set by what happens next, not by what fills today.

The operating rule is straightforward and rarely followed: introduce monetisation only after an action has been earned, observe whether attention persists over the following weeks, and adjust the load on the evidence of whether engagement survived — not on fill rate and not on yield. A useful email can carry a sponsor from the beginning. What it cannot do is let the customer form the impression that the advertising is the reason the email exists. The moment that impression forms the inventory begins to die, and it dies faster than it was built.

The metric this rule optimises deserves a name, because the wrong metric is what kills these programmes. It is not yield per open. It is attention-adjusted yield — the economic value taken today, net of the effect on the attention that will or will not be there next week. A programme optimising raw yield will always over-monetise, because the cost lands in a future quarter and the revenue lands now. A programme optimising attention-adjusted yield treats ad load as a treatment to be tested against a holdout, like everything else in this architecture.

Key point: A marketplace that consumes its own supply is not a network. It is a liquidation.

This is where the third income line does its structural work. The party holding the dial is the party whose carry depends on the attention surviving.

Where the model does not apply

One carve-out, stated here so that nobody has to discover it in a compliance review.

In regulated financial services, reward currencies and third-party advertising inside customer communications run into restrictions that are not going to be argued away. For those brands the media term is unavailable in its standard form, and the compliant path is a curated attention digest without the reward and third-party layers. The first two income lines are unaffected. The equation still works; it simply runs on three terms instead of four.

The second ledger

There is a column the revenue statement does not show, and it is the more durable half of the proposition.

Each of the three income lines produces a signal as well as a payment. The Act rung produces declared intent, preference, consent and completion behaviour — the individual data the whole personalisation architecture runs on. The Run rung produces the Decision Trace: context, treatment, counterfactual, outcome, next state. The Network rung produces cross-category declared interest that no single brand can observe from inside its own relationship, because it can only be seen where several brands’ attention meets.

This is the literal content of Email for Revenue & Data, extended to the whole ladder. Revenue makes the programme self-funding. Data makes the next decision smarter — and unlike the revenue, it cannot be copied. A competitor can replicate a payment block, an ActionAd unit, a holdout design. It cannot replicate the accumulated join of declared intent and verified outcome, because that is a function of time and volume, not of engineering.

Figure 7 — The second ledger. Every income line pays once in revenue and once in data that compounds.

The statement, completed

Email value	Mechanism	Commercial model	Rung
Action and data	Pay-in-Email, Tell-in-Email	Capability or usage fee	Act
Incremental outcomes	Recover, Protect, Grow	Beta + Alpha + Carry	Run
Trusted attention	ActionAds, NeoNet	Share of completed actions	Network

Three income lines, three mechanisms, three commercial models, three rungs of one ladder. The first is available to a brand this quarter. The second requires an operator willing to be measured. The third requires a surface people open by choice.

The question this essay cannot answer

The media term is the largest of the three and it is the only one that depends on something a brand cannot manufacture on its own.

Capability revenue needs a good product and a competent send. Outcome revenue needs an accountable operator and a holdout. Media revenue needs recurring, voluntary, daily attention — and a bank, an airline, a retailer or a coffee company does not have enough naturally occurring reasons to be opened every day. It has moments. A ritual needs something else.

So the statement ends on a question it cannot resolve from inside the B2B model. Who manufactures enough recurring inbox attention to make the media line large?

That is the next essay.

***

The old email statement asked how little the sending could cost. The new one asks how much the attention can be worth. Those are not the same question, and only one of them has an interesting answer.

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.