Rajesh Jain

III

What Bending Spoons Teaches

4   The shared platform

Don’t buy the database. Build the factory that makes operating the next one cheaper than the last.

Bending Spoons began in 2013 as a small mobile-app developer and became something else: a company that acquires established digital products and runs them on its own operating machine. It has made more than fifty acquisitions. Its main businesses include Evernote, WeTransfer, Vimeo, Eventbrite and AOL. In September 2026, two months after listing on Nasdaq, it completed its $1.285 billion purchase of Airtable, and within days agreed to buy Miro for $1.355 billion. Many of these products were famous, widely used and under-run. Their users had not left. Their economics had stalled.

The easy reading of Bending Spoons is that it buys bargains. That misses the point. The company describes itself as built on operational excellence: it acquires digital businesses, transforms them, reinvests the earnings and repeats. It keeps what it buys, and says it has never sold a material business. The transformation is the business. Acquired products move onto a common technology platform, common processes and a common way of deciding on pricing, product and cost. Each new product benefits from everything the previous ones taught. The machine is the asset; the products are what it operates.

That is the lesson worth taking. A product that is uneconomic for its own management to run well can become profitable inside a machine that runs many such products. The fixed costs of excellence, such as engineering, analytics, pricing expertise and experimentation, are spread across the portfolio. The marginal cost of operating one more product falls with each one added.

Now apply that logic to the dormant half of a brand’s database. One brand cannot justify an editorial team, a games studio, a deliverability practice and an advertising sales operation for customers who are not buying. An operator serving a hundred brands can. What is uneconomic for one brand to operate can become profitable when a shared platform operates it for a hundred. The editorial templates built for one travel brand serve the next. The advertiser who buys an action in one career Digest can buy it across five. The deliverability playbook refined on the first dormant list makes the fiftieth safer.

Applied to attention, the Bending Spoons model is not about buying databases. It is about building the factory that makes each additional database cheaper and quicker to operate than the one before.

Figure 2. The portfolio is operated through one shared machine. The machine is what improves.

5   Where the analogy breaks

The operator runs the asset. It never owns it.

Analogies are useful until they are taken literally, and this one breaks in three places.

First, customers’ consent cannot be bought. Bending Spoons acquires companies outright, along with the user relationships those products hold. No operator can acquire a brand’s customers that way, and none should try. Permission to receive email from a brand is permission given to that brand. The operator’s role is to run the dormant audience on the brand’s behalf, under the brand’s name, inside the brand’s consent, and under data-protection law that holds the brand accountable for what happens. In India, the Digital Personal Data Protection Act frames exactly this relationship.

Second, the brand stays in the room. A Bending Spoons acquisition ends the previous owner’s involvement. Here the brand remains the owner of the customer relationship, the data and the right to stop. It approves the content categories, the advertisers and the limits on cadence. The operator earns the right to rebuild attention. It does not earn the customer.

Third, the analogy carries a warning. Bending Spoons’ acquisitions have often been followed by deep workforce cuts and higher subscription prices, and the company has critics among users of the products it bought. The inbox equivalent of that error is over-monetisation: treating restored attention as inventory to harvest rather than a relationship to keep. An operator that fills a dormant customer’s Digest with advertising will lose that customer a second time, and this time for good. The discipline has to be built into the product: at most one clearly labelled partner action per edition, category exclusions that the brand controls, and a hard stop when complaint rates rise.

With those corrections, the analogy holds where it matters. A portfolio of dormant audiences is operated through one shared machine, and the machine is what improves. Every additional brand should be cheaper to onboard than the last one in its category. If it is not, there is no factory, only a services business with a clever name.

II

Why Brands Abandon the Dormant Half

3   The cost is not the send

The dormant half is not abandoned because it is worthless, but because it is expensive to operate alone.

The obvious explanation for why brands neglect dormant customers is that emailing them costs money. It is the wrong explanation. On modern infrastructure, the marginal cost of sending an email is a small fraction of a cent. A brand could email its entire dormant half every week for less than it spends on a single day of paid search.

The real cost is producing something worth opening. A dormant customer has already ignored months of the brand’s promotions. Another discount, another new arrival, another “we miss you” will not change that. What might change it is something the customer values for its own sake: a useful piece of content, a small game, a question worth answering, an object worth keeping. Producing that every week, for an audience that is not currently buying, needs an editorial team, a games studio and a content operation. No CRM team has these, and no CMO can justify building them for customers who generate no revenue.

The second cost is risk. Sending to unresponsive addresses is the fastest way to damage a sender’s reputation with the large mailbox providers, and that reputation is shared with the active programme that pays the bills. A brand that mails its dormant half carelessly can find its promotional email landing in spam for everyone. Mailbox providers have tightened their rules for bulk senders, and the complaint thresholds they tolerate are low. From the CRM team’s seat, the dormant half is not an opportunity. It is a threat to the part of the database that works.

So the rational decision for any single brand is to leave the dormant half alone and pay to reacquire whoever comes back. It is rational for the brand. It is wasteful for the system.

A dormant audience needs a publisher, not another campaign. Look closer and the problem breaks into four obstacles, each of which a single brand struggles to clear.

The first is content. The dormant customer needs a reason to open that has nothing to do with selling. That is a publishing problem: a recurring edition with something worth reading, something worth doing and something worth keeping. I described this format in The Brand Digest: Email’s Missing Publication. Brands are not publishers, and their CRM teams are measured on campaigns, not editions.

The second is the sending domain. Any serious attempt to reach dormant customers needs isolated sending, careful volume ramps, validation before every send, suppression of addresses that should not be contacted, and an automatic stop tied to complaint rates. These are specialist disciplines. Getting them wrong once can cost a brand months of deliverability.

The third is monetisation. Even when a brand restores attention, it has no way to earn from it between purchases. Commerce media networks have shown that verified attention close to intent is valuable to advertisers, but building an advertising sales operation to monetise a few hundred thousand re-engaged readers makes no sense for one brand. The attention would be restored, and then left idle again.

The fourth is measurement. The industry’s default signal of attention, the open, has become unreliable. Privacy features in some mail clients pre-load images, and security scanners open emails that no person has seen. A brand that counts opens as reactivation will move machine-opened addresses back into its promotional programme and damage the reputation it set out to protect. Real reactivation needs a verified human action, and a holdout group to show whether that action would have happened anyway.

Each obstacle carries a fixed cost: an editorial operation, a deliverability practice, an advertising sales team, a measurement system. Each can be justified on its own. Together, for one brand’s dormant customers, they fail the budget meeting. Shared across a hundred brands, each becomes small. That is the opening the rest of this essay is about.

How brands can turn forgotten customer databases into living, profitable audiences, without paying to send another email.

Every brand owns a database it has stopped operating. This essay is about who should operate it, on what terms, and why only a portfolio can make it pay.

The argument in one line. A dormant database is not a dead asset; it is an unoperated one. What is uneconomic for one brand to operate can become profitable when a shared platform operates it for a hundred brands: recovering customers at no charge to the brand, earning media revenue from the attention it restores, and taking a share of transaction uplift only where it can prove it.

I

The Graveyard in Every Database

1   The silent majority

The brand paid to build the database, and then stopped operating it.

Every large consumer brand owns a graveyard. It sits inside the customer database: millions of email addresses collected over years of advertising, registrations, purchases and loyalty sign-ups. Each one cost money to acquire. Each one belonged to someone who, at some point, had a reason to engage. Then, gradually and without any single moment anyone could point to, the relationship went quiet.

Marketing calls these customers dormant. The term is polite. It means the brand has stopped expecting anything from them. They have not unsubscribed. Their addresses have not bounced. They simply stopped opening, and the brand responded by sending them less, then nothing, then forgetting they were there. In a mature database it is rarely a fringe, and it is often a large share of the list.

The accounting is strange. Finance rarely gives dormant customer attention any separately measurable value, because it produces no revenue anyone can point to. Deliverability teams treat it as a hazard, because sending to unresponsive addresses damages the sender reputation that the active programme depends on. And the acquisition team, under pressure to grow, goes back to Google and Meta to find customers. Some of the people it pays to reach are already in the database. The brand is buying back, at auction prices, attention it once owned and let lapse. That is the reacquisition tax: the share of marketing spend that goes to winning back customers the brand already had.

So the dormant half sits in an odd position. It is too risky to email, too valuable to delete and too quiet to count. Nobody owns it. The CRM team’s targets are built on the responsive base. The acquisition team’s targets are built on new customers, and a dormant customer it reacquires looks, on its dashboard, like a new one. The dormant half falls between two budgets and belongs to neither.

This essay argues that the problem is not the asset. It is the economics of operating that asset one brand at a time. It takes its frame from an unlikely place: a Milan-based software company that has built a business valued at about $18 billion at its July 2026 listing, and above $33 billion after Berkshire Hathaway added to its stake this month, by taking over products that other people had stopped running well.

2   A database is not an audience

A dormant list is not an audience. It is the raw material for one.

Start with a distinction that most marketing dashboards blur. A database is a list of addresses the brand is permitted to email. An audience is the set of people who will pay attention when it does. The first is an accounting asset. The second is an attention asset. Real Reach measures the gap between them: the number of customers who have engaged in the last ninety days, set against the size of the list.

Consider an illustrative brand with ten million addresses. Two million read its email with some regularity. Three million are fading: they still open occasionally, but less often than last year. Five million have not engaged in months. The brand’s list says ten million. Its Real Reach says two.

Figure 1. The accounting asset and the attention asset. Illustrative split, not a market estimate.

This matters because almost every email metric is computed against the wrong denominator. Open rates and click rates are reported on sends, so a brand that stops emailing its dormant half looks healthier, not weaker. The list stays large, the rates improve, and nobody sees that the attention asset is shrinking.

It is also worth being precise about what the dormant half contains. It is not an audience waiting to be switched back on. Some of those people no longer need the category. Some have moved to a competitor. Some addresses have been abandoned, and a few have been recycled into traps that mailbox providers use to catch careless senders. Some customers withdrew their permission in spirit long ago, even if they never clicked unsubscribe. A dormant list contains a potential audience, mixed with addresses that should never be emailed again.

That is why the argument has to be made carefully. The opportunity is not to email the dormant half. It is to find the part of it that can still become an audience, earn that audience back, and leave the rest alone. Doing that well is expensive, and the expense is the reason almost no brand does it.

Published October 29, 2026

Personal AI agents will take over the chores. The opportunity lies in what people still want to do themselves.

In September, the inbox got a doorman. And the doorman acts.

The argument in one line. Personal AI agents will handle the chores of digital life: sorting, comparing, booking, paying, filing. That frees human attention, but it does not hand that attention to brands. The time people save will go to what they want to do themselves: learn, play, collect, compete and belong. Brands that build those experiences, and use AI to make them richer rather than to remove the person, will earn a place the agent is told to protect. Agents will clear the inbox. The winners will build what humans choose to come back for.

The Inbox Has Two Readers argued that every message now meets a person and the person’s AI agent, and that utility will move to the agent while experience stays with the person. That essay asked how one message can serve two readers. This one asks the next question: once the agent has handled everything it can, what does the human come back for?

1

The agents arrive

In one month, personal agents moved from reading to acting.

Picture two mornings. On the first, a person wakes to twenty-seven unread emails: a bank confirmation, a discount on something bought last week, a hotel asking for feedback, receipts, a renewal reminder, a newsletter. They skim, postpone, delete, and miss the one message that mattered. On the second, their agent has already filed the receipts, checked the renewal and set aside two decisions that need approval. The inbox is quiet. The marketer’s question is where their messages went. The more interesting question is what the person now chooses to open.

The second morning is no longer hypothetical. On 8 September, Meta introduced Muse, a personal agent that runs on its own secure virtual machine in the cloud, works across a person’s connected apps, and sends emails, books travel and shops, asking for approval before consequential steps. On 29 September, OpenAI introduced dots: always-on agents, each with its own cloud computer, reaching more than 4,000 apps and doing some work before anyone asks for it. That unprompted background work is restricted to reading: actions that touch a person’s accounts are checked against their rules before they run. Alongside them sits Instinct, an invite-only agent that people reach by text message. On 5 October it brought its agent into group chats, where it can help a group of friends plan a trip, buy tickets or decide who brings what, even if some of those friends have no account.

Mark Zuckerberg’s stated ambition is an agent for everyone, working around the clock on their behalf. Whether or not that arrives on schedule, the direction is no longer speculative. The largest consumer technology companies in the world are now competing to stand between people and their digital lives.

None of this is finished. OpenAI tells users to review consequential work because dots can still make mistakes. Early users have already posted reports of agents acting without approval, some of them still unconfirmed. Most people do not yet use any of these products, and many will never grant an agent access to their inbox.

But the test I set in The Inbox Has Two Readers has moved faster than I expected. That essay gave the two-reader thesis until 31 December 2028 and asked for three kinds of evidence: capability, adoption and sender behaviour. Within weeks, capability has largely arrived. Adoption is the open question. Sender behaviour has not started: almost every brand still writes every message as though a person will read it.

The inbox now has a doorman, and the doorman acts.

2

The paradox of delegation

The more effort AI removes, the more we notice which effort was the point.

There are two kinds of things we do. Some we do for the outcome. Others we do for the experience. An agent that files my tax return is a gift: I want the return filed, and I would happily never see the form. An agent that completes my crossword has destroyed the only reason the crossword existed.

The AI industry mostly treats human effort as waste to be removed. Much of it is. But not all effort is waste. Some of it is the reason we take part.

History is clear on this. When Deep Blue beat Garry Kasparov in 1997, many assumed chess had lost its point. Today more people play chess than ever, and the engines that beat the world champion have become the coaches that help ordinary players improve. Cars did not end the marathon. Calculators did not end Sudoku. Any computer can solve the day’s Wordle in an instant, and millions still play it every morning, because the point was never the answer. The point was solving it, and sharing the grid with friends.

Psychology explains why. Richard Ryan and Edward Deci’s self-determination theory identifies three needs behind intrinsic motivation: autonomy, the sense of choosing for oneself; competence, the sense of getting better at something; and relatedness, the sense of belonging with others. Learning, play and shared collecting satisfy all three. Delegating them to software satisfies none.

The same research carries a warning, though not a blanket one. A meta-analysis of 128 experiments by Deci, Richard Koestner and Ryan found that tangible rewards people expected to receive tended to weaken intrinsic motivation for interesting activities, while positive feedback strengthened it. Other researchers dispute how general the effect is. The design lesson survives the debate: recognition helps, and paying people to play can turn play into work. I will return to this, because it shapes how any reward for attention should be designed.

Figure 1 sorts the activities of digital life by what people are likely to hand over.

Figure 1 — The delegation spectrum. Chores move to the agent; discovery is assisted; learning, play and belonging stay with the person.

The middle column matters. Agents will help people find what to explore, coordinate plans and remind them of what they chose. The right-hand column is where the person wants to do the thing themselves. The Inbox Has Two Readers drew a similar line between routine choice, which will be delegated, and expressive choice, which stays human. Window or aisle is routine. Which Card to give my sister is expressive.

This gives every product built in the age of agents one design test.

Does the AI create value by doing the activity for me, or by helping me enjoy doing it myself?

3

Two markets

Agents save time. The next opportunity is making saved time worth spending.

AI is creating two markets at once.

The first is the market for delegated work. Machines compete to deliver results with less human involvement: fewer clicks, fewer forms, fewer decisions. It is measured in tasks completed, minutes saved and errors avoided. This is where Meta, OpenAI, Google and a crowd of well-funded start-ups now compete.

The second is the market for chosen experiences. Companies compete to offer things people want to spend their own attention on. It is measured in voluntary return, mastery and belonging. Media, games and social networks have always lived here.

The first market eliminates effort. The second makes effort rewarding.

Brands sit awkwardly between the two. They hold something most companies would envy: permission to reach millions of customers directly, in a domain those customers already care about, such as their money, their home, their car or their health. Almost all of that reach is spent on Sell and Notify: offers, confirmations, alerts and reminders. That is exactly the lane agents are taking over.

The relationship behind those messages does not disappear. An agent still needs trustworthy prices, terms and records from the brand, and it will reward brands that supply them cleanly. What disappears is the person’s attention to them.

Brands are losing the human’s attention for the messages they send most, and most have never built a message a human would choose.

4

Freed time does not belong to brands

Scarcer attention is more valuable and harder to earn at the same time.

There is a tempting version of this argument, and it has a flaw in the middle. It goes: agents free up human attention; people want experiences; brands should provide experiences; therefore brands win. The middle step is wrong.

If an agent clears the inbox chores, a person gains some minutes each day. Those minutes do not arrive with a brand’s name on them, and brands have no claim on the dividend. By default the time goes where attention already flows: short video, social feeds, games, messages with friends, creators people follow. These industries have spent twenty years engineering for attention, and they are very good at it.

So a brand’s Digest is no longer competing with other promotional emails. It is competing with everything else on the phone. The bar is the quality of good media, not the quality of a good campaign.

Most brand content does not clear that bar. A newsletter that is a catalogue in disguise will be summarised by the agent in a single line, “new arrivals, 20 per cent off, nothing urgent”, and the person will be right not to miss it.

Two disciplines follow. The reader, not the brand, should choose the frequency: a fund house may earn two visits a week, a grocer a weekend edition, a museum a season around an exhibition. And quality matters more as content becomes cheaper to make. A hundred generated facts do not make an experience. Someone still has to decide why a story matters, what makes a challenge fair and what the reader will carry away.

Brands do hold three advantages that media lacks. They already have the relationship. They operate in a domain where the customer has a stake. And they can reward with things only they can give: access, recognition, an experience tied to what the customer already owns. The question is whether they will use those advantages to build something worth choosing.

The opportunity is real, but it belongs to brands willing to be as good as media.

5

The doorman

People will not tell their agent to protect a brand. They will tell it to protect something they love.

Email has always had a delivery problem: spam filters, promotions tabs, sender reputation. The agent adds a second and more consequential gate. It decides what reaches the person at all.

That gate is already being built. Google’s AI Inbox, still in beta, reorganises email around to-dos and topics, and Google’s own help page says it surfaces information only from the Primary tab, leaving out Social and Promotions. That is one product’s rule, not a law that every agent will follow. But it shows the shape of the problem: a brand can be delivered perfectly and still never reach the person. Figure 2 shows the three fates of a brand message once an agent stands at the door.

Figure 2 — The doorman. The agent acts on transactions, summarises information and shows the person only what they have asked to keep.

One response will be to make every message sound urgent. It will fail. An agent serving the person will learn to discount manufactured urgency, just as mailbox providers learnt to reject spam.

What gets through the door is the principal’s instruction. And here the shift is subtle but important. A person is unlikely to tell an agent “always show me everything from my fund house”. They are far more likely to say: “Handle the fund’s paperwork, but don’t hide my investing Cards.”

The customer subscribes to the experience, not to the brand. In an agent-mediated world, it may be the person’s Album, their Set in progress, their weekly challenge with a friend, that earns a protected place in daily attention. The brand’s logo will not.

How many recurring experiences will each person protect? Nobody knows yet. I suspect few, but the number matters less than the mechanism. An agent learns what to surface from what its principal does and says. A customer’s engagement history may increasingly decide what their agent chooses to show them. That is the reason to start earning voluntary attention now: not because a door is about to close, but because every week of chosen engagement becomes evidence the agent will read. No position will be permanent, and a better experience can displace an older one. But every position starts with a record, and a record takes time to build.

Three implications follow.

Build explicit preference now. Let people say what they want and when. “My investing Digest on Tuesdays and Fridays” is a far stronger signal than a sender’s decision to send. Today it improves engagement. Tomorrow it is the instruction an agent can read.

Make subscriptions portable. A person’s Album, history and Sets should live beyond any single message and be reachable from the web and through other permissioned routes. Email is an excellent front door. It need not be the only room.

Do not design for one platform’s whitelist. The rules agents follow will change many times. A human preference, clearly expressed, is the one signal likely to survive every version.

There is a further reason to rely on the person’s stated choice. The doorman has its own business model to find. Instinct is free while in beta, and reports on its funding list advertising among its possible routes to revenue. A door that can be sold is a door brands should not depend on buying. The only reliable way through is to be something the principal asked for.

6

What survives the doorman

An agent can summarise what a Card says. It cannot summarise what you did to earn it.

Information alone will not survive. Any fact in a Digest can be retrieved by an agent in a second. If a brand’s relationship with its customers rests on information, the agent will absorb it.

What survives is participation history. Consider someone who has collected 37 of the 50 Cards in a Set about great investors, made predictions before each reveal, mastered 22 of them in recall challenges and traded duplicates with five friends. An agent can report that state. It cannot reproduce its value, because the value lies in having done it. That record is personal and cumulative, and it cannot be delegated without being destroyed.

That describes the destination sketched in The Next Turn. The starting line for NeoCards, as Issue a Collectible set it out, is simpler: get a Card, open it, keep it. Prediction, trading and Circles come later, as people show they want them. The argument holds at the starting line too, because even a Card kept is something the person chose and the agent did not.

This gives the Digest’s three blocks a sharper purpose.

Read must offer something a summary cannot replace. That is not only a matter of insight. A novel can be summarised in three paragraphs, a film reduced to its plot, a match to its score, and people still read, watch and follow, because the experience was the point. The test is not whether an agent can summarise the content. It is whether the summary can substitute for the experience. A story told well, a funny observation, a beautiful image or a surprising connection passes that test. A list of new arrivals does not.

Do must ask for the person’s own judgement: predict before the reveal, recall what you learned last week, choose between two possibilities, challenge a friend. These are expressive acts, not response signals.

Keep must mean that tomorrow differs because of what you did today: a Card owned, a Set closer to completion, a mastery that persists.

And the Album must belong to the person, not the brand. A fund house might issue an investing Set, a museum a history Set, a publisher a science Set and a parent a family Set, all collected in one shared Album. The issuer provides content and distribution. The person accumulates identity and history. That is far harder to displace than any single newsletter, and it gives the person something worth protecting from the doorman.

Figure 3 shows how one relationship can serve both readers.

Figure 3 — One relationship, two lanes. The fund house sends the agent facts it can verify and act on, and sends the person an experience worth choosing.

The agent receives what it needs: the SIP was processed, the portfolio value is updated, the KYC is complete, nothing needs action. The person receives a Brand Digest: which asset class did best over the past twenty years, a guess to make before the answer is revealed, a new Card in the 50 Great Investors Set, a friend to challenge. In a regulated category like this one, the human lane stays with learning and play: no wagers, no third-party offers, no investment advice dressed as entertainment.

The two lanes reinforce each other. MarTech 2027 made the point that a machine has no loyalty, but the person it serves does. A customer who values what a fund house teaches them is more likely to trust it, and that trust shapes the preferences their agent will follow when it next compares funds.

7

AI that makes people better, not idle

Most AI products remove the human. The better opportunity is AI that makes the human’s part more rewarding.

So far this essay has treated agents as gatekeepers or administrators. There is a third role, and it may be the most important. AI can make the human experience much better.

Consider a Card about a great investor. Today it carries a short biography, a famous idea and a question. With AI, the same Card can offer a two-minute challenge adapted to what I already know. If I answer wrongly, it explains the idea with an example drawn from my own situation rather than a textbook. A month later, it checks whether I still remember, and suggests a friend to challenge next.

Figure 4 — The same Card, three ways. AI makes the challenge personal and the learning lasting; the person still does the thinking.

The person still does the thinking. The AI makes the thinking more rewarding. That is the reverse of most agentic products, and a far stronger philosophy than trying to tell human clicks from machine ones. It is also the argument of When AI Gets Cheaper, What Gets More Valuable?: cheap intelligence becomes an input to an experience whose value comes from somewhere else.

The agent has a place here too, as an ally rather than a substitute. It can remind me that I am two Cards short of a Set, find the friend who holds my duplicate and arrange the swap. Instinct’s group chats, where one agent helps several friends agree a plan, show that such coordination already works. Logistics can be delegated. Achievement cannot. The agent must never make the prediction, earn mastery or claim recognition on my behalf.

There is a further loop. A person who learns about investing through a Set comes to know more, and what they know shapes what they ask their agent to look for. The human lane can inform the preferences through which agent-mediated commerce runs. That is exactly why it must never become covert persuasion: no hidden instructions to the agent, no nudges dressed up as lessons. The person’s autonomy is the value proposition. An experience used to steer the agent behind its principal’s back destroys the trust that made it worth protecting.

Two design principles protect the human lane.

Keep it Panini, not crypto. Cards should be earned through human acts and hold social, educational and sentimental value rather than resale value. The moment a collectible becomes a financial asset, bots arrive to farm, snipe and arbitrage it, as sneaker drops, ticket sales and NFT markets have shown. The motivation research adds a second reason: rewards that feel like payment can turn play into work. Rewards should recognise attention, not purchase it. The same test applies to Mu, the attention currency I have written about. Earned for completed acts and declared preferences, and shown as a record of progress, it recognises participation. Paid out simply for opening, it risks becoming the kind of reward the research warns about. Mu should be designed as the first, not the second.

Treat everything an agent reads as data, never as instructions. Any content a brand exposes to agents must be incapable of steering them. Researchers have shown repeatedly that text hidden in a web page or an email can redirect an AI agent, so the line is easy to cross. Brands that want to be trusted by their customers’ agents must be the first to respect the boundary.

8

The economics, in the right order

Earn the habit first. Price the attention later.

It is tempting to jump from “people participate” to “advertisers will pay a premium”. That jump skips two steps. The value of the human lane builds in three stages, each earning permission for the next.

Figure 5 — Three assets in sequence. Attention enables trust; together they make commercial value possible. The heights show order, not size.

Attention comes first: people return and take part by choice. Trust follows: the experience is useful, enjoyable, authenticated and hard to game, so people recall it, share it and tell their agents to keep it. Only then does commercial value become possible: sponsored Sets and in-email actions that produce measurable incremental results without damaging the first two.

The unit of account is the one Paid to Send. Paid to Keep. proposed: the Engaged Edition, meaning one person, one edition and at least one credible human action. With agents in the inbox, that unit matters more, because a message processed by software is not a message experienced by a person. It sits alongside Habit Rate, the measure Points Reward Purchases. Cards Reward Attention. proposed: the Engaged Edition counts what happens in each edition, and Habit Rate counts whether people keep coming back. One measures the visit, the other the habit.

Every CMO will ask one question here. If the Album belongs to the person and holds Sets from many issuers, who captures the value? Suppose a fund house funds fifty investing Cards, and the customer comes to love the Album and returns to it every week. Is the loyalty to the Album or to the fund house?

Both, and they must be measured separately. The Album earns the recurring visit. The issuer must earn attribution: recognition as the source, a reputation for knowing its category, trust and, in time, preference when the customer or their agent next chooses. Points Reward Purchases. Cards Reward Attention. made Issuer Recall a core measure for exactly this reason. If people remember the Cards but not who issued them, the brand has funded someone else’s entertainment. A brand will keep paying for Sets only if that benefit shows up against a holdout. Without it, the Album may still be a good consumer product; it will not be a brand-funded one.

Verified human attention is a prize, not yet a moat. As agents browse, click and buy, web traffic will fill with machine activity, and advertisers will trust impressions and clicks less. A person answering a question, making a prediction or keeping a Card sends a stronger signal than an open. It is not proof: agents can answer questions too. Any billing standard must rest on authentication, fraud checks and auditable evidence, not on a claim of perfect verification. And even real human participation commands a premium only when it reaches relevant people and produces results an advertiser can measure. The premium has to be demonstrated before it is priced.

For every brand, the strategic choice comes down to two lanes. Make Sell and Notify legible to agents: clean offers, honest terms, clear permissions, verifiable provenance. Make the Digest worth returning to as a human: things to read, do and keep. The two lanes share an identity but not a success metric. One relationship, two ways to serve it.

Neither lane can rescue a weak product. When the customer’s agent compares options, it reads the factual record: price, terms, returns, service history. A clever Set will not hide a poor one. Affection may earn consideration. Operational truth decides whether the brand is chosen.

One guardrail belongs here. The goal is not to extract as many minutes as possible from people. That is the logic of the feed, and it is the logic personal agents may help people escape. Let people choose their days and pause at will. Prefer learning remembered a month later to time spent in the email, and an unprompted invitation to a friend to a manufactured share button.

The brands that win will be those people choose to spend time with, not those that take the most of it.

9

What would prove this wrong

The architecture is elegant. Only the cohorts can say whether it is right.

Every claim in this essay can be tested, and most within a single quarter, using concurrent, randomised cohorts rather than last year’s numbers as the baseline. The test runs in two stages, because the product contains two kinds of experience.

Cards, Sets and adaptive explanations are individual experiences, and can be randomised person by person. Circles and trading are network experiences: their value depends on who else takes part. Randomise people individually and the social cohort may look weak only because its members have no one to trade with. Let them mix freely and the controls are contaminated. So the first stage tests the individual product, extending the three-arm test in Points Reward Purchases. Cards Reward Attention. with an arm for AI, and the second randomises existing groups, such as families, friends or colleagues, rather than individuals.

Stage	Cohort	What it receives	What it tests
1	Control	The brand’s existing emails	The baseline
1	A	Brand Digest: Read and Do	Does a better publication earn return?
1	B	A plus Keep: Cards and Sets	Does ownership strengthen the habit?
1	B + AI	B with adaptive Cards	Does AI improve enjoyment and learning?
2	Groups	B, randomised by existing group, with or without Circles and trading	Does belonging strengthen the habit further?

Table 1 — The pilot in two stages. Individual features are randomised by person; social features by group.

Measure Habit Rate and repeat voluntary participation at weeks 4, 8 and 12; returns to the Album without an email prompt; recall after a delay; unsolicited shares; Issuer Recall and brand preference; and attention decay against the control. Keep transactional emails identical across cohorts and add no new promotional incentives.

Ask people whether they would tell their agent to always show them this, but treat the answer as intention, not behaviour. The stronger evidence is what they do: subscribe explicitly, choose a schedule, come back unprompted. As agent use grows, add one more comparison: readers whose inboxes are largely agent-mediated against similar readers without that mediation.

Six results would weaken or kill the thesis.

People do not protect it. Few choose a schedule or return to the Album unprompted. Then the experience is something people open when it appears, not something they choose.

Keep adds nothing. Cohort B shows no better voluntary return than Cohort A by week 12. Then ownership is decoration, and the Digest must succeed as a publication alone.

AI adds interaction, not value. Adaptive Cards produce more taps without better learning or enjoyment. Then the AI is optimising the wrong outcome.

People delegate the experience too. Users happily let agents collect, trade and complete on their behalf. Then even objects can be absorbed, and the human lane is narrower than I think.

The issuer gets nothing. Participation rises, but Issuer Recall, attention decay, dormancy and incremental customer value do not move against the holdout. Then the Album may be a worthwhile consumer experience without being a brand retention tool. That would be a valid finding, not a failure to conceal.

People do not want relationships with brands. However good the design, most people reject ongoing relationships with most brands. A person-owned Album shared across many issuers is the strongest answer to this risk, because people collect what they love from whoever issues it. But it is an answer, not a guarantee.

The wider category test runs alongside the one in The Inbox Has Two Readers, scored on 31 December 2028. If personal agents remain niche, the doorman argument weakens. The attention argument does not: the Digest still competes with everything else on the phone.

**

There are two ways AI can give us back our time. One is to do what we previously had to do. The other is to help us do more of what we love doing. Almost every company in the agent race is building the first. Very few are building the second, and brands, with their direct relationships and their permission to reach people every day, are better placed than most to try.

Agents will clear the inbox. The winners will build what humans choose to come back for.

In the age of artificial intelligence, the most valuable products may be those that make us want to use our own intelligence.

Published October 28, 2026

Growth is not a number to target. It is an equation to improve.

Marketing plans for 15% because it asks for better execution. Ask instead what it would take to double each of four levers, reach, attention, action and value, and the answers change. This essay is about that shift: from 15% moves to 10X moves, who should be paid for making them, and how to prove they work.

1

The 15% Trap

It is October, and somewhere a leadership team is building next year’s marketing plan. The spreadsheet opens with last year’s actuals. The CRM team expects the database to grow by 12%. The email team promises a two-point lift in click rates. The engagement team has a new journey that should improve conversion by a tenth. Retention commits to trimming churn. Finance adds it all up, tests it against the budget, and the plan lands where plans usually land: 15 to 20% growth, funded by more campaigns, more automation and a larger technology budget.

Nothing in that plan is wrong. Every number is defensible, and every team has done its homework. Yet the plan is built to produce a small answer. The plan is small because the question is small. Ask a team for 15% and it will find 15%, by doing what it already does a little better.

The first problem is the starting point. Budgets begin with last year, so teams are asked how much better their part of the machine can get, not whether the machine is the right shape. Email optimises sends and clicks. Engagement optimises journeys and conversions. Acquisition optimises cost per customer and return on ad spend. Each improvement is real, and each is measured on its own dashboard. None is asked how it combines with the others.

That leads to the second problem: fragmentation. Reach belongs to one team, engagement to another, conversion to a third, retention to a fourth. They report to different leaders and are rewarded against different goals. Nobody owns the multiplication. Worse, a local win can damage the whole. A promotional calendar that sends harder lifts this quarter’s conversions and quietly erodes the attention next quarter depends on. A discount that lifts transactions thins the margin each transaction earns. Every team hits its number, and the business gets less than the sum.

The third problem is the database illusion. Brands count registered users, subscribers and customer records as assets. Many of them are no longer reachable in any useful sense: emails go unread, app users have not opened the app in months, customers have quietly moved on. As I argued in From Messages to Media, a database is not an audience. The number that matters is Real Reach: the people who have engaged with the brand in the last ninety days, set against the size of the list. For most brands the gap is large. The cost of that gap arrives later, in the paid-media budget, when the brand buys back customers it once had and pays a second time for a relationship it let lapse. I measure that as REACQ%: the share of customers bought through paid media who were already the brand’s customers.

The fourth problem sits underneath the other three. The technology and services marketing buys are paid in ways that reinforce the trap. Email platforms charge per message. Engagement platforms charge per contact or per active user. Agencies charge for activity. Ad platforms charge for paid reach. Each model rewards improvement inside its own category. None is rewarded for changing the economics of the customer relationship as a whole. I will come back to this, because I run one of these companies, and our pricing is part of the problem.

So change the question. Instead of asking how to grow 15% next year, ask how much of the value your existing customers could create you are realising today. Count every visitor who browsed and was never identified, every subscriber whose attention faded, every cart left behind, every form left unfinished, and every purchase made at a discount it did not need. In most businesses, that unrealised value is not a rounding error. It can be several times what the business earns from its customers now, and most of it needs no new acquisition spend at all.

This essay is about one shift: from a 15% mindset to a 10X mindset. Not 10X as a forecast, but 10X as the question that forces different moves.

Marketing has been planning with addition. It should be planning with multiplication.

2

Think in Doublings

Two times two times two times two is sixteen. Four doublings make 16X. Fall short on every one of them, and the product still lands somewhere near 10X.

The arithmetic is the least interesting part of that sentence. Whether a team finds 70%, 80% or 100% in any one place matters far less than what the target does to the team. A 15% target asks for better execution. A doubling asks for a different idea.

Ask an email team for 15% more revenue and it will refine what it already does: sharper subject lines, better send times, one more promotion in the calendar. Ask it to double how often customers give the brand their attention, and none of those moves gets close. The team has to ask a different question: why would anyone come back between purchases? That question leads somewhere new. The same is true of every lever in the business. A doubling cannot be reached by tuning, so it forces the search for a new move.

I should be clear that doubling anything in a mature business is hard. This essay does not argue that 10X is easy. It argues that most plans never look for it, because they are not organised around the levers that multiply.

Figure 1. Four doublings make 16X; even falling short on each lands near 10X. A 15% plan lands at 1.15X.

So what are the levers? For any business that lives on its relationships with customers, I would write the equation like this:

Customer value = Reach × Attention × Action × Value

Measure all four for the same customers over the same period, with value counted as contribution rather than revenue. Each lever asks one question and exposes one leak.

Reach asks how many people we can meaningfully serve. Not records in a database, but known, permissioned people the brand can reach: Real Reach, not list size. The leak is anonymous visitors who are never identified, app users with no contact route, customers with incomplete details, and former customers whose relationship has gone quiet.

Attention asks how often those people give us their attention: reading, browsing, responding, returning, voluntarily. A tracked open is weak evidence, because many opens are now registered by machines rather than people. The leak is attention that decays between purchases because the brand speaks only when it wants to sell.

Action asks how often attention becomes action: completed purchases, applications, bookings and sign-ups. The leak is abandoned carts, unfinished forms, failed searches and journeys with one step too many.

Value asks what each action is worth in contribution, after discounts, costs and returns. The leak is transactions bought with discounts they did not need, thin baskets and missed cross-sells.

In plain words: more people, more often, doing more, worth more.

Figure 2. One equation, four levers, a leak at each. Email earns attention and acts in the inbox; engagement converts across site and app.

The equation is easy to misuse, so one discipline belongs right next to it. The four levers multiply on paper; the moves that pull them do not act independently. Bring dormant customers back into reach and attention per person may fall, because the people you add are less engaged than the people you had. Pay people merely to open and you get more opens from less interested people, which is why rewards should follow completed actions rather than taps. Lift conversion with discounts and value per action shrinks. A good plan pulls each lever in a way that does not push another one back.

Headroom also differs. A neglected programme may have large gaps in three levers; a well-run one may have a gap in only one. The point of thinking in doublings is not to assume every lever can double. It is to ask the doubling question of every lever, and see which ones have an answer.

Asking is the easy part. The harder part is knowing what the 10X moves look like.

3

From 15% Moves to 10X Moves

Every lever can be pulled in two ways. There is the 15% move, which does what the team already does a little better. And there is the 10X move, which changes what the team does. Most marketing plans are made almost entirely of 15% moves.

The previous essay, 10+1 Email Innovations: Pipe to Product, Messages to Money, set out the innovations that turn a single email from a delivery into an action, scored in five stages per thousand emails delivered. This essay steps back from the email to the customer. The innovations are the same; here they appear as moves on four levers, and the score is the whole relationship rather than one send.

Figure 3. Same four levers, different moves. The 15% move refines existing work; the 10X move changes it.

Two engines make the 10X moves: email and customer engagement. They are sold as different categories of software, bought by different teams and reported in different meetings. The customer does not see categories. She sees one brand, reaching her in her inbox, on its website and in its app. The two engines are better understood as two instruments acting on the same four levers.

On Reach, the 15% move is to buy more traffic or add another sign-up pop-up. The 10X move is to identify who is already there. Engagement can recognise far more of the visitors and app users who arrive than most brands bother to. Email can capture contact details from transactions and service interactions, and reawaken quiet relationships before the brand pays to acquire new ones. The test is whether Real Reach rises and REACQ% falls.

On Attention, the 15% move is sharper subject lines and better send times. The 10X move is to give people a reason to return between purchases: a recurring Digest with something useful to read, a small poll or quiz, collectable NeoCards that reward the next open. As I argued in The Next Open, every email has two jobs, and the second is to make the next one more likely to be opened. This is where email leads, because the inbox is where attention can be earned every day.

On Action, the 15% move is to test another button or send another reminder. The 10X move is to put the action inside the message, so the customer can say Yes, Pay or Tell without leaving the inbox. That was the central claim of the 10+1 essay: email no longer only earns attention; it increasingly completes the action. On the site and in the app, engagement carries the rest, with agentic decisioning choosing the next best step for each person. Agents should be judged by the outcomes they create, not by the number of campaigns they generate.

On Value, the 15% move is to deepen the discount or push another promotion. The 10X move is to protect contribution: sell at full price where the customer did not need an incentive, recommend with margin in mind, and build better baskets. For publishers, much of Value comes from advertisers and sponsors rather than purchases. The equation in How Publishers Can 10X Their Revenue, subscribers × newsletters per reader × engagement × yield, is the same four levers in a publisher’s terms.

Email earns attention and increasingly completes the action. Engagement converts it across site and app.

Consider how this can work for one customer. Ishita has not visited a fashion brand’s site in three months. She has not complained or unsubscribed; she has simply drifted, the way most customers do long before any report calls them dormant. That drift is the subject of Before the Slide. On Sunday, she receives the brand’s weekly Digest: a short, useful read with a two-question poll at the end. She answers it. Her answer tells the brand something it did not know: she has moved to a colder city.

Her next visit, from a Digest link, opens on outerwear for her new city rather than the summer range she browsed months ago. She adds a coat to her cart, is called away, and leaves. The next morning, an email lets her complete the purchase inside the message, at full price. She is back, and in this telling no advertisement had to buy her back. Whether that holds across thousands of customers like Ishita is exactly what a holdout group is for.

Each step pulled a different lever. The Digest restored Attention. The poll refreshed what the brand knew about her. The recommendation lifted Action. The full-price purchase protected Value. And because she stayed, she remains within Reach.

This is why the two engines belong in one plan. Coordinated, the levers reinforce each other across channels. Email attention feeds engagement sessions, and what engagement learns sharpens the next email. Uncoordinated, the same two engines duplicate messages, tire customers and contradict each other’s offers. A brand that runs them as separate programmes gets the second outcome by default.

Two clarifications keep the model clean. Retention is not a fifth lever. A retained customer stays within Reach and keeps returning, which shows up as Attention and Action; counting retention separately would count the same gain twice. Likewise, Digests, agents and recovery programmes are not levers. They are the moves that pull them.

If these engines can create this much value, a question follows. Why are the companies that sell them still paid for the inputs?

4

Who Gets Paid for Which Lever

Most marketing technology is paid for activity: per message, per contact, per active user, per licence. These models built large, durable businesses. They also set a ceiling, on the vendor and on the brand. A partner paid per send is paid on volume, not on the equation. It gains when the list grows or the brand sends more. It has no stake in Attention, Action or Value.

It is worse than indifference. Call it the attention paradox. When attention improves, a brand needs fewer reactivation campaigns, fewer reminders and fewer win-back sequences. Under per-send pricing, the vendor is paid less for doing better. It should surprise no one that vendors on that model rarely push hard on attention.

A vendor’s business can be described in four revenue streams that mirror the brand’s four levers:

Vendor revenue = Volume + Attention + Actions + Outcomes

The names are deliberate. Each stream is priced at a different depth of the customer’s equation, and so each gives the vendor a reason to care about a different amount of the customer’s growth.

Figure 4. The capture ladder. The deeper a partner is paid, the more of the customer’s equation it has a reason to improve.

A partner paid on volume cares about Reach and little else. A partner paid per verified attention event cares whether customers keep coming back. A partner paid per completed action also cares whether that attention turns into purchases, bookings and applications. A partner paid a share of verified gain cares about all four, including whether each action makes money.

The further down the equation a partner is paid, the more of your growth it cares about.

Outcome pricing needs structure, or it becomes an argument. In The Alpha Playbook I described it in three parts. Beta is the contribution the brand would have earned anyway, agreed before the work begins. Alpha is the verified contribution above it. Carry is the partner’s negotiated share of Alpha. Price then follows value rather than volume, the shift I described in Paid to Send. Paid to Keep.

There is a trap to avoid here. A brand that creates 10X more value does not owe its partners 10X more money. The partner has to prove its contribution against what would have happened without it. It should take a minority share. And it must never charge twice for one outcome: a recovered customer may generate a send, an active user and a completed action, each a fair charge, but the brand’s gain from that customer is counted once.

Two facts make outcome pricing workable. First, the money does not come from the martech budget. No CMO will pay ten times a software invoice out of a fixed technology line. Outcome fees are paid from profit that would not otherwise exist, and for publishers, attention can be funded partly by advertisers. The budget ceiling that caps the martech industry does not apply to a share of new profit. Second, the brand keeps most of the gain. If a partner takes a fifth of the Alpha, the brand keeps four-fifths of money it was not making before.

The costs are real, and vendors should state them. Outcome revenue is lumpier than subscription revenue. Media revenue is shared and earns thinner margins. Attention products need real demand from brands and advertisers, not assumed demand. Above all, vendors have to become different businesses: running programmes, measuring incrementality and waiting to be paid until the result is proven. Relabelling old features does not count.

The 10X mindset applies to the vendor too. The 15% move for a martech company is to sell a bit more volume to each customer each year. The 10X move is to be paid for helping the customer pull all four levers.

For the brand, this becomes one question to put to every partner: what are you paid on, and which of my levers does that make you care about?

Aligned incentives make the promise plausible. They do not make it true.

5

The 10X Test

A doubling mindset is not a forecast. What turns it into an operating model is a discipline in four parts: an audit, a short list of moves, experiments that prove them, and a clear statement of what would show the idea to be wrong.

It begins with an audit. The brand and its partner look at all four levers in the brand’s own data, not industry benchmarks. How far is Real Reach below the size of the list? How many of the customers bought through paid media were already customers? Where does attention fade? Where do actions fail? Where does value leak away in discounts and returns? The answers differ from brand to brand. One has a Reach problem: most of its traffic is never identified. Another has an Action problem: interest is high and checkout loses it. A third converts well but at thin margins. The aim is not to double every lever. It is to find the levers where a 10X move exists.

Figure 5. Illustrative headroom assumptions. Uneven across levers, roughly 6X before interactions and testing.

In the illustration in Figure 5, the assumptions imply something like 6X, not 10X, and that is before the levers’ effects on one another are tested. That is a perfectly good answer. A brand that finds even a few-fold opportunity in its existing customers has a very different plan from one that started at last year plus 15%.

Next comes the short list. Pick the two moves with the most headroom and the strongest evidence, and test them over a quarter. Before anything starts, agree the baseline, the eligible customers and a holdout group that keeps receiving today’s experience. A recurring Digest might be tested for its effect on attention and on the purchases that follow. A next-best-action engine might be tested against the brand’s existing journeys. A recovery programme might be tested against both doing nothing and buying the same customers back through paid media.

Then the standard of proof. A message that arrived before a purchase did not necessarily cause it. A customer who returned after an email might have returned anyway. Gains have to be measured against what would have happened without the move, which is what the holdout is for. They also have to be measured in contribution, not gross revenue. Doubling transactions through incentives can look like growth while destroying value.

Finally, the essay owes its readers the conditions under which it is wrong. The four-doublings strategy fails if, across many brands, audits find 10X moves in fewer than two levers. It fails if gains in one lever are routinely cancelled by losses in another: attention falling as reach expands, or conversion rising only as margins fall. And it fails if partners cannot create enough verified gain to justify outcome fees. Any of those findings would not make ambition wrong. It would mean the big opportunity sits in one or two levers rather than four, and the plan should be built around those.

If those conditions do not hold, planning changes. It starts with the four levers instead of the budget. The team asks the doubling question of each lever, lists the 10X moves, ranks them by evidence and funds the ones with the best expected contribution. Marketing becomes organised around the economics of customer relationships rather than around campaigns, channels and tools. Its partners change role too: from delivering activity and reporting engagement to finding opportunities, running the moves and sharing in the gains they can prove.

Return to that planning meeting, a year from now. The spreadsheet opens differently. Instead of last year plus 15%, it opens with four questions. What would it take to double the customers we can reach? To double how often they give us their attention? To double how often that attention becomes action? To double what each action is worth?

The answers will not always add up to 10X. But they will show where growth is hiding, and they will produce moves that a 15% plan would never have considered.

Ten times growth is not a forecast. It is a different way to ask the question. The old plan began with a budget, divided it among channels and hoped the improvements would add up. The new plan begins with customer economics, asks the doubling question of each lever and tests how far the answers compound. For brands, the prize is more value from relationships already earned. For the companies that serve them, it is the chance to be paid for helping create it.

Growth is not a number to target. It is an equation to improve.