Subscription

When software shipped on discs, a sale was a single event. The customer paid five hundred dollars, carried away an installer, and the seller booked that money as revenue for the month and went looking for the next customer. Once the sale closed, so did the relationship. Whether that customer opened the product daily or opened it once and deleted it was unknowable from the seller's side, and there was nothing more to collect from them until the next version shipped.

In that structure revenue restarts at zero every quarter. Quarterly reviews kept stalling in the same place. Two hundred thousand dollars sold last quarter, and the first day of this quarter still reads zero; nothing done well last quarter fills any part of this one. The reward for building something good came from finding new buyers rather than from the people already using it, which quietly pushed effort toward the front of the funnel and away from everything behind it.

The customer side wasn't comfortable either. Five hundred dollars is a lot to commit to something you haven't used, so decisions got postponed or pushed up several levels of approval. Buy it, find it doesn't fit, and the money is simply gone. Sellers spent their energy convincing people not to buy, and buyers carried the risk of buying wrong, and that arrangement held for a long time because neither side had a smaller unit to trade in.

Moving products onto the web made another arrangement possible. The service now runs on a server the seller controls, so opening and closing access became something the seller does directly, and that opened the path to collecting small amounts continuously instead of one large amount once. Charge ninety-nine dollars on the first of each month and keep the service open for that month. The customer risks ninety-nine dollars to start, and the seller, knowing most of this month's customers will still be here next month, can calculate next month's number today. Once that calculation works, running the company changes. If a hundred thousand dollars came in this month and two percent leaves monthly, six months out can be sketched now, and hiring decisions get made on top of that sketch. In the sell-once structure the calculation didn't exist to be made, because nothing carried across the boundary between one quarter and the next.

One condition comes attached. Money arriving continuously means delivery continues too. The servers stay on, outages get fixed, and when a competitor ships something better the product has to close the gap. In a one-time sale, quality after the sale didn't connect to next quarter's revenue; in a subscription, a customer disappointed this month drops straight off next month's invoice. Revenue continues, and so does responsibility.

The terms under this heading fall into two groups. Cycles decide the interval the charge is cut at, and operations are the machinery that keeps that interval turning without hands on it. Choosing a cycle happens once, while pricing is designed; the operational pieces run every month for as long as a customer stays.

Cycles decide how often an invoice gets cut, and there are two.

1. Monthly plan — charged in one-month units. The customer can stop next month if it isn't working, so the amount at risk up front is as small as it gets and the barrier to starting is low. From the selling side it's a model that has to win the decision again every month. Twelve months carry twelve chances to leave, so a customer who sees no value by month four leaves in month four. That's why teams on monthly billing staff the first three months heavily: what customers do afterward splits sharply on whether they cleared that stretch.
2. Annual plan — a year collected at once. Roughly two months usually come off the price, and in exchange a year of cash arrives now. A customer who cools off mid-year has their next charge eleven months away, so they don't leave during it. Holding a year of cash means next year's engineers can be hired this year. What arrives doesn't immediately become revenue, though. The eleven months not yet delivered sit as deferred revenue and move across one month at a time as the months pass. Deferred revenue is money collected for service not yet provided, which accounting cannot recognize as revenue yet.

Operations are the pieces that keep the cycle repeating without manual work, and there are three.

3. Auto renewal — the setting where the next cycle starts on its own when the current one ends and the card on file gets charged. Without it, customers repeat the same payment by hand every month, and one forgotten month cuts off the service. Convenience of that size brings disputes of matching size, so advance notice a few days before renewal and a switch that can be turned off at any time come as standard. Put the setting on an annual plan and the dispute gets twelve times larger. A month slipping through is tolerable; a year leaving the account unannounced turns into refund demands and card chargebacks, and handling one of those can cost more than the customer paid.
4. Cancellation flow — the process of ending a subscription together with whatever sits in that process to talk people out of it. The screen asking why, the option to pause for a month, and the offer of a cheaper tier all live here, and so does the question of how many of them a person has to pass before the cancel button works. What comes back from that screen is raw material for reducing churn, churn being the term for paying customers leaving. Push the retention devices too hard and hide the cancel button, and the customer who wanted out gets out anyway and leaves a bad review on the way, and in some countries the practice is outlawed. So the screen gets designed as a retention device and a data intake at once. Let people pick a reason, collect the answers monthly, and what the product is failing at accumulates in its most honest form.
5. Grace period — the stretch of days where the service stays open even after a payment fails. Most failures are a card briefly over its limit or a reissued card with a new number, so cutting access on the spot loses customers who had no intention of leaving. Seven to fourteen days is typical, with a few retry attempts and notification emails across it. That retry procedure has its own name, dunning. How long to set the window is a profit-and-loss calculation. Too short and recoverable customers get lost; too long and server costs keep running on accounts that aren't paying. The usual balance comes from narrowing what stays available during the window: files stay readable, new work gets blocked, and the account holder sees a notice on every screen until the card goes through.

The two groups are attached. How the cycle is set decides what matters operationally. Bill monthly and there are twelve charges, so twelve chances to fail, and the grace period and retries become what protects revenue. Bill annually and there's one charge, but a year rides on it, so notice at renewal time matters far more. It's why a team that starts with a single monthly plan and later adds an annual one ends up rewriting its renewal notices at that moment and not before.

"Should we move this customer to annual and give them two months off?"

Deciding how much discount to trade for a year of cash from someone currently paying monthly.

"Forty failed charges last month and the grace period saved thirty of them."

Retries and notices held on to most of the customers who were about to be cut off over a card problem.

"That's one-time revenue, not subscription, so it stays out of recurring."

Money that arrives once and doesn't repeat gets excluded when counting what will come in again next month.

The number Grace put up at Pagedeck's pricing review was the monthly recurring revenue graph, the total of only the amounts that repeat each month. Three months nearly flat, while new teams kept arriving steadily. Arrivals and departures were cancelling out, which is a shape that never appears on a signup chart on its own.

Sofia named one cause. A large share of accounts that started monthly quit around month four, and that timing overlaps with the person who championed the rollout getting pulled onto other work and drifting away from the product. Mia brought the answers left on the cancellation screen. "Stopped using it" is more than half; "too expensive" is under one in ten. The problem is failing to take hold, not price, and the two call for entirely different fixes.

Two decisions came out of the meeting. One was a month-three email summarizing how the account had been using the product. The other was widening the annual discount from two months to three, so more teams get carried past the first four months. What Theo flagged was that changing the discount also changes how mid-cycle plan switches get calculated. Recalculating by the number of days like that is proration.

Applying the decisions took a month. Changing a discount means also deciding when existing customers start paying the new amount and how long they stay on the old one. Price changes in a subscription don't finish the moment they're decided, because the next invoice of every current customer hangs on them.

Reading a subscription as money that arrives by itself once it's set up is the most common mistake. The charge happens automatically and the customer does nothing, so it genuinely looks that way. What's automatic is the payment date; the relationship gets re-evaluated every month. Every billing notice prompts a customer to think once about whether to keep paying, and in a month where the product barely got opened, that thought turns into a cancellation. Miss the structure and everyone gets assigned to bringing in new customers while nobody owns helping existing ones take hold. Six months of that produces a graph where signups climb monthly and total revenue stays flat. The two lines rise together for the first few months and then split, and the point where they split is where the rate of leaving matched the rate of arriving.

Reading the amount charged as that month's revenue goes wrong just as often. Twelve thousand dollars landed in the account, so it feels like twelve thousand dollars earned this month. If it's payment for an annual plan, a thousand of it is this month's revenue and the remaining eleven thousand is service not yet provided, carried as a liability. Mix the two and months heavy with annual renewals look like revenue spiking and then collapsing, and hiring or raising ad spend on the strength of that graph produces an empty bank account a few months later. So teams running subscriptions keep three separate numbers: cash received, revenue recognized this month, and the amount that will repeat next month.