Evergreen webinar scheduling: scheduled, just-in-time, or on demand.

You are not picking a setting. You are picking a promise.

Every evergreen timing model is the same recording. What changes is the sentence the viewer hears at the moment they hand over an email address — and that sentence sets an expectation the rest of the funnel either keeps or contradicts.

That framing matters because it moves the question away from the one operators usually ask. “Which model converts best” has no general answer: it depends on your price, your traffic, how well your audience knows you, and how long your argument takes to land. “Which promise can I keep, and which one fits what I am selling” has an answer you can actually reason your way to, and it happens to be the question that protects you later.

The failure that does real damage is not choosing the wrong model. It is choosing one promise and delivering another: live language over a recording, a countdown that is permanently about to reach zero, a chat window with nobody in it. Those are not aggressive marketing decisions. They are claims, and claims get checked.

The test that settles most of this

Before you commit to a model, read your own registration page and ask one question: would this page read the same way if the viewer knew exactly how it worked? If the answer is yes, the model is fine and the rest of this guide is about fit. If the answer is no, you have not found an optimisation — you have found the part of the funnel that will eventually cost you a customer who was paying attention.

The registration becomes an appointment.

A set of recurring start times, the viewer picks one, and the room opens then. It is the closest thing to running the session live, which means it inherits both the behaviour that makes live sessions work and the problem that makes them hard.

The behaviour it inherits is commitment. A chosen time is a small promise to yourself, and reminders have somewhere to point: tomorrow at eleven, in an hour, now. A no-show becomes a real event you can respond to rather than an absence you cannot see. None of that is available in a model where everybody starts whenever they like.

The problem it inherits is the gap. Between registering and starting, interest decays — and unlike a live session, you have chosen to introduce that decay on purpose. Every day of distance between the click and the start is a day for the problem to feel less urgent, for the calendar to fill, and for the person to forget why they cared.

Calendar realism: people pick times they cannot keep

Monday at nine looks completely free when you are looking at it on a Thursday afternoon. This is not a character flaw in your audience; it is how everybody reads their own calendar, and it means the slot somebody chooses is a statement about their intentions rather than their availability. Two things follow. First, pick slots by when your audience actually shows up, not by when you are free — the recording plays either way, which is the entire point of the model. Second, expect the drop and build for it: the reminder sequence is not a courtesy here, it is load-bearing.

Timezone spread: across the day, not across the week

The instinct is to add days. The better move is to add hours. Somebody who registers at ten at night wants tomorrow morning, not Saturday, and a funnel that offers three times all inside one working day in one hemisphere quietly discards half of an international audience. Offer times that cover the day, display all of them in the viewer’s own timezone with the timezone named, and make sure at least one start is reachable within a few hours of arrival at any hour.

There is a limit in the other direction. More slots is not automatically better: if the session implies a room full of people and you have spread the same traffic across twenty starts a week, you have created a contradiction the viewer can see. Match the number of starts to the volume you actually have, then read the number that tells you whether you need more: registration-to- attendance split by how far away the chosen slot was.

Starting in fifteen minutes. Nobody has time to change their mind.

The page offers a start that is minutes away, so the distance between deciding and watching collapses. It is the model with the best case for it and the worst reputation, and both are earned — usually by different operators.

The case for it is not psychological trickery. It is that the strongest single predictor of whether somebody watches is how close the viewing is to the decision. The person clicking your ad is, at that moment, as interested as they will ever be. Every hour you put between that moment and the session is an hour for something else to take the attention — and for a lot of traffic, particularly cold paid traffic, there is no second visit in which to recover it.

What it actually does to show-up behaviour

Registration-to-attendance goes up, and it goes up for a mechanical reason rather than a persuasive one: you removed the interval in which people fail to turn up. That is real, and it is worth having. But read the second-order effect before you celebrate the first: some of those new attendees are people who, given three days, would have decided this was not a priority. They arrive anyway, and a portion of them leave in the first few minutes.

So the attendance rate improves while the composition of the room changes underneath it. That is not an argument against the model — it is an argument against judging the model by the number it is guaranteed to improve. Watch depth, offer-stage behaviour and revenue per registrant are the honest scoreboard here.

Where it strains honesty

The wait is manufactured. The room could open the instant they arrive; you are asking for twelve minutes to give the session a beginning. That is defensible when the wait buys the viewer something real — they arrive at a start rather than mid-sentence, they can get a drink, the session has a shape, somebody is watching the inbox. It stops being defensible the moment the wait is dressed as something it is not.

Three specific lines, and they are the common ones. A counter showing how many people are “watching” when nobody is. A chat window seeded with scripted messages on a timer. Copy that says going live. Each of those converts the mechanic from a scheduling choice into a claim about the world, and the claim is false in a way that a single screenshot proves.

Sometimes the right answer is to just let them watch it.

No wait, no appointment, no start time. The model with the least borrowed energy and the least to explain — and in more situations than most funnel advice admits, the one that beats manufactured urgency outright.

It wins clearly in five cases. When the audience already knows you — your own list, past customers, people who have bought before — because they know it is a recording, and making them wait for a start you control is a small tax on a relationship you spent years building. When the purchase is small enough that the ceremony costs more attention than it buys. When the link will be forwarded to somebody who never opted into the theatre. When the offer’s urgency already comes from somewhere real — a cohort start, a price change, an event — so you do not need to invent any. And when the traffic is mobile, distracted and unlikely to return, in which case the delay is not a filter, it is simply the loss.

What you give up, and what to do about it

Three things. There is no appointment, so there is no reminder chain and no no-show segment — a whole branch of behavioural follow-up simply does not exist. Attention is more fragile, because a video that can be watched at any time is a video that can be paused forever. And attendance stops carrying information: essentially everybody who registers starts watching, so the metric that anchors the other two models goes flat.

The replacements are straightforward. Segment on watch depth and exit point rather than attendance, so the follow-up still knows whether somebody reached the offer. Anchor any deadline to the registration or the first view rather than to a session start that no longer exists — which is exactly the case the deadline guide is built around. And accept that the content now carries the whole thing: with no event framing to lean on, the first three minutes have to be worth the next forty.

Four axes, three models, and no percentages.

There are no numbers in this table on purpose. Every published conversion figure for these models was measured on somebody else’s offer, price and traffic, and borrowing one produces a decision with false precision attached. What travels between businesses is the mechanism.

Read the table for the trade rather than the winner. Scheduled buys attention and pays for it in decay. Just-in-time buys attendance and pays for it in composition and in the honesty it has to be careful with. On demand buys reach and pays for it by removing every structural advantage, leaving the content to do the work alone.

The fifth axis nobody puts in the table

What each model demands of you afterwards. Scheduled sessions need somebody to notice when a slot stops filling and when the reminder sequence quietly breaks. Just-in-time needs the countdown checked on a real device, in a fresh browser, on a schedule — because the failure mode is invisible from the inside. On demand needs the least maintenance and gives the least warning when it decays. All three belong on a recurring maintenance check, because an evergreen funnel repeats whatever you built, including the part that stopped working in March.

Start from what you are selling, not from what converted for somebody else.

These are starting positions, not verdicts. Each one is a reasoned default for a common situation, and each one should lose to your own data the moment you have any.

One pattern runs through the whole table. The colder the traffic and the smaller the second chance, the more the model should compress time. The warmer the relationship and the larger the decision, the more the model should respect the viewer’s own calendar. Manufactured urgency is worth most where there is no relationship to spend, and costs most where there is.

The models are not exclusive, and the good funnels rarely pick one.

Each single model has one bad trade in it. Most of the useful combinations exist to avoid exactly that trade while keeping what the model was good at.

1. 01Imminent start, real times beside it. Offer the next start in a few minutes and two named times later in the week on the same page. Nobody is forced to wait and nobody is forced to hurry; people self-select on how they actually buy. This is the pattern that keeps most of just-in-time’s conversion benefit without any of its honesty problem.
2. 02Scheduled first, on demand after the window. Run scheduled sessions for the first stretch of the funnel, then make the recording available on demand to anyone who never attended. You keep the appointment structure where it earns attention and stop pretending a two-week-old registration is still an event.
3. 03On demand for the warm list, scheduled for cold traffic. One recording, two entry paths, split by source. The people who know you skip the ceremony; the people who do not get the structure that makes a stranger stay. This one costs you a second registration page and nothing else.
4. 04Scheduled with a same-day slot, always. Whatever else you run, make sure a viewer arriving at any hour can find a start within the same day in their own timezone. Most of the loss attributed to “scheduling” is really the loss from a three-day gap, and this removes it without changing the model.

The combination to avoid

Running all three with different promises on the same page. A viewer who sees “starting in twelve minutes”, a list of scheduled times and a watch-now button has been given three contradictory statements about what this is, and the honest conclusion — that the start time is arbitrary — is the one they will reach. Pick a default, offer one alternative, and let the page make sense.

Whatever you combine, keep one recording and one follow-up spine branched on behaviour rather than four parallel funnels that drift apart. The evergreen email templates are written for exactly that shape, and the evergreen funnel playbook covers where the branches go.

Draw the line once, in writing, before the numbers tempt you.

Nobody sets out to build a dishonest funnel. It happens one setting at a time, each one slightly more effective than the last, until a customer takes a screenshot. Decide where the line is while it is still an abstract question.