How to set up an automated webinar: seven phases, in build order.

Never automate a session that has not sold anything.

Automation is a multiplier, and it multiplies whatever you point it at. A session that converts becomes a machine. A session that does not becomes the same failure repeated several hundred times, quietly, in front of traffic you paid for.

The temptation is understandable: recording once and selling forever is the entire appeal, and the fastest route there looks like recording immediately. But an evergreen funnel removes the thing that makes a weak live session survivable — you standing there, reading the room, and fixing it mid-flight. Nothing gets fixed mid-flight in a recording.

Three tests, not a run count

“Three live runs” is the advice you will read everywhere and it is the wrong shape. What you are looking for is evidence, and evidence arrives on its own schedule. These are the three signals worth waiting for.

1. 01The stay test. People are still there when the offer starts. If the room empties before minute forty live, it will empty earlier on a recording, where leaving costs nothing socially.
2. 02The purchase test. Somebody bought, at the price you intend to charge evergreen. A session that converts at a launch price and has never been tested at the standing price has not been tested.
3. 03The question test. The questions have stopped surprising you. When you can predict the three objections before anyone types them, you can answer them inside the presentation — which is the only place an evergreen viewer can hear an answer.

Running those sessions is a separate job with its own clock. If you have not run one yet, start with the live webinar guide and come back here when you have a run that cleared all three tests. If you already have several and you are choosing between them, the live-to-evergreen conversion guide handles that decision in detail.

Take out everything that only worked because people were in the room.

A live recording is not an evergreen asset yet. It contains a room, a date and a set of mechanics that assume both — and each of those quietly tells the viewer that they are watching something that already happened to somebody else.

Three kinds of thing come out, in ascending order of difficulty. Dead air is the easy pass: waiting for latecomers, the promise to give it another two minutes, the pause while poll results come in, the tech hiccup at minute nine, the silence before the first question arrives. None of it is content and all of it is expensive, because a viewer alone at a screen leaves during silence.

Live-only mechanics are harder. Anything that asks the room to do something — type in the chat, raise a hand, answer a poll — reads as a lie when the room is one person. So does naming attendees, referencing the number of people present, or promising to get to questions at the end. Some of these cut cleanly; some are wrapped around a sentence that carries the argument, and those need a re-record rather than a snip.

Anything that dates it is hardest, because it hides. A season, a specific date, a reference to a recent announcement, a price you will change, a tool that may not exist next year, “as we head into the back half of the year”. A funnel that ages is not a maintenance problem, it is a credibility problem: the viewer works out they are watching something old, and everything else you said gets older with it.

Re-record the frame, keep the teaching

The default is to keep. The teaching is what converted, and re-performing it alone in a quiet room usually makes it worse — you already know where it goes, so you rush the parts that earned their pauses. What does get re-recorded is the frame around it: the open, any transition that spoke to the room, the offer section, and a short FAQ block covering the questions every live run produced.

The open in particular has a job the live version did not: say what this is, say how long it takes and what they will be able to do afterward, and say when the offer arrives. That last line costs nothing and buys back all the attention people otherwise spend wondering whether they are about to be sold to.

Which run to use, what comes out in which order, and how to match a re-recorded segment to the original are the whole subject of the live-to-evergreen guide. This phase is done when you have one versioned final cut and the timestamp the offer begins at, written down where the rest of the build can read it.

Scheduled, just-in-time or on-demand — pick the trade, not the default.

This is the decision that shapes every number after it, and it is usually made by accident — whichever option was selected by default the afternoon the funnel got built. Each model buys one thing and sells another.

Match the model to the traffic

Cold traffic that arrived from an ad and has no relationship with you tends to suit just-in-time: the attention you have is the attention you get, and asking it to come back on Thursday is asking for something it will not do. A warm list that already opens your email suits scheduled sessions, because those people will hold a slot and the reminder sequence has something real to remind them about. Existing customers and long-consideration offers suit on-demand, where the viewer is deciding rather than discovering and being made to wait reads as theatre.

Run one model deliberately before you run several. Offering all three at once is the most common way to end up unable to say which one produced the result — and the numbers in Phase 07 stop being readable the moment the funnel has three different clocks in it.

Two smaller decisions belong to this phase because everything downstream assumes them: whether a replay exists at all and for how long, and how far ahead the next available session is allowed to be. A scheduled funnel whose next session is five days out is a registration page with a waiting list attached — most of those registrants are gone before it starts. The scheduling guide covers slot density, replay windows and the hybrid patterns in full.

Same shape as a live sequence. Completely different clock.

A live campaign runs on one clock that everybody shares: the session date. An evergreen campaign runs on a clock per person, starting the moment they register — which is why a live sequence copied across verbatim produces reminders for a session that already happened to someone else.

Six segments, split on behaviour

The sequence before the session is straightforward — confirmation, a reminder or two if the model has a gap to fill, and the join link. Everything interesting happens afterward, and it depends entirely on what the viewer did.

• →Registered, never started. The evergreen no-show, and usually the largest group. They have not seen anything yet, so the sequence is another chance to attend rather than a follow-up. Different message, different clock, and never the buyer path.
• →Started, left early. Something was wrong in the first stretch — the fit, the pace, the promise. Short, low pressure, and worth one attempt at the specific idea they left before reaching.
• →Watched, left before the offer. They liked the teaching and never saw the thing you were selling. Send the offer, not the whole recording again. This is the segment most funnels waste, because on aggregate reporting it looks identical to the group above.
• →Reached the offer, did not click. The warmest non-buyer you have. They heard the price and the terms and decided against them, which means the objection is nameable — and you already know what it is, because the live runs told you.
• →Clicked, did not purchase. A checkout problem, a terms problem or a timing problem, and the three are distinguishable. Restate the deadline, make replying to a human easy, and check the checkout yourself before assuming it was the price.
• →Purchased. Out of the sales sequence the moment the payment lands, into onboarding. Nothing reads worse than a deadline reminder arriving after someone paid.

Those six are not a preference. They are the minimum set that produces messages a person can read without noticing they were written for somebody else, and the difference between them is worth more than any subject-line work you will ever do on a single blast.

What has to be true for any of it to work

Every one of those splits depends on a watch-time signal — started, reached minute twenty, reached the offer mark, clicked — arriving somewhere the email system can act on, attached to the same record as the registration and the eventual purchase. If the watch data lives in a video dashboard and the emails live somewhere else, you do not have six segments. You have one sequence and a report you cannot send from.

The sequences themselves, written out for each branch, are in the evergreen email templates. The strategy behind branching this way — and why aggregate funnel reporting hides the segment worth the most — is in the evergreen funnel playbook.

A deadline that never arrives teaches people to ignore all of them.

In a live session the deadline is a shared moment: everyone heard it at the same time and it passed for all of them at once. Evergreen has no shared moment, so a fixed date either expires and kills the funnel, or gets quietly extended forever and stops meaning anything.

The answer is a window per registrant. It is a genuine deadline — it starts, it runs, it closes, and it closes differently for each person because each person started at a different time. Four decisions make it real.

1. 01Pick the anchor. Registration time or viewing time. Viewing time is fairer and matches what the viewer experienced; registration time is simpler and keeps a clock running for people who never watched. Pick one, and say which one in the emails.
2. 02Pick a length you can hold. Short enough to create a decision, long enough to survive a weekend. Purchase lag from Phase 07 is the honest input here — if almost everything lands in the first day, a longer window is buying you nothing and costing you urgency.
3. 03Make expiry real. The checkout has to stop honouring the price, not merely the timer. An expired countdown above a working purchase link is the version people screenshot, and it is worse than having no deadline at all.
4. 04Decide what happens after. A real page saying the offer has closed and what is still available, not a broken link. Some of those people come back at full price later, and they will not if the last thing you showed them was an error.

Then the offer section itself has to work with nobody reacting to it. Every objection the live room raised out loud still arrives — you simply cannot hear it — so it has to be answered inside the recording, in order. Say the price out loud rather than only putting it on a slide, because some of these viewers are on a phone. And put the offer somewhere persistent beside or beneath the video: in a live room you paste the link into chat while you talk, and evergreen has no chat to paste it into.

The full mechanics — anchor choices, timer behaviour, what to do about people who register twice, and how to keep scarcity claims defensible — are in the evergreen deadlines guide.

Register as a stranger and go all the way through.

Everything before this is inspection. This is the only test that finds the failures that matter, because the failures that matter live in the joins between systems — and no individual system reports them.

Run the funnel three times with three behaviours: register and never start, register and leave a few minutes in, and register and watch through to a completed purchase. Use a fresh email address for each one. Testing on your own record is the classic way to prove that a funnel works for exactly one person who is already tagged as a customer.

Do it in a private window, on a phone at least once, and in a timezone that is not yours if you can borrow a colleague for ten minutes. Timezone handling is the single most common evergreen defect and it is invisible from the seat of the person who built it.

Seven numbers, and an interval that keeps them true.

Published webinar benchmarks average across industries, price points and traffic sources that have nothing to do with yours. Definitions travel between businesses. Numbers do not. Define these precisely and compare them only to your own previous month.

Read them as a sequence rather than a scorecard. Each one hands the problem to the next: registration rate judges the page, start rate judges the schedule model, watch-through judges the recording, click rate judges the offer, and everything after that judges the checkout and the terms.

Then maintain it on a schedule

Evergreen funnels do not announce their failures. A link rots, a price changes, an integration stops passing watch data, a reference ages badly — and the funnel keeps running, producing a slightly worse result every week, which is indistinguishable from a slow month until somebody looks.

A monthly read of these seven numbers and a quarterly pass through the content and the links is a reasonable working cadence. Put both in a calendar rather than in your intentions; the maintenance checklist is the version you can run in half an hour without deciding what to look at first.