River
Y CombinatorBacked by Y Combinator
FREE TEMPLATE

Welcome Email Sequence Template

Four documents and four sheets that time the sequence to the day your own subscribers order, rather than to a generic five-day cadence.

Free download  ·  No account needed

Conversion Timing Curve  ·  by acquisition source

Two peaks, seventeen days apart

Tallow & Ash  ·  12,000 subscribers, one month, tracked 90 days to a first order

Day bandPopup 7,320Guide 3,480Blended 12,000
0 to 13.361%0.805%2.400%
2 to 31.301%0.637%1.033%
4 to 61.017%0.525%0.828%
7 to 100.868%0.410%0.687%
11 to 140.701%0.471%0.596%
15 to 210.647%5.588%2.150%
22 to 340.636%2.881%1.288%

The sequence sends on days 0, 1, 2, 4 and 6. Shaded rows are the days it says nothing.

What the blended column cannot say

Blended, day 15 to 21 reads 2.150% against 2.400% on day 0 to 1, so the second peak looks like a bump on one decaying curve. It is a second population. 5.588% is the highest hazard in the table, it belongs to 29.0% of the list, and the sequence has been silent for nine days by the time it arrives.

Hazard, and why it is not a conversion rate

First orders in the band, divided by the subscribers who still have no first order at the band's start. The denominator shrinks as people convert, so the number can rise. A cumulative conversion curve only rises, which is why it can never show a dead zone.

What the quiet cohort is worth

The guide cohort converts at 12.90% over 90 days against the popup cohort's 9.34%, and at $78 rather than the $66 the day-4 discount pair sells at. It is the better half of the intake, and not one message is timed for it.

Every hazard from day 7 onward was measured with nothing in market, so 5.588% is an unassisted rate.

Tallow & Ash sends five welcome emails over seven days. Nobody there chose that schedule: it arrived with the sending platform's starter flow in 2021, and the copy has been rewritten twice since without the timing being touched. It is also the cadence in every published welcome-series template, which is a fact about the templates rather than a finding about anybody's list. Run the numbers on one real month of signups and the sequence is timed for the 61.0% who arrived through a discount popup, and silent on the days the other 29.0% buy.

A cumulative conversion curve only rises, so it cannot show you a dead zone. A hazard can. For each day-band it asks what share of the subscribers who still have no first order place one in that band. The blended answer here runs 2.400% on day 0 to 1, down to a 0.596% trough on day 11 to 14, then back to 2.150% on day 15 to 21. Split by acquisition source, that last figure is not a bump on one curve at all.

Popup signups peak at 3.361% and decay from there. Care-guide signups open at 0.805% and reach 5.588% on day 15 to 21, the highest hazard in the table, nine days after the sequence stops. So this pack ships two sequences rather than one longer one, alongside the lifecycle email programme pack that decides what one subscriber may receive in a week and the customer segmentation template. Send River two exports, or take the files blank from the template library.

The curve, the schedule read off it, and the message that gets deleted

The Conversion Timing Curve, the Sequence Map it implies, and Performance by Step.

Conversion Timing Curve

Cohort defined on signup date. First orders only, so a customer with four orders contributes one event. The denominator is re-derived every band.

Day bandDaysPopup ordersPopup hazardGuide ordersGuide hazardBlended hazardCumulative
0 to 122463.361%280.805%2.400%24.00%
2 to 32921.301%220.637%1.033%34.08%
4 to 63711.017%180.525%0.828%42.08%
7 to 104600.868%140.410%0.687%48.67%
11 to 144480.701%160.471%0.596%54.33%
15 to 217440.647%1895.588%2.150%74.67%
22 to 3413430.636%922.881%1.288%86.58%
35 to 5925470.700%421.354%0.867%94.50%
60 to 8930330.495%280.915%0.607%100.00%
90 days6849.34%44912.90%10.00%1,200

The two cohorts do not share a peak, or anything close to one. Popup opens at 3.361% and never returns above 1.302%. Guide opens at less than a quarter of that and hits 5.588% seventeen days later, which is 6.95 times its own opening hazard. A single sequence fitted to the average of the two is timed for neither of them.

Guide is the smaller and better half. 3,480 subscribers against 7,320, converting at 12.90% against 9.34%, and it takes only 12.5% of the orders the day-4 discount produces. It is not price-motivated and it is being discounted nine days early anyway.

A third cohort of 1,200 (checkout opt-in, in-store, referral) produced 67 first orders and is held aside rather than scheduled against, because a cohort that size gives a hazard table that is mostly noise. 684 plus 449 plus 67 is the 1,200 total.

Sequence Map

Nine authored messages across two sequences. Any one subscriber receives four or five. Audience is the cohort's still-unconverted population at the start of the band.

SequenceMsgDayHazardAudienceSubject
Popup and other103.361%8,520Your skillet questions, answered in order
Popup and other213.361%8,520How a pan gets its surface
Popup and other321.301%8,260The three pans people actually start with
Popup and other451.017%8,16110% off your first order, no expiry theatre
Guide100.805%3,480The care guide, and where people go wrong with it
Guide220.637%3,452Seasoning, in four paragraphs and no myths
Guide360.525%3,430Which size, for how many people
Guide4130.471%3,398What a cast-iron pan costs over ten years
Guide5175.588%3,382Ready when you are, and here is what ships today
Total950,603against 58,894 today

The first three guide messages sell nothing, because at a hazard under 0.9% they could not. Their job is to earn the next open, and that is measurable: a subscriber who opens nothing in fourteen days converts at 0.626% over the following ten weeks, against 7.731% for one who has opened once.

Day 13 is placed before the peak on purpose, not inside it. A message arriving into a peak reaches somebody who has not heard from you in a fortnight. One four days earlier means the day-17 message lands on a reader who was active this week.

Sends fall 14.08% and the sequence is live in bands holding 59.17% of first orders against 42.08% today. That is a placement measure, not a revenue projection: it says a message was in market when the order happened, and the holdout is what would say it caused anything.

Performance by Step

The sequence currently in production, on the same cohort. Orders are last click before the order, within 72 hours.

MsgDaySubjectSendsOpenOrdersOrd / 1kRev / sendUnsubUnsub / order
10Welcome, and the one thing worth knowing12,00061.80%18715.58$1.2160.092%0.06
21How a skillet is actually made12,00044.20%968.00$0.6240.142%0.18
32What most people start with11,71233.50%716.06$0.4730.213%0.35
44Here is 10% off your first order11,59130.10%625.35$0.3530.336%0.63
56Your code expires tonight11,59124.80%342.93$0.1940.518%1.76
Total58,8944507.64$0.5760.258%0.34

Message 5 is 39.5% of the sequence's unsubscribes and 7.6% of its orders. Its 2.93 orders per 1,000 sends is 18.8% of message 1's, and it costs 1.76 unsubscribes for every order it produces. It is deleted in the retimed design rather than moved, because nothing in the curve justifies a slot for it on any day.

Read down the orders-per-thousand column and this looks like an ordinary decay. Each message earns less than the one before, which is what everybody expects and the reason nobody looks further. The problem is not in this table. It is that every one of these five days belongs to one of the two cohorts.

505 first orders actually occurred in days 0 to 6. 450 are attributed to a message. The 55-order gap had no click inside the window and is left unattributed rather than assigned. No holdout was run on any of the five, so none of these figures is an increment.

What's in the pack

01

Conversion Timing Curve

Nine day-bands with the at-risk population, first orders and hazard for each source, plus the blended rows underneath to read against.

02

Engagement and Stop Rule

Where each sequence ends, split on fourteen-day opens and split again by source, because one threshold gets one cohort wrong.

03

Sequence Map

Nine messages with the send day, the hazard of the band it sits in, the audience size and one sentence justifying the date.

04

Performance by Step

What the sequence in production actually earns per message, including the unsubscribes-per-order column that condemns one of them.

05

Timing Method

How the hazard is computed, the four caveats that belong on the sheet, and the four ways the table comes out wrong.

06

Sequence Brief

The current sequence, the two that replace it, and what the change is worth stated as sends and placement rather than projected revenue.

07

Drafted Emails

Nine drafts with subject, preheader, body direction and a must-not line, which is the field that blocks the discount reflex.

08

Testing Plan

Four tests in order, each with the events per arm per month worked out first, so an unresolvable test never gets queued. The email A/B testing plan template does that sizing for a whole programme.

How to use it

  1. 1

    Open in River, or take it blank

    Let River read your exports and do the arithmetic, or take the four documents and four sheets away and compute the curve by hand.

  2. 2

    Send two exports

    Subscribers with a signup date and an acquisition source, and orders with a customer id and an order date. Four columns is enough to compute the curve.

  3. 3

    Read the curve before the schedule

    You get the hazard table, then the same table split by source, and a straight answer on whether your curve has one peak or two.

  4. 4

    Take the schedule off it

    One sequence per cohort with a peak nobody else shares, each message carrying the hazard that justifies its day, and the stop rule derived per source.

Frequently asked questions

Is this template free?

Yes. Download the Word documents and CSV sheets with no account, no card and no email gate. Edit with AI is the optional half: River computes the curve from your exports, splits it by source and reads the schedule off it. Every pack sits in the template library.

How many emails should a welcome sequence have?

As many as your curve has high-hazard bands, which is the only honest answer. The worked example here needs nine authored messages across two sequences, and any one subscriber receives four or five. A brand with a single early peak needs three or four and nothing later.

What is a hazard, and why not just use conversion rate?

A hazard is first orders in a band divided by the subscribers who still have no first order at that band's start. Because the denominator shrinks, the number can fall and rise. Cumulative conversion only rises, so it can never show you a window where nothing is happening.

Can I not just extend my sequence to three weeks?

Sometimes, and check the split first. A blended curve with a second bump is usually two cohorts with one peak each, and extending one sequence sends the late messages to everybody, including the cohort whose hazard has been under 0.7% for a fortnight.

Should I stop mailing subscribers who never open?

Derive it rather than inherit it. Here, fourteen-day non-openers convert at 0.626% over ten weeks against 7.731% for openers, but that splits to 0.315% and 2.276% by source. Complaint volume also has a ceiling: Google asks senders to keep the Postmaster Tools spam rate below 0.10%.

Does a longer sequence change anything about compliance?

One thing that is easy to miss. The FTC's guidance requires an opt-out mechanism to work for at least 30 days after the message that carried it, and requests honoured within ten business days. Stretch a sequence to day 17 and a day-0 unsubscribe link is suddenly load-bearing.

What format are the downloaded files?

Four .docx documents and four .csv sheets, in a zip. They open in Word, Pages, Google Docs, Excel, Numbers and Sheets with no conversion. Add ?format=pdf to the download link if you want the documents as PDFs to circulate instead.

Find out when your subscribers actually buy

Take the Word documents and CSV sheets blank, or open this exact pack in River and send it a subscriber export and an order export.

Edit with AI