4. Building Sequences

Senders, cohorts, copy, and the four gates between a draft and a sent email.

What a sequence is

A sequence enrolls a group of people in a series of timed, personalized touches, and stops the moment someone replies.

A marketing blast sends everyone the same thing. A sequence reaches each person with something relevant to them and follows up until they respond. If you only need to send one email, you do not need a sequence. A touch can be an email or a LinkedIn action, and you can mix both in one sequence.

Three touches is a safe default. You can extend to four or five. Past five, returns drop and complaints rise. Use widening gaps: two to three days, then four to five, then five to seven. Even spacing reads as automated to both humans and mail providers.

Before you write anything

Three setup items. Skipping them is the most common reason a sequence will not publish.

Pick how you are sending email

Connected account. Your own work email. Right for warm plays and lower volume, because the message comes from the real you. Check the Connected Account toggle in your Sequence settings.

Warmed sending accounts. Dedicated accounts on separate domains, for cold volume. Upload them also in your Sequence settings

If you are running anything in Group 3 of the Play Library, read the Sending Safely article now rather than later. A new domain cannot carry cold volume for at least its first two weeks.

Connect LinkedIn, if your sequence uses LinkedIn

LinkedIn sequences are supported. Connection requests and direct messages both work as steps, and you can mix them with email in one sequence.

Go to Settings > Sequences > your name and connect LinkedIn at the bottom. Do this before you build. A sequence containing LinkedIn steps will not publish until LinkedIn is connected, and the message you get at that point does not always make the cause obvious. If a sequence refuses to publish and you cannot see why, check this first.

Fix your signature

Plain text. No links, no logos, no images, no banner. This is deliberate. Graphics and links in a cold email signature read as marketing automation to spam filters and to humans, and they cost you delivery. Sequence emails are plain text throughout for the same reason.

Decide who is sending, first

Before any copy exists, carve up your contacts by seniority and assign senders. A CEO writing to a CEO is a different email from a VP of Sales writing to a CEO. The same words in the wrong mouth read as a mismatch, and no amount of copy craft fixes it.

Decide explicitly:

• Who sends to founders and C-level

• Who sends to VPs and directors

• Who sends to practitioners

• Who owns LinkedIn

Every sender has to exist in Lightfield first. If you are assigning a teammate, they have to accept their invite and log in before you build. Until they do, the system cannot author or send from that name, and a sequence that depends on them produces nothing rather than telling you why. This is a common reason a launch stalls.

On LinkedIn, notoriety matters more than title

If one person on your team has an audience and a track record people recognize, their connection requests get accepted and their DMs get read at a completely different rate than anyone else's. Route LinkedIn through whoever has the reach, even when the email lane belongs to someone else. Acceptance rate is not just a vanity number here: below roughly 30%, LinkedIn quietly tightens your weekly limit, so the person with the best acceptance rate is also protecting a scarce resource.

Build cohorts before you build copy

Type create sequence in the Sequences panel or in any chat, and tell the agent which list you are working against.

Say whether it is an account list or a contact list. The agent can read one as the other and take a wrong turn, and one word up front saves the correction.

Then have it recommend cohorts. This is the highest-leverage move in the whole process.

Example prompt: Look at the contacts on [list name], recommend cohorts based on meaningful segmentation by persona, and write a personalized sequence for each so the right message goes to the right people.

You can cohort by persona, by situation (recently funded, just launched, hiring for a specific role), or by warmth. The test for whether something deserves its own cohort: if the value prop changes, or the signal changes, it is a different cohort. If both are the same, it does not need one.

Variants, where the ask is genuinely uncertain

Within a cohort you can run two or three variants against the same persona. The most useful axis is usually the CTA: a value-led ask ("want the two changes that did most of the work?") against a time-led ask ("worth 15 minutes?"). Run both and see which pulls. You can also vary by value prop or by signal. Do not vary for the sake of it. Two variants you can read a result from beats five you cannot.

Frameworks

The agent works generatively by default, reasoning from the knowledge base you built in the Setup: Your Knowledge Base article. You can also name a template.

None of these is a default. The right one should feel chosen for the specific person.

The rules that decide whether copy works

Most of this is already built in. Lightfield has the craft encoded: length limits, opening on a real signal, one idea per email, one ask, and a list of the phrases that make an email read as automated. You will get reasonable copy without knowing any of it. The reason to read this section anyway is that knowing what the rules are is how you steer, and how you tell whether a draft is good. When a preview reads badly, it is almost always breaking one of these, and naming which one is faster than asking for something vaguely better.

One idea per email

The single biggest lever. If a sentence introduces a second concept, cut it. More than one idea is the most common reason a sequence underperforms.

Have a trigger

A specific, observable reason you are reaching out right now, traceable to a real CRM field or a real search result. Build this into the sequence instructions rather than researching contacts by hand: tell the agent to find a real signal per contact, and give it a specified fallback line for when it cannot find one. Then every contact gets a real reason, not only the ones you personally got to. Cold outreach without a trigger is a much harder sell, and no amount of copy craft fixes it.

Relevance beats personalization

Their job title and a compliment on their post are decoration. Something true about their world right now, tied to a consequence they already feel, is a reason to reply.

Earn the reply, do not ask for the meeting

A meeting is the highest-friction thing you can request on a first touch. Ask a question they can answer in one line, or offer something they get by replying: a benchmark, a teardown, a comparable example. One ask per email. Never two.

Under 80 words

Most email is read on a phone. If it needs scrolling, it needs cutting. High-performing cold emails run 50 to 125 words, and the shorter end of that is usually right.

Every follow-up changes the angle

New proof point, different pain, a customer example, a real question. "Just circling back" trains people to ignore you. A short break-up note as the last touch often pulls more replies than the ones before it. And two to three follow-ups can generate a large share of all your replies, so the sequence that stops at one email leaves most of its pipeline on the table.

Ban the tells by name

The agent calibrates far better against concrete bans than against "make it sound good." Hand it this list:

• Throat-clearing openers: "I hope this email finds you well," "I wanted to reach out," "My name is"

• Filler adjectives: innovative, cutting-edge, seamless, game-changing, world-class

• Fake warmth: "Love what you're doing," when it is generic enough to send to anyone

• More than one CTA, and any request for "just 30 minutes" up front

• Em dashes

You set the voice, Lightfield handles the rest

The mechanics above are handled for you. Voice is the part that needs you. Drop in one or two of your own good emails as a reference and it will write in your register instead of a neutral competent one. This takes two minutes and it is the single highest-return thing you can do to the copy.

Calibration: the same prospect, two ways

Before. Reads like a template, asks for too much, says nothing specific:

Subject: Quick question

Hi Jordan, I hope this email finds you well! I came across your company and was really impressed by the innovative work your team is doing. We've built a cutting-edge platform that helps companies like yours streamline operations and unlock real value. I'd love to grab 30 minutes to walk you through it. Does Tuesday or Thursday work better?

After. Signal-led, one specific proof, one low-friction ask, about 60 words:

Subject: 3 SDRs

Jordan - saw you're hiring three SDRs this quarter. That usually means the number went up and your reps are already stretched. We helped [Similar Co] add 70 booked meetings a quarter without new headcount by fixing how leads get routed. Want the two changes that did most of the work? I can just put them in a reply.

It opens on a real signal, names a consequence they already feel, backs it with a number instead of adjectives, and asks for a reply rather than a meeting. That is the bar for every cohort.

Subject lines: two strategies that work

Internal camouflage. Two or three lowercase or plain words that look like a coworker sent them.

Offer-led punch. The offer is the subject. Only works if the offer is genuinely striking.

Never: feature lists, company names, "Quick question," or anything that smells like a sequence.

LinkedIn copy is different

Do not paste a cold email into a DM.

• The connection request is not the place to pitch. One line or none at all. It caps around 300 characters, so it cannot carry a pitch anyway. Earn the connection, then start the conversation.

• Once they accept, the DM is a conversation. Drop the subject-line thinking, the signature and the formal opener. Two or three sentences.

• No links in a first message. They suppress reach and read as spam.

Running both channels together

Do not say the same thing twice in two places. A pattern that works: a light connection request with no pitch, a short DM once they accept giving the real reason, then an email carrying the fuller context alongside it. Showing up thoughtfully in two places reads as a real person paying attention. One more identical email reads as a machine. Hold one CTA across the whole sequence rather than one per channel. You are running one conversation on two channels.

Where to spend LinkedIn

Your weekly invitation budget is the scarcest resource in your outbound stack, and it does not scale with money or headcount. Roughly 100 requests per rolling seven days, the same on every paid tier. Lightfield defaults you to 10 a day and caps you at 20, which is that ceiling divided across a working week. So do not spend invitations on warm contacts. Someone who already knows you will open an email from your real inbox. Save LinkedIn for people you cannot otherwise reach: net-new prospects with no verifiable email, champions whose new address you do not have, genuinely unreachable inboxes. Details in the Sending Safely article.

If you are already connected to them, skip the invite

You cannot send a connection request to someone you are already connected to. LinkedIn rejects it outright, and the step fails. So for an audience you are mostly already connected to, drop the invite step and build the sequence as DM only. This is not a workaround, it is a genuinely strong play: your existing connections who went cold are one of the warmest audiences you have, they need no invitation budget at all, and a DM to an existing connection is a message inside a relationship rather than a request from a stranger. Teams running this at scale see it work well.

Test, enroll, activate

Four gates. Each one stops something from going out, which is the point.

1. Confirm the plan

The agent surfaces a plan before generating anything: goal, audience, motion, angle, CTA, step count and a label per email. This is your checkpoint. Read it and make sure it does what you intended, because everything downstream is built on it.

2. Publish

On the first pass, sequences are drafts, and drafts do not appear in the Sequences panel. Nothing you change sticks until you publish. Publish early, and republish after every edit. The agent usually republishes on its own. Confirm that it did.

3. Test run

Test runs are previews. They send nothing to anyone. They show your copy rendered against real contacts so you can read it, give feedback, and have the agent revise. Republish, run again, repeat. Test runs live in two places: the chat where you created the sequence, and inside the sequence itself (click in, click the diagonal arrow at top right, open the Test Runs tab). Read at least five, across different cohorts. One preview tells you the template works. Five tell you whether the personalization is real or whether the same opener is being recycled with names swapped.

4. Dedupe, then enroll, then activate

Run a dedupe pass first. Ask the agent to check your list against existing customers, open opportunities, and anyone already in another active sequence. This costs two minutes and is worth it every single time. Cold-prospecting an existing customer is the kind of mistake that gets forwarded internally. Then tell the agent to enroll the contacts from your list into the published sequence. Nothing sends until you hit Activate. It is the only action that starts sending.

Changing a live sequence

You can edit a sequence after it is running. To swap a phrase, reroute a sender, fix a step, or change the timing, pause the sequence, make the change with the agent, then resume it. Contacts partway through pick up the new version at their next step. Editing a live sequence without pausing first is what strands people mid-flow, so the order is pause, change, resume.

What happens when someone replies

A reply pulls that person out of the sequence so they stop receiving messages. From there it is on you.

• Email-only sequence: an email reply removes them

• LinkedIn-only: a LinkedIn response removes them

• Both channels: a response on either one removes them from the whole thing

The reply is the goal, not the meeting

You are not trying to get a meeting from a first email. You are trying to start a conversation. A meeting is a big commitment to make to a stranger, and asking for one immediately is the most common reason good outbound gets ignored. So the first email asks for something small, and you earn the meeting in the exchange that follows. This is why the copy rules above push toward one easy question rather than a calendar link.

Expect most replies to be no

Roughly half of all replies are out-of-office messages, "not interested," "wrong person," or an unsubscribe. This is normal. A 4% reply rate where 2% are genuinely interested is a working campaign, not a failing one. If you are new to this, that ratio is the thing most likely to make you conclude outbound does not work when in fact it is working.

The four replies you will get, and what to do with each

Interested. Reply fast. Speed matters more than polish here, because interest decays. Answer their actual question, then propose a specific time rather than asking when suits them.

Wrong person. This is a good reply, not a dead end. They have told you the person you want exists and is not them. Ask who it is. People usually tell you, and a referral inside the company converts far better than a cold email to the same person would have.

Not now. The most undervalued reply in outbound. "Not until next quarter" is not a rejection, it is a date. Note the reason and the timing, and go back then with a reference to what they said. Most teams treat these as losses and email them again six months later from scratch, which wastes the one thing they were given.

Not interested. Take them at their word and stop. Remove them, and do not send them a clever follow-up asking why. The reply rate you gain is not worth the reputation you spend, and complaints are the fastest way to damage your sending.

One thing not to do

Do not pitch in your reply. They wrote back to a short message asking one question, and a three-paragraph product overview in response is a bait and switch. Match the register they replied in.

Checklist

• Senders assigned by seniority, LinkedIn routed to whoever has the reach

• Connected Account toggle set, or warmed accounts uploaded

• Signature added, plain text, no links or graphics

• Cohorts built by persona or situation

• Variants limited to something you can read a result from

• Ban list handed to the agent

• One or two of your own emails given as voice reference

• Plan confirmed before generation

• Five-plus previews read across different cohorts

• Published, and republished after the last edit

• Dedupe pass run

• Contacts enrolled

• Activate