Start Here: Build and Run Pipeline in Lightfield

How to build and run pipeline in Lightfield. Every step happens by talking to the agent in chat. Read this first, then work through the numbered articles in order.

Already know what you want to send? Build up your knowledge base first, then go straight to the Building Sequences article. Skip the target list work until you need new companies.

Start here: build your APG knowledge base

Everyone builds this first, whatever you plan to run. It is steps 1 and 2 of setup.

Step 1: Build ICP

Lightfield mines your CRM directly - won deals, churned deals, lost deals, and the accounts behind them. It does not cold-interview you. Where your CRM fields are thin, it deep-dives a sample of 20 to 30 lost deals and pulls the real loss reasons out of transcripts and email history, because loss-reason fields are almost never filled in reliably enough to trust on their own.

It writes five reference files:

• icp.md - vertical buckets ranked with won and lost counts, headcount and revenue bands, funding stage mix, the incumbent tools you displace, and trigger events observable in data

• buyer-persona.md - winning titles by role bucket, persona by company size band, acquisition channel mix, champion versus economic buyer patterns, and the persona mistakes that predicted losses

• anti-icp.md - your explicit do-not-sell-to list ranked by signal strength, real loss reasons, recurring feature gaps, who beats you, and the warning signals that precede a silent loss

• win-loss.md - conversion patterns by buying trigger and prior tool, a prior-tool index, a persona index, and an honest record of which CRM fields were too sparse to conclude anything from

• index.md - the lookup hub the rest hang off

Feed it more than your CRM

The CRM scan is the floor. Treat adding your own context as part of the build rather than an optional extra. Your CRM knows what you sold and to whom. It does not know the value prop you shipped last month, the vertical you started moving into in Q2, or the objection your reps began hearing in April. That material changes who you should target and how you should talk to them, and none of it is in a closed-won record.

Two ways to add it, and you can keep adding at any point:

• In chat. Tell the agent to update the APG folder with the context, then paste it or attach the file.

• In the folder. Open the APG folder and use the chat box inside it.

Worth feeding:

• Updated decks and positioning that reflect a new value prop

• Competitive intel and objection handling you have picked up since your CRM data was written

• A market you are moving into that your deal history does not represent yet

• Qualitative knowledge your team carries that no CRM field ever captured

• Named accounts or segments you want treated as priorities for reasons the data cannot see

If you add something substantial after setup is done, tell the agent to refresh your scoring rubric so it accounts for the new context.

Step 2: Build scoring rubric

This step does ask you two questions: any strategic priorities that sit outside fit (logo prestige, geographic focus, partner ecosystem, public-company status), and how much weight fresh trigger events should carry against baseline fit.

It then writes a 0 to 100 scoring model across three pillars - ICP fit, outbound timing, strategic - with weights summing to 100 and A through D bands cutting at 80, 60 and 40. Alongside it, a signals catalog: the observable facts that justify reaching out now, each mapped to the buyer pain it implies. Scores are additive. A weak account earns low pillar scores rather than getting marked down, so a low score means weak fit rather than a penalty you need to go hunt for. Your do-not-sell-to list does its work earlier, screening what enters the list at all.

What the folder does from then on

It filters what you already have. Point the agent at your CRM with the knowledge base loaded and ask it to surface accounts worth retargeting or reengaging. Closed-lost from eighteen months ago, churned logos, deals that stalled at security review. It reads those records against who you sell to today, against the buyer persona, and against your anti-ICP, and returns the subset that fits now. A lot of what looked like a bad fit two years ago was a timing problem or the wrong contact in the room.

It drives everything new. Sourcing searches against your sweet-spot verticals and screens candidates on the do-not-sell-to list. Scoring ranks survivors against the rubric and writes a score, its rationale, and signal chips onto each account. Contact selection ranks people against the buyer persona. Copy draws on the signals catalog and the pain each signal implies, plus whatever it can find on the specific person. One folder, both jobs. That is why it comes before any decision about cold outbound.

One folder holds more than one market

You do not need a second APG folder. Your verticals sit inside the ICP as ranked buckets, each carrying its own weight in the scoring rubric, so a second or third market shows up as a different band rather than a separate setup. Feeding in context about a market you are moving into is how you get it represented before your deal history catches up. Where it strains: two markets with genuinely different buyers and genuinely different loss patterns get averaged into one persona and one disqualifier list. Say so during the step 2 interview so the rubric separates them. The practical separation happens at the list level. Run target-account sourcing once per market, keep a separate target list for each, and sequence each with its own angle. Same knowledge base underneath, different lists on top. Refreshes re-derive from your whole deal history rather than layering onto the last version, so expect your rankings to move as your customer base does.

The Setup: Your Knowledge Base article has the build steps.

Then two tracks, and you can run both

Nothing after the knowledge base is sequential. Most teams should have both of these moving in the same week.

Track 1: sequence people you already have

Fastest path to replies. What goes in:

• CRM accounts the knowledge base surfaced as worth retargeting or reengaging

• Event and conference lists you imported

• Existing customers, for expansion or advocacy plays

• Anyone with a concrete reason for the message to exist: shared investor, referral, website visit, champion who changed jobs

Sends from your connected work inbox. No domains, no warm-up, no DNS. Replies land in days. See the Play Library article for what to run, then the Building Sequences article to build it.

Track 2: build a cold target list

Steps 3 through 7 of setup. Source net-new companies against your ICP, create them, score them against the rubric, prune everything below your threshold, then find the right people at every account that survives. Verify email coverage before you write a line of copy. Slower to stand up, and it is the only track that eventually needs its own sending domains.

Two things to expect here. Steps 4 and 6 stop at a draft on purpose. They build the scoring automation and the contact-finding automation, and both are left unpublished so you can test them before they touch real records. Step 7 stays blocked until step 6 is published, because a draft automation cannot run. Setup is resumable. Stop halfway, come back in a new session weeks later, and it picks up at the next incomplete step. See the Setup: Your Target List article.

Why the gates exist

There are checks in this process that feel like friction. Verification, previews, publish before activate, daily caps. Here is the reasoning, once, so the rest of the collection makes sense.

Your sending reputation attaches to your domain, and every message you send shares it. Prospecting, live deals, renewal conversations, support replies, invoices. Providers score the domain, not the campaign. That reputation is the one asset in the stack you cannot repurchase. A new domain is cheap. Rebuilding trust on your primary is slow, and while it rebuilds, mail to customers lands in spam alongside everything else.

So the gates map to the four ways this goes wrong:

• Unverified lists bounce, and bounce rate is the fastest signal that a sender is careless.

• Unread previews send broken copy to a few hundred people at once, and a bad merge field is visible to every one of them.

• Publish and Activate are separate because a send cannot be recalled.

• Daily caps and send gaps exist because volume spikes and metronome timing are exactly what spam filters look for.

Lightfield enforces the caps and the pacing for you. The checks are the part that needs a person.

Sending limits

Defaults are set conservatively. Hard caps are enforced.

Warmed mailbox numbers are per mailbox, so a fleet of three at default settings carries 150 a day. LinkedIn invites cap at 20 a day, which makes LinkedIn a precision channel. Spend those on people worth a personal note.

Sizing, worked backwards

Cold converts at roughly 1% of people emailed into meetings booked. So 10 meetings a month means reaching about 1,000 people a month, which is 50 a day across 20 working days.

• Connected inbox at the default 20 a day gets you about 400 a month, or roughly 4 meetings.

• Connected inbox at its 50 a day cap gets you to 1,000, running pinned at the ceiling with no headroom.

• One warmed mailbox at its 50 a day default gets you to 1,000 with room to grow to 100.

Volume should come from many small campaigns rather than one large one. Small, sharp campaigns reply at close to triple the rate of large ones, so more good campaigns beats bigger ones.

When you need sending domains

Two separate reasons, and either one is sufficient. Volume: above roughly 1,000 emails a month, your connected inbox runs out of ceiling. Isolation: cold outbound on your primary domain puts customer mail and live deals inside the blast radius. Separate domains keep that contained, at any volume. Warming takes four to six weeks and the clock runs whether or not you have started. If either reason applies to your plan, start it now and run Track 1 while you wait. If you already have warmed sending accounts elsewhere, upload them and use them here. The Sending Safely article has the setup.

A note on vocabulary, because it generates the most support tickets: a warm audience is people who already know you, and a warmed domain is a sending account prepared to carry cold volume. Same word, unrelated things. Warm audiences run from your normal work address with no setup at all.

What Lightfield does, and what you own

Lightfield handles the labor: research, sourcing companies and people, enrichment, scoring, drafting copy. It works at a scale and consistency no person matches, and it performs the same on candidate 300 as on candidate 3.

Three decisions stay with you.

• Which markets you point it at, and how you separate them. Ranked inside the ICP, separated at the target-list level.

• Your angle and your register. The copy craft is already built in: length limits, opening on a real signal rather than a compliment, one idea per email, one ask, plus an explicit blocklist of phrases that read as automated. You supply the specific reason you are reaching out. Paste in two emails you have actually sent and it writes like you. Two minutes.

• When to push and when to hold. Volume, pacing, and killing a campaign that is not landing.

What the sequencer does not do

Worth knowing up front.

• Sequences run in a straight line. No conditional logic or branching.

• Email and LinkedIn only. No SMS, no dialer.

• Step timing and content are edited by talking to the agent.

• Lists are static. Re-run them to pick up new signals.

• Newsletters, product announcements and design-template blasts belong in your marketing tool. Lightfield sends each person their own short series of messages and stops the moment they reply.

• Built for a few hundred specific people at a time. A high-volume BDR floor with inboxes rotating behind each rep is the wrong fit, and we would rather say so now.

Two things that look like bugs

A sequence is a draft until you publish it, and publishing sends nothing. Sending starts when you press Activate. Any edit after publishing needs a second publish to go live.

A finished sequence looks identical to a running one. Nothing turns red. Sending stops, and replies dry up two or three weeks later. The Keeping It Running article covers the weekly check that catches this.

The seven articles in this collection

1. The Play Library - What you can run, easiest to hardest, and what each play requires.

2. Setup: Your Knowledge Base - Building your ICP files and scoring rubric. Everyone starts here.

3. Setup: Your Target List - Sourcing companies, scoring them, finding the right people. Track 2 only.

4. Building Sequences - Writing messages that earn replies, and getting a sequence live safely.

5. Sending Safely: Domains, Inboxes, and Limits - Volume, reputation, and setting up domains properly.

6. Keeping It Running - The weekly hour, and how to keep pipeline from quietly drying up.

7. Troubleshooting and Known Limits - What to do when something is stuck, and an honest list of current gaps.

Before you activate

Each article carries its own checklist. This is the summary.

• All five ICP files built, and read

• Your own context added: decks, positioning, competitive intel, new markets

• Do-not-sell-to list reviewed against your own judgement

• Scoring rubric built after your context was added, weights and bands sanity-checked

• Separate target list per market, if you sell into more than one

• Every email address verified

• Cold only: SPF, DKIM and DMARC configured on your sending domains

• Signature added as plain text, no links or images

• Contacts grouped so different personas get different messages

• Several test runs read, across different groups

• List checked against existing customers and live deals

• Sequence published, and published again after your last edit

• Next campaign already decided

• Press Activate