---
title: Preparing to send | Lightfield
description: Build the knowledge base your sequences read from, and turn it into a scored target list with the right people at each company.
---

Before you write a sequence, give Lightfield two descriptions: the kind of company that buys from you, and the kind of person inside it who decides. Lightfield uses the first to find and rank companies, the second to pick who to contact, and both to write the copy.

If these are wrong, everything downstream is wrong, and you can’t fix it in the copy. When something changes later, you change it here, once, and everything built afterward picks it up.

This page has two parts:

- **Your knowledge base** (steps 1 and 2). Everyone does this, whatever they plan to run.
- **Your target list** (steps 3 to 7). Only for cold outbound to companies not yet in your CRM.

Each step starts from a chat message. Each step also checks that the earlier steps’ output exists, so if you stop partway you can come back later and ask for the next step.

## Part 1: your knowledge base

Your outbound context lives in its own **APG** folder in workspace [Knowledge](https://crm.lightfield.app/crm/knowledge), separate from the rest of your knowledge. You don’t create it by hand: the first step creates the folder and writes files into it. Keeping it separate means you can update targeting and messaging often without disturbing the knowledge the rest of your CRM relies on.

Workspace knowledge can be edited by admins only, so an admin may need to run these steps or add to the folder.

### Do you have enough closed-deal history?

Step 1 reads the won and lost deals in your CRM. A hundred closed deals give it plenty to work from. Five don’t, and it will still hand you a confident description built on very little.

If your history is thin:

- Give it everything with real signal: the deals you closed, the ones you lost and why, named companies you know are a great fit even if they never entered your pipeline, and any positioning or research docs.
- Tell it you’re early and want a hypothesis, not a summary. Ask it to reason from the best-fit customers you named rather than averaging a handful of deals.
- Treat the first version as a draft, and correct it hard once your first sends come back.

### Step 1: Build your ICP

In chat, type `build my ICP`.

Lightfield reads your CRM rather than asking you to fill in a form. It pulls the deals you won, lost and churned, joins them to what it knows about each company, and reads the actual email and call history of a sample of lost deals to work out why they really died, because loss-reason fields are rarely filled in reliably.

It writes these files to the APG folder:

| File               | What’s in it                                                                                                             |
| ------------------ | ------------------------------------------------------------------------------------------------------------------------ |
| `icp.md`           | Verticals, size and funding bands, incumbent tools you displace, trigger events                                          |
| `buyer-persona.md` | Winning titles and seniority, persona by company size, and the persona mistakes that predicted losses                    |
| `anti-icp.md`      | Who not to sell to, real loss reasons, recurring feature gaps, who beats you                                             |
| `win-loss.md`      | Conversion patterns, a prior-tool index, a persona index, and which CRM fields were too sparse to conclude anything from |
| `index.md`         | The lookup hub the other files hang off                                                                                  |

When it finishes, read the files properly. Pay particular attention to `anti-icp.md`: who not to sell to is usually where a team learns something new. Lightfield also flags analysis it couldn’t finish because CRM data was patchy, and suggests fixes such as recording loss reasons. Each one makes the next refresh better.

Refresh your ICP every quarter, or whenever you add around ten closed-won deals. A stale ICP quietly degrades everything built on it.

### Feed it more than your CRM

Your CRM knows what you sold and to whom. It doesn’t know the value proposition you shipped last month, the market you’re moving into, or the objection your reps started hearing recently. Add that context as part of the build:

- Updated decks and positioning
- Competitive intel and objection handling
- A market you’re moving into that your deal history doesn’t represent yet
- Knowledge your team carries that no CRM field captures
- Named accounts or segments you want treated as priorities

To add it, tell the agent in chat to update the APG folder with the context, then paste it or attach the file. You can do this at any point.

### Step 2: Build your scoring rubric

Type `build my scoring rubric`.

This turns your ICP into a score from 0 to 100 for any account, plus a catalog of signals. It asks you two questions. Answer both properly: this is where your judgment enters the system.

1. **Strategic factors beyond fit**, such as logo prestige, geography, partner-ecosystem fit or public-company status.
2. **Fit versus timing**: how much fresh trigger events count against baseline fit. A split such as 70/30 favors accounts that look right; 50/50 favors accounts that look ready. If you’re unsure, start closer to fit and weight timing more once you trust your signals.

It writes two more files:

- `scoring-rubric.md`: dimensions across three pillars (ICP fit, outbound timing, strategic), with weights summing to 100 and bands at A 80+, B 60 to 79, C 40 to 59, D below 40.
- `signals.md`: labels that can attach to accounts (recently funded, hiring spike, leadership change) and the buyer pain each implies. That pain mapping tells your copy what to lead with.

Scoring is additive: a weak account earns low marks rather than being penalized. Clearly wrong accounts are screened out earlier, when companies are sourced.

If you add something substantial to the folder after this step, ask the agent to refresh your scoring rubric so it accounts for the new context.

### More than one market

You don’t need a second folder. Your verticals sit inside the ICP as ranked buckets, each with its own weight in the rubric. Where two markets have genuinely different buyers and loss patterns, say so when step 2 asks its questions, so the rubric separates them. Then keep a separate target list for each market and sequence each with its own angle.

### Changing things later

When positioning shifts or you learn something new about who buys, change it in the knowledge base, then re-score and let new sequences inherit it. Either edit a file directly, or hand the agent new evidence (a strong call, a new use case) and ask it to fold it into the APG files.

Don’t hand-edit scores or patch one sequence to work around a stale knowledge base. Patch downstream and every future sequence inherits the stale version.

| What changed                       | Where to change it                 |
| ---------------------------------- | ---------------------------------- |
| Who your ideal customer is         | `icp.md`, by re-running step 1     |
| Who the buyer is inside an account | `buyer-persona.md`                 |
| What disqualifies an account       | `anti-icp.md`                      |
| How accounts get prioritized       | `scoring-rubric.md`, then re-score |
| What signals you act on            | `signals.md`                       |

### Knowledge base checklist

- ICP built, and all five files read, `anti-icp.md` especially
- Your own context added: decks, positioning, competitive intel, new markets
- Data gaps the agent flagged written down somewhere you’ll see them
- Scoring rubric built after your context was added, with real answers to both questions
- Signals catalog reviewed, and the pain mapping makes sense to you

## Part 2: your target list

You need this only to reach companies that aren’t in your CRM. If you haven’t used up your warm plays yet, start there; they work without any of this, and faster. See the [Outbound playbook](/sequences/playbook/index.md).

Steps 4 and 6 create automations, and only workspace admins can create automations. If you aren’t an admin, ask one to run those two steps.

### Plan for the list to shrink

Every stage loses some. A typical first list looks like this:

| Stage                                                   | Count     |
| ------------------------------------------------------- | --------- |
| Companies found                                         | 400       |
| People found at the best-fit companies, about five each | 600       |
| After removing addresses that don’t verify              | about 420 |
| People you actually sequence, two or three per company  | about 300 |

Start with 50 companies and you finish with too few contacts to learn anything. That’s why a first list should be 200 to 500 companies. These figures are illustrative; yours will vary.

### Step 3: Find companies

Type `find target accounts`. The agent asks two things:

- **How many.** Ask for 200 to 500 for a first list. If you don’t give a number, it defaults to a small batch. Lightfield screens a wider pool of candidates before creating accounts, so this takes a while.
- **What to call the list.** Pick a name you’ll recognize in three months. Every later step asks which list to use.

The list is created as a target list and appears under **Demand** in the sidebar, where its table shows each account’s score, status, signals and contacts. If a list isn’t showing as a target list, open its options and choose **Make target list**.

Lightfield searches for companies that match your ICP, screens out obvious mismatches using `anti-icp.md`, and deliberately keeps borderline ones: a weak company that ranks low later costs nothing, while a good one thrown out early is gone. Companies already in your CRM are skipped as duplicates. For a small batch, the new accounts may come to you as a draft to approve before they’re created.

When it finishes, read what it rejected and why. If it threw out companies you’d have kept, or kept ones you’d have cut, adjust your ICP now. This is the cheapest moment to find out.

### Step 4: Set up scoring

Type `create the score prospect account automation`.

This creates the **Score prospect account** automation, which scores one account against your rubric and records a score out of 100, signal labels, and a short written rationale on the account.

It’s saved as a draft. Test run it on one account, check the output, then activate it. Step 5 can’t use a draft.

### Step 5: Score the list

Type `score the target account list` and name the list.

Scoring ranks the list so you know which companies to work first. It runs in three stages so you can catch a broken rubric before it runs hundreds of times:

1. **One account.** Read the score and the rationale. Does it match your own read of that company? The rationale shows whether Lightfield is reasoning about the company or just matching on a couple of fields. If it’s wrong, fix your rubric before continuing.
2. **The next ten.** Look at the spread. If everything lands between 70 and 75, your rubric isn’t telling companies apart. Go back to step 2 and put more weight on fewer things.
3. **The rest**, in batches of 100.

### Step 6: Set up people-finding

Type `create the find contacts automation`.

This creates the **Find contacts for account** automation, which finds the people at one account who best match your buyer persona (about five) and adds them as contacts on that account.

It’s also saved as a draft. Test it, then activate it before you go on. This is where most people get stuck: step 7 can’t run while this automation is a draft, and the message you see mentions step 7, not this one.

### Step 7: Find the people

Type `find target contacts`. The agent asks which scored account list to work from, and what score threshold to use (for example, accounts scoring 70 or more). If some accounts are unscored, it asks whether to score them first, include them, or skip them; nothing is silently dropped.

It then runs the people-finding automation once per selected account. The contacts it creates are linked to their accounts, so they show on your target list, and they’re also collected into a contact list. Before starting a large number of runs, the agent tells you how many and asks you to confirm.

Sequence two or three people per company, not one. If you bet a company on a single inbox, you lose some to people who are away, have left, or turn out to be the wrong person.

### Before you write anything: verify reach

Ask the agent to run a verification pass across your contact list to confirm email addresses and LinkedIn profiles. Never enroll a list you haven’t verified. Lightfield also checks addresses when contacts are enrolled and again at send time, and marks addresses that fail as **Undeliverable**, but a clean list protects your sending reputation from the start.

Coverage is usually best when the list was built through the steps above, because enrichment has the account context to work from. A raw list imported from elsewhere usually comes back thinner.

### Target list checklist

- Company list built (200 to 500), clearly named, showing as a target list
- Rejection reasons read, and they match your judgment
- Score prospect account automation tested and activated
- Whole list scored, and the spread of scores makes sense
- Find contacts for account automation tested and activated
- Contacts found, highest-scoring companies first, two or three per company
- Every address verified

## Next steps

- [Building sequences](/sequences/building-sequences/index.md): write and test your first sequence.
- [Sending safely](/sequences/sending-safely/index.md): read this before sending anything cold. If you’ll need sending domains, start warming them now.