Skip to content

creatop-gtm/apollo-operator / .claude/skills/apollo-icp-builder

apollo-icp-builder

Verified

Turn a business and its ICP into a real Apollo search, a shared profile.yaml, and one to three custom scoring signals, validated on a sample. Use to define who to target on Apollo before building a list.

FileKindSize
SKILL.mdskill md13.0 KB
references/angles.mdmarkdown11.3 KB
references/apollo-search-reference.mdmarkdown4.9 KB
references/scoring-starter-frameworks.mdmarkdown3.7 KB

references/angles.md

View on GitHub
# Angles: running more than one against a single ICP

An **ICP** is who to reach. An **angle** is *why they are relevant right now*. One ICP usually supports several angles, and picking between them is one of the highest-leverage decisions in the whole motion. It is also nearly free, because people search costs nothing.

---

## The three kinds, and why the difference matters

The kind of angle determines how long a list stays good, which determines how you should spend on it.

### Timing signal

Something just happened. They are hiring an SDR, they raised a round, someone changed jobs, they adopted a tool.

- **Sharpest relevance available.** The reason to write is obvious and specific.
- **Decays fast.** Once they hire, the moment is gone. A hiring-signal list is stale in weeks.
- **So: rebuild it, do not stockpile it.** Enrich close to send, and re-run the search rather than reusing a saved list.

### Technographic fit

They already use a tool your offer wraps around, replaces, or complements.

- **Self-qualifying.** Owning the product proves budget, intent, and category awareness at once.
- **Decays slowly.** Companies do not swap core tooling often.
- **Usually small.** A technographic universe is a fraction of the firmographic one. Treat it as a precision list, not a volume list.
- **Strongest when you have a credible claim on that tool** (a partnership, a certification, deep expertise). Then it is a moat, not just a filter.

### Structural fact

Something permanent about how the company is built. 11 to 200 employees with 0 to 2 people in sales. No marketing team. One location. Founder-led.

- **Does not decay.** True until the company changes shape, which is slow.
- **Usually the largest universe of the three**, so it is the volume workhorse.
- **Weakest single reason to write**, because nothing just happened. It needs the copy to do more work, and it pairs well with a vertical to stay specific.

| | Relevance | Decay | Typical size | Use as |
|---|---|---|---|---|
| Timing signal | Highest | Fast | Small to medium | Sharp, rebuilt often |
| Technographic | High | Slow | Small | Precision play |
| Structural | Moderate | None | Large | Volume workhorse |

---

## Size several before committing to one

Because people search is free, size every candidate before spending anything. Use `per_page: 1` and read `total_entries`.

**Always compute the unfiltered baseline too.** It shows what choosing an angle actually buys you. In a real session, one ICP produced:

| Angle | Universe |
|---|---|
| No angle (baseline) | 203,908 |
| Structural: 0-2 salespeople | 127,751 |
| Structural + vertical: B2B SaaS, 0-2 salespeople | 2,837 |
| Timing: hiring an SDR or BDR | 3,243 |
| Technographic: already uses the tool | 1,044 |

Seven angles sized in a few minutes, zero credits, and the decision made on numbers instead of instinct.

**Expect roughly 85% attrition** from raw universe to sendable, after drift cuts, deduping, verified-only filtering, and catch-all removal. Two real runs landed at 15% and 13%. So a 1,000-person angle is a 150-person campaign, and a 3,000-person angle is a first month of sending. Say that out loud before anyone gets attached to the big number.

### Check what the filter actually admits

A signal filter can let in a second population you did not ask for. On 2026-09-14 a `latest_funding_date_range` angle, meant to find companies that had just raised, was **34% companies that had just been acquired**, because Apollo's "latest funding" is the latest funding *event* and an acquisition is one. Pairing it with `organization_latest_funding_stage_cd` fixed it (see `apollo-signals`). The universe size looked reasonable the whole time; only a sample of the companies behind it showed the problem.

So before enriching an angle, look up a handful of the companies it returns and check they carry the signal the angle is named after. Budget for it: a company lookup is 1 lead credit each, and ten of them is cheap next to enriching a list built on the wrong premise.

---

## One angle per campaign

Two angles in one campaign means one message trying to be relevant for two different reasons, and a result you cannot attribute to either. Separate campaigns cost nothing extra and tell you which angle actually works.

Copy has to change with the angle, not just the list:

- **Timing:** lead with the event. "Saw you're hiring an SDR."
- **Technographic:** lead with the tool. "You're already running X."
- **Structural:** lead with the consequence. "Most 40-person companies without a sales team end up with the founder doing outreach at 9pm."

Same offer, three genuinely different first lines. If the same email would work for all three angles, the angles were not different enough to be worth separating.

---

## Suppress between angles, every time

Angles against one ICP **overlap**, and the overlap is invisible unless you check. In a real three-angle run: **85 people in the second list had already been enriched for the first**, and **19 more appeared in both new lists.** Every one would have been paid for twice and emailed twice.

- Keep one suppression file per business of every person id already enriched or enrolled.
- Subtract it from each new list **before** enriching, since that is where the money is.
- When two angles both want the same person, assign them to the **higher-intent** angle and remove them from the other, deliberately.
- **Person-id suppression does not stop company overlap.** Two angles can reach the same company through two different people. Apply the per-company cap across every part and angle of the ICP, not within one file: in a live run on 2026-09-14, **47 of the 59 contacts removed by the cap were removed because a higher-priority part already had someone at that company.** And give each campaign on the same ICP its own subject lines, so two colleagues comparing inboxes see two different conversations rather than one blast.
- **Put competitor exclusion in the angle definition when what you sell is something your competitors also buy**, such as data or outbound services. They reply because they are buyers, not because they are your ICP: in the same run, the first positive reply to a data-service campaign came from a competitor. Exclude them at search time, by keyword, NAICS, or a named domain list, not after they reply.

```bash
comm -12 new_ids.txt already_enriched.txt | wc -l    # what the overlap would have cost
```

Full mechanics in `apollo-list-builder`, step 3.

---

## Sequencing several angles

Do not launch four campaigns in one week. With one live campaign a result is attributable; with four it is noise, and a deliverability problem in one is hard to trace.

1. **Start with the sharpest**, usually a timing signal or a technographic fit. It is the fastest read on whether the offer lands at all.
2. **Add the volume angle** once the first has produced replies worth learning from.
3. **Keep the rest staged.** A free-stage list costs nothing to hold and everything to enrich early.

**Sending capacity sets the number of angles, not credits.** Every angle queues on the same pool of warmed mailboxes. Sizing is free and enrichment is affordable; the constraint that actually binds is mailboxes multiplied by sends per mailbox per day. A fifth angle on a pool that is already full is a list that waits, and a waiting list decays (see the timing-signal drift rule in `apollo-list-builder`). Do the capacity math before building another angle, not after. The volume math is in `sending-infrastructure`.

If the sharpest angle fails, that is information about the **offer**, not the angle. Widening the list will not fix it, and a bigger list of the wrong people is just a more expensive version of the same result.

**An angle cannot rescue the wrong persona either.** A good timing signal aimed at the wrong buyer still returns nothing: in a live run, a recently-hired-or-promoted signal sent to sales leaders, for an offer that founders buy, produced **706 sends and zero positive replies.** Check the persona before blaming the angle, and check the angle before blaming the copy.

---

## A fourth kind: first-party intent

Added v1.3. The three kinds above are all inferred from Apollo's data about the company. **First-party intent is different in kind: it is behaviour on your own property**, and person-level visitor filtering made it reachable headlessly (see `apollo-signals`).

Someone from the account read your pricing page four times last week. That is not a firmographic fact or a public event, it is an action aimed at you.

- **Highest relevance of any angle**, because the prospect started it.
- **Decays fastest.** Faster than a hiring signal. A pricing-page visit is worth writing about within days, not weeks. Sort by `last_visited_at` and work the top.
- **Smallest universe by a wide margin**, and hard-capped by your own traffic. Treat it as a queue to work continuously, not a list to build once.
- **Two hard gates**: the tracking script must be installed, and person-level identification is **United States only**. A non-US ICP gets company-level signal at best.
- **Filter `website_visitors_people_confidence_tier` to `high`** before you reference the visit in copy. Referencing a visit that Apollo guessed wrong is worse than not referencing it.

| | Relevance | Decay | Typical size | Use as |
|---|---|---|---|---|
| First-party intent | Highest | Fastest | Smallest | A queue, worked continuously |

It does not replace the volume angle, and it cannot be scaled by trying harder. It is capped by traffic, so it is the one angle where the growth lever sits outside outbound entirely.

---

## Mapping the advanced filters onto the taxonomy

Most of the filters that have no CLI flag are exactly the ones that build a sharp angle, which is why lane 3 matters here (see `apollo-operator`). What each is actually good for:

| Filter | Kind of angle it builds | Note |
|---|---|---|
| `organization_naics_codes` / `organization_sic_codes` | **Structural**, vertical slice | The precise way to cut a vertical. Far tighter than keyword tags, which match prose. NAICS is prefix-matched, so a shorter code is deliberately broader. |
| `organization_headcount_growth_range` + `..._past_n_months` | **Timing** | Growth is a decaying signal even though headcount is structural. Rebuild it, do not stockpile it. |
| `person_days_in_current_title_range` | **Timing** | New person in seat. A `max` of 180 finds someone still forming their plan; a `min` of 730 finds someone who owns the current mess. Opposite angles from one filter. |
| `person_total_yoe_range` | **Structural**, about the person | Separates a first-time head of sales from a fourth-time one. Changes the register of the copy more than the offer. |
| `organization_founded_year_range` | **Structural** | Company age as a proxy for how set the processes are. |
| `market_segments` | **Structural** | Matched against tags and name, so it is fuzzy. Size it before trusting it. |
| `person_linkedin_urls` | Not an angle | A lookup path for a list you already have, not a way to discover one. |
| `website_visitors_people_*` | **First-party intent** | See above. |

Two of these deserve the warning attached to `organization_department_or_subdepartment_counts`: an unrecognised department key, and a market segment that matches nothing, are both **accepted by the schema and silently return no matches** rather than erroring. A zero-result angle is usually a typo, not an empty market. Size it against the unfiltered baseline before concluding anything.
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 194kUpdated Claude CodeCodexVerifiedInstall