gtmskills.com

Reference file

README.md

readme-md.md

1:1 ABM Ads Skill

The complete skill for running 1:1 ABM (Account-Based Marketing) ads on LinkedIn - paid LinkedIn advertising aimed one company at a time. You build one ad set per named target company (targeting employers = {that company} plus a narrowed buyer layer) and run personalized single-image ads - the company's own name/logo in the creative - each driving one click to an asset built for that account. This kit is everything needed to plan, size, target, design, and launch that ad campaign end to end. Self-contained; no dependencies on any other repo.

Safety first. Every build step creates objects in DRAFT (no spend). Activation is a separate, explicit step. Nothing is ever deleted without an explicit human "CONFIRM DELETE". The scripts read credentials from scripts/.env - never commit or share the real .env, only env.example.txt.

What a 1:1 ABM campaign is (the model)

  • One LinkedIn ad set per target company. Targeting = employers = {that company} plus a narrowed layer (geo / function / titles / seniority) so the ad reaches the buyers, not the whole company.
  • All ad sets live under one campaign group, managed and reported together.
  • Creative is personalized per account - the company's name/logo in the creative is the whole point (5-10x CTR). Each ad drives one click to a landing page built for that account.
  • Not the same as 1:few / 1:many (a single uploaded company list with title targeting). This kit is for the true 1:1 whale/enterprise play.

Full strategy and the "why" -> sops/01-abm-strategy.md.

What's in the box

1-to-1-abm-ads/
├── README.md                        ← you are here: the full pipeline, end to end
├── SKILL.md                         ← one-screen index / router
├── requirements.txt                 ← Python deps (requests, python-dotenv)
│
├── sops/                            ← the knowledge: read these to understand the decisions
│   ├── 01-abm-strategy.md           ← strategy & the why: campaign types, list vs 1:1, sizing rules, sales orchestration
│   ├── 02-audience-sizing.md        ← the 300-member floor, sizing minimums, bidding for small audiences
│   ├── 03-build-and-launch-sop.md   ← THE build procedure: Phases 0-7 (inputs → size → group → ad sets → conversions → UTMs → ads → QA), plus the copywriting rules for the 3 layers (creative / commentary / landing page)
│   ├── 04-audience-narrowing-sop.md ← how to narrow each audience to the target band via the geo/function/titles/YoE levers + the entry-level rule
│   ├── 05-ad-creative-skill.md      ← how to design the personalized single-image ABM creative (copy structure, hierarchy, never-fabricate rule)
│   └── 06-image-generation-skill.md ← gpt-image-2 mechanics the creative skill depends on (endpoints, resolution, references, masks)
│
├── scripts/                         ← the executable pipeline (Python 3)
│   ├── env.example.txt              ← copy to .env, fill LINKEDIN_ACCESS_TOKEN + OPENAI_API_KEY
│   ├── resolve_and_size.py          ← STEP 1: company list → org URNs → audience size → runnable (≥300)  [read-only]
│   ├── narrow_audience.py           ← STEP 2: walk one account through the levers → recommended targetingCriteria  [read-only]
│   ├── render_creatives.py          ← STEP 4: generate personalized creatives via gpt-image-2  (template - adapt CONFIGS + brand)
│   └── build_campaign.py            ← STEP 5: config-driven builder - group + ad sets + conversions + UTMs + ads, all DRAFT
│
└── config/                          ← templates you copy and fill
    ├── account_list.example.csv     ← input for STEP 1 (name, linkedin_url)
    ├── narrow_config.example.json   ← input for STEP 2 (region, functions, titles, band, geo mode)
    └── build_config.example.json    ← input for STEP 5 (account/org IDs, budget, targeting, conversions, UTMs, per-company copy + image + landing page)

Prerequisites

  1. Python 3.9+. python3 -m venv .venv && source .venv/bin/activate && pip install -r requirements.txt
  2. LinkedIn Marketing API access token with r_ads, rw_ads, r_ads_reporting on an app that can reach the target ad account. Put it in scripts/.env as LINKEDIN_ACCESS_TOKEN.
  3. OpenAI API key (for creative generation) in scripts/.env as OPENAI_API_KEY.
  4. cp scripts/env.example.txt scripts/.env and fill both in.
  5. A landing page per account (personalized asset the ad links to) - you build these separately; the pipeline just needs the URLs.

The full pipeline (run in order)

Step 0 - Understand the decisions

Read sops/01-abm-strategy.md (why) and sops/03-build-and-launch-sop.md (the master build procedure - Phases 0-7). Every later step maps to a phase there.

Step 1 - Resolve + size the account list → scripts/resolve_and_size.py

Turn a list of company names/LinkedIn URLs into resolved org URNs with their addressable audience, and see which clear LinkedIn's 300-member floor.

cd scripts
cp ../config/account_list.example.csv ../config/account_list.csv   # then edit it
python3 resolve_and_size.py                       # region defaults to North America + Europe
# → writes ../config/account_list_sized.csv  (org_id, audience_size, can_run_300, match_basis)

Rows flagged first_unverified matched a same-name entity - eyeball those before trusting them. Anything under 300 isn't runnable as a 1:1 ad set (expand region or drop it). Details: sops/02-audience-sizing.md.

Step 2 - Narrow each audience to the band → scripts/narrow_audience.py

For each runnable account, decide the targeting so the ad set lands in the sweet spot. The tool sizes the account through the levers, prints the seniority mix + entry-level %, and outputs a ready-to-use targetingCriteria.

cd scripts
python3 narrow_audience.py --org 3254263 --geo-mode light
# or, with a config:  python3 narrow_audience.py --slug robinhood --config ../config/narrow_config.json

Target band: > 300 (hard floor) · sweet spot 300-1,000 · up to ~1,200 fine · 2,000+ trim hard toward 1,000. Entry-level ≤ 5% (ideally 0%), decision cluster (Manager/Senior/VP) on top. Full logic + the geo strategic fork: sops/04-audience-narrowing-sop.md.

Step 3 - Write the copy (3 layers)

The ad spans creative + commentary + landing page, designed together. The ad's only job is one click to the page. Rules, the point-of-view, and a worked example are in sops/03-build-and-launch-sop.md → "Writing copy for 1:1 ABM ads". Never invent a play, number, or claim that isn't on the page.

Step 4 - Generate the personalized creatives → scripts/render_creatives.py

Design the single-image ABM ad per account with gpt-image-2 (company name/logo as the hook). render_creatives.py is the working template - adapt the CONFIGS list and the brand values (colors, fonts, logos) to your brand, then run. Design rules: sops/05-ad-creative-skill.md; image-gen mechanics: sops/06-image-generation-skill.md.

cd scripts
# edit CONFIGS + brand values inside render_creatives.py first
python3 -u render_creatives.py     # renders in parallel to ./output/{...}

Step 5 - Build the campaign (DRAFT) → scripts/build_campaign.py

Config-driven builder. Fill build_config.json from Steps 1-4 (org IDs, budget, the narrowed targeting from Step 2, conversions, UTMs, and per-company landing page + image + headline + commentary). Runs Phases 1-7 of the build SOP with a verify gate after every create.

cd scripts
cp ../config/build_config.example.json ../config/build_config.json   # then fill it
python3 build_campaign.py --config ../config/build_config.json               # DRY RUN (no writes)
python3 build_campaign.py --config ../config/build_config.json --execute     # builds everything DRAFT

Mandatory and easy to forget: attach the account's active conversions to every ad set, and set UTMs at the ad-set level (not baked into ad URLs). Single-image link ads must be built as an article post (not media) or they lose the destination URL + CTA. All in sops/03-build-and-launch-sop.md (Phases 4, 5, 6).

Step 6 - QA + activation

GET each ad set + creative back and confirm targeting, budget, DRAFT status, conversions, UTMs, and that each ad has a destination URL + CTA + headline. Share the preview URL per ad set. Only after QA passes, activate (group → ACTIVE, then each ad set → ACTIVE) - real spend, explicit go required. Do NOT use personalized name/logo ads in Germany (privacy regulation).

The target-band cheatsheet

Audience Verdict
< 300 Won't deliver - not runnable
300-1,000 Sweet spot
1,000-~1,200 Fine
~1,200-2,000 Acceptable, trim toward 1,000
2,000+ Too broad - trim hard

Levers, in the order the narrowing skill applies them (light-geo default): light geography → job function (+ exclude junior seniority) → job titles → years-of-experience → final geography. Geo can also be the first lever if geo is the strategy - that's a decision you make up front (see sops/04-audience-narrowing-sop.md).

Non-negotiable safety rules

  • Everything is created DRAFT / PAUSED. Never create ACTIVE.
  • Never delete anything via the API without an explicit fresh human "CONFIRM DELETE".
  • Never fabricate copy, numbers, or claims - every word on an ABM ad must trace to the landing page or something the client provided.
  • Keep credentials in scripts/.env only. Share env.example.txt, never .env.

By Ivan Falco - Frontal