fivos-icp-segmentation

Fivos ICP + Segmentation

First step in every Workflows client build. Produces a structured ICP matrix and a tiered segmentation plan that downstream skills consume. Nothing else runs until the ICP is signed off.

Inputs

• Client name (required) - used for tagging in the company DB.
• Product one-liner (required) - what the client sells, in one sentence.
• Prior customer logos (3-10, optional but strongly recommended) - closed-won accounts.
• Bad-fit logos (optional) - churned, lost, or poor-fit accounts.
• Known ICP hypotheses (optional) - the client's own view of their ICP.
• Geo scope (default: global) - regions in scope.

If the engineer doesn't have customer logos, ask for them before proceeding. Sampling 5-10 real closed-won accounts beats any hypothesis the client can articulate.

Outputs

1. ICP matrix (Markdown doc) with firmographic, technographic, and fit-signal columns filled in per tier.
2. Tier plan - Bullseye / Expansion / Experimental - with named thresholds, target account count per tier, and reasoning.
3. Filter JSON block - copy-paste input for fivos-account-sourcing. Uses AI Ark structured filter syntax (see feedback memory: AI Ark uses structured filters, not NL prompts).

Approval gate

Before handing off to sourcing, get explicit sign-off from the engineer's manager on:

• Tier thresholds (what makes something Bullseye vs Expansion).
• Expected account count per tier.
• Any signal that materially changes spend (e.g. funding triggers that pull in non-ICP accounts).

No sourcing runs without this sign-off.

Workflow

Step 1 - Extract the ICP matrix from prior customers

If customer logos exist, fetch firmographic + technographic data on each:

• Headcount
• Revenue (estimate is fine)
• Industry (both NAICS and plain-English)
• Geo (HQ + operational regions)
• Tech stack (key platforms in the buyer's workflow)
• Funding stage (for venture-backed ICPs)
• Growth signals at time of close (hiring, expansion, funding)

Look for the tightest intersection. A tight ICP is better than a broad one - broad ICPs produce noisy campaigns.

Step 2 - Build the ICP matrix

Fill this table (inline in the output doc):

Dimension	Bullseye	Expansion	Experimental
Headcount
Revenue band
Industry
Geo
Tech stack
Funding stage
Key fit signals
Expected close rate	High	Medium	Unknown

Tier definitions:

• Bullseye - the tight intersection. These are the "we close these when we find them" accounts.
• Expansion - one step out on any dimension. Worth running at lower personalisation depth.
• Experimental - hypothesis accounts. Run in small volumes to validate before scaling.

Step 3 - Size each tier

For each tier, estimate the total addressable count using internal-DB knowledge first, then bounded counts from AI Ark / Discolike / Ocean.io. Output a range, e.g. "Bullseye: ~2,400 globally, ~1,100 in-geo."

Step 4 - Encode as filter JSON

Output one filter block per tier, using AI Ark's structured filter schema. Example structure (not NL):

{
  "tier": "bullseye",
  "account": {
    "industry_codes": ["518210", "541511"],
    "employee_count": {"gte": 50, "lte": 500},
    "hq_country": ["US", "CA", "GB"],
    "tech": {"any_of": ["Snowflake", "Databricks"]},
    "funding": {"stages": ["series_b", "series_c"]}
  }
}

Structured filters are faster, cheaper, and more precise than NL prompts for AI Ark - use them every time.

Step 5 - Segmentation cuts

Within each tier, propose segmentation cuts for downstream campaign differentiation:

• By industry vertical (so the copywriter can write industry-specific hooks).
• By geo (so inbox assignment and sending hours can be region-aware).
• By size band (so the copywriter can differentiate messaging to 50-person vs 500-person companies).

Each cut becomes a downstream campaign. Recommend no more than 3-4 active cuts per tier or you lose statistical signal.

Step 6 - Signal library

List the 5-10 triggers that, when they fire, materially raise the fit score. Standard starters:

• New funding round in the last 90 days.
• Target-role hiring (e.g. "hiring VP of Data").
• Tech adoption (new logo in the stack).
• Headcount growth >15% QoQ.
• Competitor churn signals.
• Product launch / expansion announcement.
• Regulatory / compliance deadlines (if relevant).

These feed fivos-campaign-strategist (which plays to run) and fivos-lead-scoring (how much weight to give each).

Output template

# ICP + Segmentation - <Client>

**Date:** <YYYY-MM-DD>
**Owner:** <GTM engineer>
**Status:** Draft / Approved

## Product one-liner
<one sentence>

## Tier matrix
<the table from Step 2>

## Tier sizes
- Bullseye: ~X accounts (in-geo: ~Y)
- Expansion: ~X
- Experimental: ~X

## Segmentation cuts
- Bullseye / Vertical A
- Bullseye / Vertical B
- Expansion / Geo - EMEA
- ...

## Signal library
- ...

## Filter JSON blocks
<one block per tier>

## Approvals
- [ ] Manager sign-off on tier thresholds
- [ ] Manager sign-off on expected account counts
- [ ] Client sign-off on ICP (if required by contract)

Handoff

Once signed off, invoke fivos-campaign-strategist with this doc as input. The strategist picks plays and sequence shapes per tier. After that, fivos-account-sourcing runs the filters.

References

• ~/.agents/skills/gtm-meta-skill/finding-companies-and-contacts.md - for provider mix context.
• ~/.agents/skills/gtm-meta-skill/provider-playbooks/ai_ark.md - filter schema.
• Feedback memory: AI Ark uses structured filters, not natural language.