Ideal customer profile (ICP)

Your ICP is the structured answer to "what does a great Hydra-customer-of-yours look like?" It lives on your tenant alongside your Company Profile and gets used in two places now (with more on the way):

1. The AI brief that every Hydra AI reads before it answers, designs, or writes for you.
2. Lead Discovery (Marketing → Outreach) — when productized lead-gen ships, it'll source new prospects matching your ICP.

It is separate from your own classification. "Industries" on your Classification card describes your business — you might be a SaaS company. "Target industries" inside ICP describes the businesses you sell to — which might be construction firms, retail, agencies, or anything else.

Where to find it

Settings → Company Profile, second card from the top. The fields are:

• Target industries — comma-separated. Industry tags for the businesses you sell to. (e.g. saas, b2b-services)
• Target company sizes — multi-select chips. Toggle the segments that match your customers:
  • Startup (<50 employees)
  • SMB (50-250)
  • Mid-market (250-2,000)
  • Enterprise (2,000+)
• Buyer roles — comma-separated. Decision-maker titles. (e.g. Head of Support, VP CX, Operations Director)
• Competitors (current vendors) — comma-separated. The tools your target customers use today. These are who Hydra's lead-gen will source new prospects from — the "they currently use Zendesk; you can sell them Hydra" wedge. Use product names as commonly written (Zendesk, Intercom, HubSpot).
• Problem solved — one or two sentences. The specific pain you solve. Be concrete: not "improve customer experience" but "support teams drowning in low-value tickets a bot could deflect."
• Disqualifiers — comma-separated. Anti-fit signals — types of companies you explicitly do NOT want as customers. (e.g. agencies serving end-clients, consumer brands, regulated healthcare)

Edit any field directly. Edits save on blur and bump your profile version.

How it gets populated

Hydra auto-derives all six fields from your onboarding answers — primarily the dedicated onboarding question "Who are your best-fit customers, and what tools do they use today?" The Sonnet classifier reads that answer (plus your Customer segments and Business description) and structures it for you.

If the auto-derivation isn't quite right, edit the structured fields directly. Your manual edits override the AI-derived values; they won't be overwritten by future regenerations unless you click Save & regenerate brief at the bottom of Company Profile, which re-runs the classifier from scratch.

How it's used today

1. The Sonnet context brief that every AI in Hydra reads before it does anything for you (designing reports, drafting flows, building mini-apps, writing bot prompts). When the brief mentions who you target and what they currently use, downstream AIs make smarter recommendations — e.g. mini-app suggestions calibrated to your actual buyer.
2. Lead Discovery (scripts/leadgen/discover.ts) — when run for your tenant, the script reads your ICP and uses it as the rubric. Your competitors list drives the BigQuery filter (which sites in HTTP Archive to surface). Your target_industries, target_company_sizes, target_buyer_roles, problem_solved, and disqualifiers are interpolated into the Sonnet scoring prompt so each prospect is judged against your fit criteria, not Hydra's. The personal_line cold-email opener is written in the voice of someone selling your product to your buyer roles. Run as: npx tsx --env-file=.env.local scripts/leadgen/discover.ts --tenant <your-slug> --limit 50 --dry-run first to estimate cost.
3. The Marketing → Outreach page — the prospects table reflects whichever discovery runs have been performed for your tenant.

How it'll be used soon

• Imported lead scoring — when you import a CSV of leads, Hydra will optionally score each row against your ICP and flag the ones outside it.
• Bot persona tailoring — bots will know what kind of business is on the other end of every chat and adjust tone/recommendations accordingly.

Why this is separate from your tenant Classification

Two adjacent concepts that look similar but solve different problems:

They drift independently. A SaaS company that sells to construction has industries=['saas'] and icp_profile.target_industries=['construction']. Conflating the two would leak your customers' industries into your own tenant's suggestions, and vice versa.

Tips for high-quality ICP

• Be specific in Problem solved. "Support teams drowning in low-value tickets a bot could deflect" beats "improve customer service." The downstream AI uses this verbatim when describing what you do.
• List 1-3 competitors, not 8. The first three are who Hydra's lead-gen will source against. Long lists dilute targeting.
• Use real disqualifiers. "Agencies, consumer brands, healthcare" is more useful than leaving the field empty. Disqualifiers prevent false-positive matches that waste outreach budget.
• Update when your positioning shifts. ICP isn't set-and-forget. Every time you re-evaluate your wedge, revisit this card.