Master ICP & B2B Outbound Qualification Framework

START HEREHow to use this document

THE ICP IN ONE PARAGRAPHWho Softvole is for

Every clause is testable and is scored separately, so an account can pass one and fail another: a perfect-fit agency with no timing signal is Priority B; an agency hiring a WordPress developer that sells $2,500 websites is Nurture, not a lead.

§1Partner economics — what an agency must charge

Everything in this document derives from one relationship: the agency's client price must cover Softvole's production cost plus the agency's own sales, strategy, account management, revisions, overhead and risk.

agency_client_price  ÷  softvole_package_price  =  markup_multiple

≥ 3.0×   comfortable — agency keeps ≥ 67%; no pressure on Softvole's price
2.5×     healthy — the working target
2.0×     thin — possible for retainer-led agencies; expect negotiation
< 1.8×   unviable — the agency cannot sell this profitably, so it will
         either haggle, squeeze scope, or churn after one project

The five agency price bands

§2Why this version exists

Two audits produced this document. v2.0 fixed what ICP v1 got wrong about need. v2.1 fixes what v2.0 got wrong about money.

What v1 got wrong, and v2.0 fixed

Pilot Evidence Batch 001: 6 of 16 records (38%) that passed vendor “marketing services” filters were not agencies — a beverage brand, a course business, an ad network, a membership community, a personal brand, an outsourced SDR firm. Site verification is therefore a hard gate, not an option.

What v2.0 got wrong — the commercial gap

Garbage leads the old rules produced

• High-volume local-business agencies selling $2,000 sites — perfect capacity fit, impossible margin.
• Volume scoring that ranked a cheap 60-site-a-year shop above a studio doing four $25,000 builds.
• Capacity A agencies that subcontract everything at the lowest price they can find.
• Agencies whose portfolio is builder-based (Wix, Squarespace, GoDaddy) reading as “sells websites”.
• 20-person social-media agencies scoring on headcount while selling two websites a year.
• SEO agencies whose real demand is small hourly tasks — an offer mismatch that looks like a price objection later.

Good leads the old rules rejected

• Capacity D agencies with real dev teams and high prices — budget, process and peak-load pain. Now Tier 2-Premium (§5), not −8 points.
• Branding and design studios doing three high-value brand-plus-site projects a year — low count, high value, perfect for build-from-design.
• Independent agencies of 51–150 people with a small web function and Band 4+ pricing.
• Agencies under NDA with little public portfolio — Unknown now routes to research, never to rejection.
• Mixed-stack agencies that use Webflow for marketing sites and WordPress for content-heavy or WooCommerce clients.

Hard filters vs soft signals

Automation vs human

§3The true ICP — the business situation

Softvole is valuable in one situation: web demand exceeds reliable production capacity inside an agency that owns recurring client relationships and charges enough to pay for production.

§4The ten fit dimensions

A company that matches the profile is not the same as a company that can buy, and neither is the same as a company that needs help now. Each dimension is judged separately and stored in its own column.

§5Account tiers

Tiers describe account type and fit. Priority classes (§17) combine tier with evidence, commercial gate and timing. A Tier 1 account can still be Priority B or Nurture.

Tier 1A — design and branding studios FIRST PRIORITY

Tier 1B — marketing and digital agencies that sell websites

Tier 1C — vertical specialists

Tier 2 — secondary

Tier 2-Premium — overflow partners Recommended Addition

Tier 3 — experimental

Exclusion

§6Agency-type segmentation

One question sets the ranking: how directly does this agency's business create WordPress build work that it cannot deliver itself, at prices that pay for production? Rankings are ordinal and are re-ordered by results (§35).

§7Internal development capacity

§8Web / development service fit

What Clay should check on the agency's website, and what each finding does to qualification.

§9Commercial-fit architecture

Fifteen new fields, four derived. Every one is populatable from public sources or an explicit Unknown — nothing here is theoretical.

New raw fields Recommended Addition

Derived fields

Minimum viable partner

partner_value_estimate ≥ $7,500 / year     (≈ 3 build-from-design websites,
                                             or 1 design+build + 1 build,
                                             or 10 Growth care sites)
AND at least one of:
  · ≥ 3 website projects per year, or
  · ≥ 5 sites they maintain (care book), or
  · Band 4/5 with ≥ 2 projects per year

Proof demand — how much evidence they'll want before buying

Proof demand never disqualifies. It selects the motion and sets expectations for cycle length — and high-proof accounts are usually the highest-value partners.

§10Price-fit model

Step 1 — estimate the band from multiple signals

No single signal decides the band. Each contributes weighted evidence; the highest-weight available signals set the estimate, and agreement sets the confidence.

Step 2 — confidence

Step 3 — Price Fit

required_price = softvole_price(offer_fit)
band_midpoint  = {1: 2,000 | 2: 4,500 | 3: 8,000 | 4: 17,500 | 5: 35,000}
multiple       = band_midpoint ÷ required_price

multiple ≥ 3.0                        → Strong
multiple 2.2 – 2.99                   → Acceptable
multiple 1.8 – 2.19                   → Acceptable ONLY if recurring_model = retainer-heavy
                                         with evidence (care plans, multi-year clients);
                                         otherwise Weak
multiple < 1.8                        → Weak
band confidence = Unknown             → Unknown
price_sensitive_positioning = yes     → Weak (and DQ_PRICE_BAND if Band 1)

Price Fit is computed per offer. The same agency can be Weak for design+build and Strong for build-from-design — which is exactly why offer_fit must be set before Price Fit.

Step 4 — offer fit

Step 5 — how Price Fit enters scoring

Price band contributes 12 of the 25 Commercial Fit points, and Price Fit itself acts as a gate on the priority class. Full weights in §17.

§11Business model fit

Do not equate higher revenue with better fit. A $3M agency selling 30 small WordPress sites a year with one developer is a better partner than a $12M agency with a 10-person engineering team. Client lifetime-value indicators to capture: client logos with multi-year relationships, “since 20XX” statements, care plans, retainer pricing pages, and case studies that describe ongoing work.

§12End-client profile

Who the agency's clients are sets both how often websites get built and how much the agency can charge. v2.1 rates verticals by budget first, frequency second — the reverse of v2.0.

Scoring: vertical points apply only when price_band ≥ 3. Record client_verticals and measure performance by vertical in §35.

§13Buyer personas (person-level ICP)

Account fit asks “is this the right company?”; persona fit asks “is this the right person?”. Different titles feel different pain and hold different authority.

Rule: contact at most 2 people per account, sequenced (primary first, secondary 5–7 days later only if no reply). At ≤ 15 employees, contact only the founder/owner. Recommended Addition

Exclude titles: interns, juniors, individual contributors (designers, developers, SEO specialists), HR/recruiters, finance, sales reps/BDRs, “freelance” titles, advisors/board members. DISQUALIFIER at person level.

§14Geography

Geography is chosen for qualified opportunities, not TAM size. Timezone gaps assume a delivery team on UTC+6 (To Validate). Compliance notes are general information, not legal advice.

Change from v1: UK raised to P1 alongside the US; UAE/GCC moved from “secondary” to a separate experiment because its buying behaviour needs a different channel mix, which would distort email-based learning if mixed into core cohorts. Recommended Addition

§15Why now? Buying-trigger framework

Timing signals answer “why contact them this month?”. Each signal records a date, a source URL and a strength. Recency window: count full value ≤ 90 days, half value 91–180 days, zero after 180 days.

A signal never disqualifies on its own, except the negative events above, which downgrade or move to nurture. DISQUALIFIER status only comes from fit rules.

§16Signal hierarchy

§17Lead scoring system

Scores are sums of evidence-backed points. Priority is assigned by gates, not by totals. Every point has a business reason. Weights are a starting Hypothesis — recalibrate after Phase 2 (§32).

Order of evaluation

1. Hard disqualifiers   → any true = DISQUALIFIED (stop; no further credits)
2. Fit scores           → Account (20) + Capability (25) + Commercial (25) = FIT /70
3. Commercial gate      → price_fit Weak = max Nurture · Unknown = max Priority C
4. Persona score        → /15  (person level)
5. Intent score         → /15  counted ONLY if Account ≥ 12 AND Capability ≥ 12
6. Negative adjustments
7. Priority class by gates — totals only break ties

Account Fit — 20 points

Capability Fit — 25 points

Commercial Fit — 25 points

Persona Fit — 15 points (per contact)

Intent / Timing — 15 points

Negative adjustments

Priority classes

Signal Strength is a label, not points: Direct · Indirect-2+ · Indirect-1 · Weak · None · Negative.

§18What is a Goldmine Softvole lead?

Priority classes and their gates are in §17.

§19Negative ICP — exclusions and exceptions

§20Data confidence

Manual review is mandatory when: HUMAN REVIEW

• the lead is Priority A or Goldmine (review before first send);
• the price band was estimated from proxies only and the account would become Priority A or B;
• price signals disagree by two bands or more;
• is_agency, sells_web or capacity_class is Low confidence but the lead would otherwise be A/B;
• AI sources disagree with provider data (e.g., provider says 40 employees, team page shows 6);
• a competitor or network-ownership flag is uncertain;
• a founder post is used as the personalization hook (check meaning and tone);
• the prospect is in Canada (CASL consent basis) or a P3 market.

§21Outreach readiness gate

“Email verified” is a deliverability fact, not a qualification. A lead enters Instantly only when every line is true.

§22Final decision tree

Is the website live and is it a client-service AGENCY (evidence of client work)?
├─ NO ───────────────────────────────────────────────▶ DISQUALIFY (DQ_NOT_AGENCY / DQ_NO_SITE)
└─ YES
   Competitor, dev firm (Capacity E), network-owned, excluded category,
   out of geography, solo, or >200 staff?
   ├─ YES ───────────────────────────────────────────▶ DISQUALIFY (reason code) · exceptions → review
   └─ NO
      CAN THEY AFFORD US?  price band from multiple signals (§10)
      ├─ Band 1, or "affordable/budget" positioning ──▶ DISQUALIFY (DQ_PRICE_BAND)
      ├─ Band 2 ─────────────────────────────────────▶ NURTURE (landing page or care plan only)
      ├─ UNKNOWN ────────────────────────────────────▶ PRIORITY C (research; resolve on the call)
      └─ Band 3 / 4 / 5
         Do they SELL websites (service page + portfolio)?
         ├─ NO, but redesigns/migrations via SEO ────▶ Tier 2b
         ├─ NO web relevance ───────────────────────▶ NURTURE or DISQUALIFY (DQ_NO_WEB)
         └─ YES
            Who designs?  Who builds?  → set offer_fit
            ├─ Design in-house, no build capacity ──▶ build_from_design ($2,400) · Tier 1A
            ├─ Neither, band 4+ ───────────────────▶ design_and_build ($3,500)
            └─ Own dev team, band 4–5 ─────────────▶ overflow_capacity · Tier 2-Premium (proof-gated)
               Is the package affordable at ≥ 2.5× their client price?
               ├─ NO ──────────────────────────────▶ try a smaller package; else NURTURE
               └─ YES
                  Internal development capacity?
                  ├─ A (sells web) / B / C with overflow ─▶ HIGH capability fit
                  ├─ UNKNOWN ────────────────────────────▶ RESEARCH (max Priority C)
                  └─ D, no overflow evidence ────────────▶ NURTURE (revisit on dev departure/hiring)
                     Partner value ≥ minimum?
                     ├─ NO ──────────────────────────────▶ NURTURE
                     └─ YES
                        Relevant decision-maker with valid email?
                        ├─ NO ───────────────────────────▶ NURTURE (find the person later)
                        └─ YES
                           Timing signal ≤ 90 days?
                           ├─ Direct or 2+ Indirect ─▶ PRIORITY A ─▶ human QA ─▶ outreach
                           │                            + goldmine criteria? ─▶ GOLDMINE handling
                           ├─ 1 Indirect / Weak / None ─▶ PRIORITY B ─▶ segment campaign
                           └─ Negative event ──────────▶ NURTURE (6 months)

§23TAM → SAM → ICP pool → qualified → sales-ready

Five numbers that must never be confused. Only the last goes to Instantly.

What the funnel looks like after the commercial gate

1,000 raw US/UK agency records
  →  620   real, site-verified agencies          (62% — Pilot Evidence)
  →  280   sell websites                          (~45% — Hypothesis)
  →  170   meaningful WordPress share             (~60%)
  →  110   capacity A/B/C, or D with overflow     (~65%)
  →   40   price_fit Strong or Acceptable         (~35% — the commercial gate)
  →   30   with a contactable decision-maker      (~75%)
  →  ~30   Priority B pipeline
  →   ~6   Priority A (direct or 2× indirect signal ≤ 90 days)

Two consequences. Priority B is the volume engine, not Priority A. And 10,000 accounts are not needed: at $7,500–$30,000 per partner per year, 10–20 active partners is a real business, which means processing roughly 2,000–4,000 records properly — far below the 16,000 the original sourcing plan assumed, and far below its credit budget.

§24Clay execution blueprint

Exactly what to discover, in what order. Each step ends with a pass condition used as the run-condition of the next step's columns.

Step 1 — Find accounts

Step 2 — Clean accounts

• Domain normalization: lowercase; strip protocol, www., paths, query strings, trailing slash. Store as domain.
• Name normalization: strip LLC, Ltd, Inc, Pty, Co., “Agency”, “Studio” suffix for matching only; keep the display name.
• Deduplicate on normalized domain, then LinkedIn company slug, then normalized name + city. Check against the persistent registry and CRM suppression list before any enrichment (Pilot Evidence: registry already exists).
• Dead-domain detection: HTTP status + redirect target. Parked pages, “for sale”, redirects to another company → DQ_NO_SITE or merge.
• Agency / non-agency pre-screen: keyword rule on description; ambiguous rows go to Step 4 AI validation.
• Geography validation: HQ country from provider vs site footer/contact page; mismatch → flag for AI check.

Step 3 — Company enrichment

Step 4 — AI research (account)

Run one bundled account-research call (Prompt 1 + 2 + 3 + 4 combined, §28§) on rows that passed Step 3. It answers:

1. Is this a client-service agency? What evidence of client work exists?
2. What agency type is it, and what services does it sell?
3. Does it sell websites? Which platforms? How many website projects are visible, and how recent?
4. Does it appear to have internal web development capacity? (class A–E/U)
5. Is there evidence of outsourced, contracted or white-label development?
6. Does it appear to have recurring client relationships (retainers, care plans, long-term clients)?
7. Does it have enough client/project volume for a white-label partnership?
8. Is it independent? Is it a competitor?

Every answer returns answer, evidence (quote), source_url, confidence, reasoning. Unsupported answers must be Unknown. AI RESEARCH

Step 4b — Commercial research and the price gate

Run on every account that passed Step 4. Bundle into one AI call: pricing page, directory profile (Clutch and similar), portfolio sample with tech lookup, design and development capability. Derive the band, then price_fit and offer_fit. Full waterfall in §25, weights in §10.

Gate: continue only when price_fit is Strong or Acceptable, or Unknown at Tier 1 with capacity A/B/C. Weak stops here as Nurture.

Step 5 — Find people

Max 3 found, max 2 contacted per account. Exclude: intern, junior, associate, coordinator, specialist, freelancer, advisor, board, recruiter, HR, finance, sales rep/BDR. Only run on accounts that passed the Step 4b price gate, Tier 1–2, no DQ.

Step 6 — Person enrichment

Never construct emails from a domain pattern and send unverified. Pilot Evidence (data-integrity rule already in the pipeline).

Step 7 — Signal research

Only on accounts with FIT ≥ 48 and a contactable persona. Order, cheapest first:

1. Job postings on the company (provider lookup) → dev/designer/PM roles, dates.
2. Headcount trend + engineering departures (people data already pulled).
3. Portfolio recency (already in Step 4 output — reuse, don't re-run).
4. Founder/company LinkedIn posts, last 90 days → AI classifies into signal types.
5. News/announcements (AI web search) — only for rows still lacking a signal and FIT ≥ 55.

Step 8 — AI qualification

One final structured call (Prompt 9) answers the ten questions: fit? compatible services? plausible problem? capacity limitation? web demand? person relevant? reason to contact now? supporting evidence? what's uncertain? contact / review / nurture / reject? Its output feeds scoring formulas — it does not replace them.

§25Commercial enrichment waterfall

Runs after v2.0 Step 07 (AI account qualification) and before people search, so contact credits are never spent on a commercially impossible account. Each variable: first source, fallback, AI step, manual trigger, confidence rule, scoring effect.

Order and cost discipline

07  AI account qualification (v2.0)            → is_agency, sells_web, capacity
07b PRICING PAGE FETCH            cheap        → run on all rows that passed 07
07c DIRECTORY LOOKUP (Clutch etc.) medium      → run only where 07b found no price
07d PORTFOLIO TECH + QUALITY       cheap/med   → run on Tier 1–2 only
07e DERIVE band → price_fit → offer_fit  free  → gate here
08  FIND PEOPLE                               → ONLY price_fit ∈ {Strong, Acceptable}
                                                 or (Unknown AND tier 1 AND capacity A/B/C)
09  CONTACT ENRICHMENT                        → unchanged
10+ SIGNALS, AI QUALIFICATION, SCORING        → unchanged, plus commercial fields

Additional AI prompt — commercial fit AI RESEARCH

Company: {{company_name}} ({{domain}}). Clutch/directory URLs: {{directory_urls}}.
Portfolio client domains: {{portfolio_client_domains}}.

Estimate what this agency charges its own clients for a website. Use only
public evidence. Do not guess a number when no evidence exists.

Check, in order: the agency's pricing/packages page; any "starting at" or
"investment" statements; Clutch or similar directory minimum project size and
hourly range; the type and size of named clients; the complexity of portfolio
work; any "affordable/budget/cheap" positioning.

Return JSON:
{
 "published_price_min_site": "currency or Unknown",
 "published_pricing_quote": "", "published_pricing_url": "",
 "directory_min_project_size": "", "directory_hourly_range": "", "directory_url": "",
 "client_tier": "micro_local|smb|lower_mid|mid_market|enterprise|mixed|unknown",
 "named_clients": [],
 "portfolio_quality": "custom_high|custom_standard|template_builder|unclear",
 "price_sensitive_positioning": "yes|no", "price_positioning_quote": "",
 "estimated_site_value_band": "1|2|3|4|5|unknown",
 "band_evidence": [{"signal":"","value":"","source_url":""}],
 "band_confidence": "High|Estimated|Unknown",
 "design_internal": "yes|no|unknown", "dev_internal": "yes|no|unknown",
 "reasoning": "max 3 sentences, facts separated from inference"
}

§26Master field schema

Column names in snake_case for Clay, CRM and future agents. “Stage” = workflow step that fills it (§31§). Unknown company data → Unknown; unknown contact data → Not Found (Pilot Evidence convention).

§27Clay credit efficiency — cheap filters first

Credit prices change and differ by provider; check current costs in your Clay workspace. Relative cost tiers below are what matter. To Validate: exact credits per column on the trial plan.

§28AI research prompts

Reusable templates for Clay AI / Claygent columns. {{field}} = Clay column reference. All prompts share the evidence rules below; paste the rules block at the top of each prompt. In production, combine Prompts 1–4 into one call (§27§).

You are a B2B research analyst qualifying agencies for Softvole, a white-label
WordPress web design and development production partner for agencies.

EVIDENCE RULES
1. Use only evidence you can find on the company's website, its LinkedIn page,
   job postings, or reputable public sources. Visit pages; do not guess.
2. Separate FACTS (directly stated, quote them) from INFERENCES (your reasoning).
3. If evidence is missing or unclear, answer "Unknown". Never fill gaps with
   assumptions. "No evidence found" is not the same as "No".
4. For every answer give: evidence (short quote), source_url, confidence
   (High = direct first-party evidence or two sources agree; Medium = one
   credible source or strong inference; Low = weak inference).
5. Prefer the company's own pages: /services, /work, /portfolio, /case-studies,
   /about, /team, /careers, /partners, footer.
6. Return ONLY valid JSON matching the schema. No prose outside JSON.

Company: {{company_name}}  Website: {{domain}}  LinkedIn: {{company_linkedin_url}}
Description: {{description}}  Employees (provider): {{employee_count}}  Country: {{hq_country}}

Determine whether this is an independent CLIENT-SERVICE AGENCY (it delivers
marketing, design, web or SEO services to other businesses) and what kind.

Disqualify signals: sells its own product/SaaS, courses, memberships, events,
media inventory, staffing/recruitment, PR-only, printing, outsourced sales/SDR,
software/app development as core business, or white-label development sold to
agencies (competitor). Check for network/holding-company ownership.

Return JSON:
{
 "is_agency": "yes|no|unknown",
 "is_agency_evidence": "", "is_agency_source_url": "", "is_agency_confidence": "High|Medium|Low",
 "agency_type": "full_service|digital|seo|ppc|social|content|branding_design|web_design|wordpress|webflow|saas_mkt|b2b_mkt|lead_gen|advertising|not_agency|unknown",
 "services_list": [],
 "client_type": "smb|mid_market|enterprise|mixed|unknown",
 "client_verticals": [],
 "named_clients_or_case_studies": 0,
 "ownership": "independent|network|pe_backed|public|unknown",
 "hq_country_on_site": "",
 "competitor_flag": "yes|no|unknown", "competitor_evidence": "",
 "disqualifier": "none|DQ_NOT_AGENCY|DQ_DEV_FIRM|DQ_COMPETITOR|DQ_CATEGORY|DQ_NETWORK|DQ_GEO|DQ_SOLO",
 "reasoning": "max 2 sentences, facts vs inference labelled"
}

Company: {{company_name}} ({{domain}}). Total employees (provider): {{employee_count}}.
Engineering/developer titles found in people data: {{engineering_headcount}}
({{engineering_titles}}).

Assess INTERNAL WEB DEVELOPMENT CAPACITY. Read /team, /about, /careers and
LinkedIn. Count people whose role is building websites (developer, engineer,
front-end, WordPress, webmaster, "designer & developer"). Note contractors,
freelancers, "network of specialists", "trusted partners", or credits such as
"development by X" on portfolio sites.

Classes:
A = no visible developer, but the company sells websites
B = one developer (or a designer who builds), likely insufficient for volume
C = 2-6 developers AND evidence of outsourcing, contractors or dev hiring
D = strong internal dev team (4+ incl. senior/lead) with no outsourcing evidence
E = software/web development is the company's primary business
U = unknown / conflicting evidence
Do NOT assign A only because no developer was found in a database; you must
have checked the team page or equivalent. Otherwise answer U.

Return JSON:
{
 "developers_found": 0, "developer_names_titles": [],
 "contractor_or_outsourcing_evidence": "", "outsourcing_source_url": "",
 "capacity_class": "A|B|C|D|E|U",
 "capacity_confidence": "High|Medium|Low",
 "sources_agree": "yes|no|single_source",
 "evidence": "", "source_url": "",
 "reasoning": "max 2 sentences"
}

Company: {{company_name}} ({{domain}}).

Does this agency SELL websites to its clients? Check services pages and the
portfolio/case studies. Count distinct client WEBSITE projects (not logos,
not social campaigns). Note dates where visible. List up to 3 portfolio
client domains so their technology can be checked.

Return JSON:
{
 "sells_web_design": "yes|no|unknown", "sells_web_dev": "yes|no|unknown",
 "web_service_evidence": "", "web_service_url": "",
 "platforms_mentioned": {"wordpress":"yes|no|unknown","elementor":"yes|no|unknown",
   "woocommerce":"yes|no|unknown","webflow":"yes|no|unknown","shopify":"yes|no|unknown"},
 "sells_landing_pages": "yes|no|unknown",
 "sells_maintenance_or_care_plans": "yes|no|unknown", "maintenance_evidence": "",
 "sells_seo_or_cro": "yes|no|unknown",
 "web_projects_visible_24m": 0, "most_recent_web_project_date": "YYYY-MM or unknown",
 "portfolio_client_domains": [],
 "recurring_model_evidence": "", "recurring_model": "retainer|project|mixed|unknown",
 "confidence": "High|Medium|Low",
 "reasoning": "max 2 sentences"
}

Company: {{company_name}} ({{domain}}). Portfolio domains: {{portfolio_client_domains}}.

Look for evidence that this agency uses external development help OR sells
white-label development itself.
Evidence of BUYING: "partner network", "freelance developers", contract dev job
posts, footer credits on client sites naming another studio, partner pages,
founder posts asking for dev recommendations.
Evidence of SELLING (competitor): "white label for agencies", "we build for
agencies", "hire dedicated developers", "offshore development team".

Return JSON:
{
 "white_label_status": "buys_wl|sells_wl_dev|sells_wl_other|none|unknown",
 "evidence": [{"quote":"","source_url":"","date":""}],
 "named_vendor_if_any": "",
 "confidence": "High|Medium|Low",
 "reasoning": "max 2 sentences"
}

Company: {{company_name}} ({{domain}}). Today: {{today}}.
Job postings: {{job_postings}}
Recent company/founder LinkedIn posts (last 90 days): {{recent_posts}}
Headcount change 6m: {{headcount_growth_6m}}. Developer departures: {{dev_departures}}.

Classify every dated event into signal types. Only count events with a date
within 180 days. Quote the exact text. Do not infer a signal from generic
marketing language.

Types and strength:
DIRECT: hiring_web_dev, contract_dev_post, asks_for_dev_partner, at_capacity_post,
        dev_departure, new_web_service_line
INDIRECT: hiring_designer_or_pm, headcount_growth, multiple_site_launches,
          new_client_win, outsourcing_language, new_ops_leader, acquisition
WEAK: award, new_office, generic_growth_post, platform_usage
NEGATIVE: layoffs, building_inhouse_dev_team, acquired_by_network,
          new_whitelabel_partner

Return JSON:
{
 "signals": [{"type":"","strength":"DIRECT|INDIRECT|WEAK|NEGATIVE",
   "date":"YYYY-MM-DD","quote":"","source_url":"","days_ago":0}],
 "strongest_signal": "", "signal_strength_label":
   "Direct|Indirect-2+|Indirect-1|Weak|None|Negative",
 "confidence": "High|Medium|Low"
}

Person: {{full_name}}, title "{{title}}" at {{company_name}} ({{employee_count}} employees,
agency type {{agency_type}}). LinkedIn: {{person_linkedin_url}}.
Start date: {{role_start_date}}. Last post: {{last_post_date}}.

Is this person likely to own or strongly influence a decision to use a
white-label web production partner? Consider company size (founders decide at
under ~30 people), their function, and whether they still work there.

Return JSON:
{
 "persona_tier": "primary|secondary|tertiary|exclude",
 "authority": "decider|influencer|user|none|unknown",
 "currently_employed_evidence": "yes|no|unknown",
 "relevant_pain": "one phrase",
 "exclude_reason": "",
 "confidence": "High|Medium|Low"
}

Inputs: signals JSON {{signals_json}}, capacity {{capacity_class}},
web projects {{web_projects_visible_24m}}, last project {{most_recent_web_project_date}}.

Write the single strongest, evidence-backed reason this agency may need web
production help NOW. If no dated evidence within 90 days exists, return
"No timing evidence" - do not invent urgency.

Return JSON:
{
 "why_now": "one sentence or 'No timing evidence'",
 "why_now_source_url": "",
 "days_since_event": 0,
 "confidence": "High|Medium|Low"
}

Given: agency_type {{agency_type}}, services {{services_list}}, platforms
{{platforms}}, capacity {{capacity_class}}, outsourcing {{white_label_status}},
client verticals {{client_verticals}}, recurring model {{recurring_model}}.

Softvole offers: WordPress development, Figma/design to WordPress, Elementor,
WooCommerce, responsive builds, rebuilds, migrations, maintenance, UI/UX
implementation - delivered white-label under the agency's brand.

Which Softvole service best solves a plausible problem for this agency, and why?
Do not claim a problem exists; state it as a hypothesis tied to evidence.

Return JSON:
{
 "softvole_opportunity": "build|overflow|figma_to_wp|technical_seo|maintenance|migration|none",
 "pain_hypothesis": "one sentence, labelled as hypothesis",
 "supporting_evidence": [""],
 "partnership_potential": "high|medium|low|unknown",
 "partnership_reason": "",
 "confidence": "High|Medium|Low"
}

You are the final qualification step. Use ONLY the structured inputs below;
do not browse. Apply Softvole Master ICP v2 rules: intent never rescues a bad
fit; fit does not imply timing; unknown key fields cap the lead at "review".

Inputs: {{account_json}} {{capacity_json}} {{web_json}} {{person_json}}
{{signals_json}} {{scores}}

Answer each question with answer + evidence reference + confidence:
1 Does the company fit the ICP?
2 Does it sell services compatible with Softvole?
3 Does Softvole solve a plausible operational problem?
4 Is there evidence of development-capacity limitations?
5 Is there evidence of relevant web demand?
6 Is the person relevant?
7 Is there a reason to contact them now?
8 What evidence supports this?
9 What information is uncertain?
10 Contact, review, nurture, or reject?

Return JSON:
{
 "answers": [{"q":1,"answer":"","evidence_ref":"","confidence":""}],
 "ai_icp_fit": "strong|moderate|weak|none",
 "ai_recommendation": "contact|review|nurture|reject",
 "recommendation_reason": "max 2 sentences",
 "unknowns": [],
 "overall_confidence": "High|Medium|Low"
}

Prospect: {{first_name}}, {{title}} at {{company_name}}.
Why now: {{why_now}}. Pain hypothesis: {{pain_hypothesis}}.
Recent web work: {{recent_web_projects}}. Recent posts: {{recent_posts}}.

Produce material for a human to write a first email. Do not write the email.
The hook must reference something specific and verifiable that relates to web
delivery (a project, a hire, a post). Avoid flattery, awards, and anything
that implies we monitored them personally. Never mention negative reviews.

Return JSON:
{
 "personalization_hook": "one factual sentence",
 "hook_source_url": "",
 "relevance_bridge": "how the hook connects to web production capacity",
 "suggested_angle": "margin|capacity|reliability|figma_fidelity|technical_seo|overflow",
 "proof_point_needed": "which Softvole case study/sample would fit",
 "do_not_mention": [],
 "confidence": "High|Medium|Low"
}

§29Personalization data

Collect only what changes the message. If a data point wouldn't change what you write, it's database noise.

§30Messaging by band and offer

§31Clay workflow order (01–22)

The operational sequence. Each step lists what to learn, why, inputs, outputs, what happens next, and what not to do yet. Open a step for beginner and expert notes.

01 Find Accounts → 02 Clean → 03 Cheap Filters → 04 Company Enrichment → 05 Tech/Service Research
→ 06 Capacity Research → 07 AI Account Qualification → 08 COMMERCIAL FIT RESEARCH (price gate)
→ 09 Find Decision Makers → 10 Person Enrichment → 11 Signal Detection → 12 Intent Research
→ 13 AI Qualification → 14 Scoring → 15 Segmentation → 16 Personalization Research → 17 Human QA
→ 18 Instantly → 19 Outreach → 20 Response Tracking → 21 Revenue Feedback → 22 ICP Refinement

§32Test plan — 100 → 500 → 1,000 → production

Do not generate 10,000 leads first. Each phase has exit criteria. Thresholds are starting Hypothesis values.

§33100-account validation framework

The point of the first 100 is to separate four different failures that all look like “they said no”.

What to record for every account

The highest-value field in the whole table is actual_site_price_stated — the number the prospect gives on the call. It is the ground truth that calibrates the entire band model. Ask it in every conversation: “what does a website typically run for your clients?”

The four-way diagnosis

Decision thresholds after 100 accounts

Cohort design for the first 100

40  Tier 1A  design / branding studios, Figma-led, band 3–5, US + UK
30  Tier 1B  marketing & digital agencies, band 4–5, thin dev capacity
15  Tier 1C  B2B / healthcare / multi-location specialists, band 3–5
10  Tier 2-P web & dev agencies, band 4–5, overflow angle (proof-heavy)
 5  Control  band 2 agencies that v2.0 would have called Priority A
             → the control tests whether the commercial gate is real
             or whether Softvole is leaving money on the table

The five-account control cohort matters more than its size suggests: if band 2 agencies convert as well as band 4, the entire premise of v2.1 is wrong and you should know that within a month.

§34Golden dataset

A small, manually labelled set of companies that acts as the answer key for every prompt, filter and score change.

How to use it

Grow it continuously: every Priority A that becomes a qualified opportunity and every QA-caught error is added (with its label) after the fact.

§35Feedback loop

The ICP is a hypothesis that outcomes correct. Optimise for qualified opportunities and profitable long-term partnerships, never for open rate and not primarily for reply rate.

Metric hierarchy

North star:   partner revenue at 6 and 12 months per 100 accounts contacted
Primary:      qualified opportunities per 100 contacted  ·  partnerships won
Diagnostic:   qualified meetings  ·  positive reply rate  ·  proposal rate  ·  win rate
Guardrails:   bounce < 2%  ·  spam complaints ≈ 0  ·  unsubscribe rate  ·  QA correction rate
Ignore:       open rate (unreliable with privacy protection; says nothing about fit)

What to track and how it changes the system

Change rules

• Minimum evidence before changing a rule: ≥ 150 contacted in the segment, or ≥ 5 qualified opportunities, whichever comes first. Smaller samples produce noise.
• Change 2–3 variables per iteration; version the ICP (v2.1, v2.2) with a dated change log entry stating evidence.
• Monthly review: segment table, signal table, lost reasons, QA corrections. Quarterly: re-order tiers and agency-type ranking.
• Win/loss interviews: ask every new partner which trigger made them reply — this validates or kills the “why now” hypotheses.
• Keep a holdout: 10% of qualified Priority B leads contacted without personalization research to measure what personalization actually adds.

§36Future ICP expansion — kept separate

§37Change log — v1 → v2.0 → v2.1

§38Ten things to understand before building in Clay

1. Money gates need. Price Fit is checked before people search. An agency that cannot resell Softvole at 2.5× is not a lead, whatever its capacity pain.
2. Vendor categories lie. 38% of “marketing services” records in the pilot were not agencies. AI agency validation runs before any contact spend.
3. Selling websites is the Tier 1 gate — not being a marketing agency. No web offer means no current web pain.
4. “No developer found” is not “no developer”. Capacity needs positive evidence; otherwise it is U.
5. One overloaded developer beats zero developers. Capacity B and C are the strongest fits; D at band 4–5 is an overflow partner, not a reject.
6. Value beats volume. Four $25,000 builds a year outrank sixty $2,000 sites.
7. “Uses WordPress” is fit; “hiring a WordPress developer” is timing. Different score components.
8. Unknown price data goes to research, never to rejection. Most small agencies publish no pricing; a strict gate would delete the best accounts.
9. Priority A needs human QA plus a written problem hypothesis with a source URL before it enters Instantly. Priority B carries the volume.
10. Judge the system by qualified opportunities and partner revenue after at least 150 contacts per segment — not by opens, and not by the first week of replies.

§39Assumptions and weak evidence

Assumptions inside this model

Where I'd push back on your direction

1. “Premium” has to be operationalised or it's only a price. Band 4–5 buyers will test it: named senior developers, a written QA checklist, accessibility standard, performance budgets, staging and version control, response-time commitments, warranty period, NDA and security posture. If those documents don't exist yet, the premium price is a claim rather than a product, and the first 100 accounts will tell you so through proof objections. This is the single biggest risk to v2.1 — bigger than the ICP itself.
2. The retainer-led counterexample is real. Some agencies knowingly run 1.8–2.0× on builds because the website is the entry point to a long retainer. A rigid 2.5× rule rejects them. That's why the uplift rule exists — but watch it, because it is also the rule most likely to let scope-creep partners back in.
3. Moving Capacity D up may matter more than any other change. v2.0's instinct — agencies with developers don't need you — is commercially backwards. They have budget, process and peak-load pain. They are slow to win, and you are not ready for all of them yet, but writing them off is how you end up selling only to the agencies that can least afford you.
4. SEO agencies are probably an offer mismatch, not a price mismatch. Keeping them in the core list will generate replies that never convert, which will look like a pricing problem in your dashboard. I'd cut them from the core cohort.
5. Your site's claim counters (300+ projects, 40+ agency partners, 5.0 rating) still conflict with the proof gap you describe. At band 4–5 the first thing a buyer does is verify. A mismatch between those numbers and a handful of new reviews will read as fabrication and will cost you the exact accounts v2.1 is designed to win. Fix that before outreach starts.
6. The case-study figures on your site follow suspiciously smooth weekly patterns. If they're illustrative, label them. If they're real, publish enough method detail that a sceptical buyer can accept them.
7. Don't let “no discounts” become “no flexibility”. The pilot, milestone payments, a defect warranty and a fidelity commitment are your negotiation currency. Use them, so price never becomes the only lever available in a conversation.
8. The pool is small. Precision at this price point means your constraint becomes list volume and trust-building, not conversion. Plan outbound capacity accordingly (§40§).

Carried over — ICP assumptions from v2.0

§40Final verdict

What the funnel looks like after the gate

1,000 raw US/UK agency records
  →  620   real, site-verified agencies          (62% — Pilot Evidence)
  →  280   sell websites                          (~45% — Hypothesis)
  →  170   meaningful WordPress share             (~60%)
  →  110   capacity A/B/C, or D with overflow     (~65%)
  →   40   price_fit Strong or Acceptable         (~35% — Hypothesis, the new gate)
  →   30   with a contactable decision-maker      (~75%)
  →  ~30   Priority B pipeline
  →   ~6   Priority A (direct or 2× indirect signal ≤ 90 days)

Two consequences. First, Priority B is your volume engine, not Priority A — at this price point there simply aren't enough signal-bearing, commercially-qualified accounts to run an A-only pipeline. Second, you don't need 10,000 accounts. If a partner is worth $7,500–$30,000 a year, 10–20 active partners is a real business. That's roughly 2,000–4,000 raw records processed properly, not 16,000 — which also cuts the credit budget the sourcing plan assumed.

Is the pricing itself wrong?

On the evidence available: no, with one flag. The build-from-design tiers ($1,200 / $2,400) sit below common onshore white-label project pricing and are the natural wedge. Design + development at $3,500 and WooCommerce at $4,500–$6,500 are consistent with the market. The one number I would still question is care at $599 as a default — the new three-tier structure resolves that, provided Essential and Growth are profitable for you at volume, which I can't verify.

The order of investigation in your own decision principle is the right one, and the data so far supports it: the ICP was the problem (#1), the offer structure needed the dev-only wedge and the pilot (#2), proof is genuinely insufficient (#3), and pricing (#4) is not currently the binding constraint. Nothing here justifies moving Softvole into the $500–$1,500 full-website market.

Conditions for this verdict to hold

1. The commercial gate is implemented as a gate, not as extra points.
2. Unknown routes to research, never to rejection — otherwise v2.1 deletes half your market.
3. Premium is documented as a service standard (QA, accessibility, warranty, SLAs, confidentiality), not asserted.
4. Public claims on the website match verifiable reality before outreach begins.
5. The band model is recalibrated against real stated prices after the first 100 accounts.
6. The five-account band-2 control cohort is actually run, so the premise stays falsifiable.

If, after 100 accounts, band 4–5 agencies convert no better than band 2–3, and the objections are about price rather than proof, then the pricing hypothesis — not the ICP — is what needs revisiting. That is the only evidence I would accept for lowering prices, and it does not exist yet.