252 Singapore DTC prospects, one evidence-gated pipeline: 146 near-term A+/A targets flow through nine stages before a single email sends

The pipeline, end to end

Nine stages left to right. Blue = evidence work inside Keenable; orange = hard gates that can suppress a prospect; dark = external execution (scanners, verification, sequencer). Hover a stage for detail.

What a11 actually sells — and why the fit model must be redesigned

a11 Studio's current offer (a11studio.pro) is 16 static ads per month, four per week, copy included, unlimited revisions, delivered through WhatsApp — plus a free redesign of one stale ad and a 72-hour CTR refund test. No video, no ad management, no engineering remediation. The source report's creative-service fit model and the user's examples (engineering hires, design stacks, WCAG, UI performance) therefore shift to supporting conversion context. The principal commercial signals become: observable paid-social creative need, promotion/launch cadence, visual-product suitability, repeat-purchase economics, contactability, and evidence quality.

The 100-point model: evidence in, score out

Weights of the eight components; a component scores above zero only when a sourced, dated evidence quote exists. Engineering hires and accessibility gaps live inside the friction/capacity components, never as primary weight.

Three persona routes

Keenable MCP enrichment patterns

Keenable SELECT is read-only DuckDB over WEB_SEARCH/WEB_FETCH and saved result sets — the evidence and extraction layer. Lighthouse/PageSpeed, axe-core, email verification and sequencer writes all run outside it, then merge in the warehouse or Clay.

-- Stage 2–3: identity + commercial signals, ≥6 diverse queries per WEB_SEARCH, batched by company
SELECT url, title, snippet FROM WEB_SEARCH(queries => [
  '"{company}" site:{domain}', '{company} Singapore promotion sale',
  '{company} new launch product', '{company} subscription bundle',
  '{company} careers hiring marketing', '{company} founder OR "head of growth" linkedin']);

-- Stage 3b: fetch official pages, extract one company-level object
SELECT SEM_EXTRACT(content, 'offer, promotions, subscription_or_bundle,
  decision_maker_name, title, public_business_email, public_whatsapp,
  recent_company_milestone — null when absent') AS company_obj
FROM WEB_FETCH(urls => [home, product, promo, careers, accessibility]);

-- Repeated facts: launches, jobs, products
SELECT SEM_EXTRACT_ALL(content, 'each product launch: name, date, url') FROM WEB_FETCH(...);
-- Relevance gate  WHERE SEM_MATCH(snippet, 'DTC physical visual product sold online in Singapore')
-- Canonical dedupe: SEM_NORM only inside GROUP BY; final GROUP BY yields one row per company
GROUP BY SEM_NORM(company_name)

-- Stage 7: one gated company row out
SELECT prospect_id, company_name, canonical_domain, icp_score, score_band, persona,
       specific_pain_hook, evidence_url_1, evidence_quote_1, confidence_1, ready_to_send
FROM merged WHERE evidence_url_1 IS NOT NULL AND observed_at >= freshness_floor
  AND suppress_reason IS NULL;

Payload schema — every claim carries URL, quote, date, confidence

{ "prospect_id":"sg-0001", "company_name":null, "canonical_domain":null, "website_url":null,
  "source_tier":"A", "original_score":null, "niche":null, "country":"SG", "source_url":null,
  "verification_status":null, "observed_at":null,
  "ecommerce_platform":null, "cms":null, "analytics_tags":null, "ad_pixels":null,
  "page_builder":null, "checkout_provider":null, "design_system_evidence":null,
  "lighthouse_mobile_performance":null, "LCP_ms":null, "INP_ms":null, "CLS":null,
  "wcag_automated_fail_count":null, "wcag_issue_types":null, "tech_stack_gap":null,
  "recent_launch":null, "recent_promotion":null, "subscription_or_bundle":null,
  "product_count":null, "recent_hiring_signal":null, "funding_or_revenue_signal":null,
  "meta_ad_activity":null, "active_ad_count":null, "latest_ad_date":null,
  "decision_maker_name":null, "title":null, "persona":null, "linkedin_url":null,
  "public_business_email":null, "public_business_phone":null, "public_whatsapp":null,
  "email_validation_status":null, "recent_company_milestone":null, "specific_pain_hook":null,
  "evidence_url_1":null, "evidence_quote_1":null, "confidence_1":null,
  "score_components":{}, "icp_score":null, "score_band":null,
  "suppress_reason":null, "ready_to_send":false }

Null is required when absent. Never inferred: employee count, revenue, ad spend, compliance status, personal email. active_ad_count only when directly observed.

Dynamic-variable contract

Three templates, under 75 words rendered

Base lengths 58, 55 and 56 words; variable budgets (milestone ≤18 words, hook ≤22) keep rendered totals under 75. Lint after render; route 3 falls back to route 2, then 1.

Execution: Clay orchestrates, one sequencer sends

Clay is the orchestration/QA table. Pick one sequencer — Smartlead or Instantly — not both initially. Sync only ready_to_send=true rows; prospect_id is the idempotency key; one campaign per persona × priority; upload as drafts; seed-test; activate only after QA; webhook replies/bounces/unsubscribes back to the master suppression table. Append-only audit fields; never overwrite source evidence. Confirm current account-specific docs and rate limits before production.

Smartlead (api_key as query parameter)
POST /campaigns/create?api_key=KEY               -- create campaign
POST /campaigns/{id}/sequences?api_key=KEY       -- save sequences: [{seq_number, email_body, delay:{days}}]
   body supports double-curly custom fields: {{first_name}}, {{specific_pain_hook}} ...
GET  /campaigns/{id}/leads?api_key=KEY           -- leads return custom-fields object, max 100/page

Instantly API v2 (Bearer / API key)
POST  /api/v2/leads                               -- create lead
PATCH /api/v2/leads/{id}                          -- update custom variables payload
POST  /api/v2/campaigns · PATCH /api/v2/campaigns/{id} · POST /api/v2/campaigns/{id}/activate
POST  /api/v2/campaigns/{id}/variables            -- campaign variables

Deliverability and Singapore compliance

Measurement loop

Track delivery, bounce, complaint, unsubscribe, positive reply, meeting, opportunity and client — sliced by persona, score band, hook and source confidence. Hold out 10–15% per high-volume segment, or A/B exactly one element at a time; never claim significance on tiny samples. Feed reply labels and outcomes into the weights monthly with versioned scoring_model_version and message_version.

Two-week sprint and launch checklist

Evidence base — all 31 source rows

Several rows are duplicate official URL variants of the same document; treat those as corroborating documentation, not separate independent sources. No URL backs more than half the rows.