From Website Intent to 1:1 Personalization with AI Agents

On this episode of Knocking to 10, John DiLoreto sits down with Marwan "Mars" Aziz, founder of sixtysixten, to turn AI agents from buzzword into workflow. Below, explain what agents are and how to build them, then walk through a Knock2 → Clay flow that auto-generates 1:1 WordPress/Webflow landing pages from website intent.

1) Agents 101: What they are and how to build one

What’s an AI Agent (in GTM terms)?

• Classic automation = rigid if/then rules → fixed inputs/outputs.
• AI agent = a goal-driven workflow that thinks with an LLM, uses tools, and returns variable outputs based on context.
• In practice, agents:
  • Read/contextualize inputs (visitor, account, meeting transcript, page history).
  • Call tools and data (CRM, Knock2, web scraping, calendars, docs).
  • Produce structured outputs (JSON, updated CRM fields, a page, a task).

Where to start: pick outcomes, not tech

1. Shadow the reps (2–3 hrs). Watch the click-by-click reality. Identify repetitive research, note-taking, CRM updates, and follow-up tasks.
2. Choose a single outcome (save 2 hours/week per rep or add 5 SQLs/month) and work backward.
3. Crawl → Walk → Run (ship value fast):
   • Crawl: one job per agent (e.g., ICP filter on a 10k lead list).
   • Walk: chain jobs (ICP → research → draft copy).
   • Run: orchestration + human-in-the-loop edits + auto-routing.

Split agents (don’t make a mega-prompt)

• Better quality when you separate jobs:
  1. ICP Detector (binary/score + reason).
  2. Researcher (company/person/industry signals; outputs JSON).
  3. Writer/Assembler (headlines, bullets, CTA from the research JSON).
• Tip: give models room to think (higher max tokens, explicit structure, strict JSON schemas).

Two proven agent patterns from the episode

• AI Battle-Card Generator — Auto-preps reps before meetings: pulls the calendar, researches the account & personas, outputs competitor context, value angles, and exact talk-tracks.
• Post-Meeting Analysis Agent — Consumes the call transcript (e.g., Fireflies), extracts pains, next steps, objections, tone/sentiment, and writes back to CRM in the format your team uses.

Recommended stack

• No/low-code: Clay (tables, enrich, LLM steps), Zapier/Make (routing), Google Sheets/Notion/Airtable (light DB), WordPress or Webflow (templates).
• Models: Small fast models for classification (e.g., ICP), stronger models for research/copy. Use function-style prompts with tight schemas.

Guardrails & etiquette

• Don’t be creepy. Use Knock2 (website visitor) data thoughtfully—delay outreach 24–72 hours after the visit. Reference needs, not surveillance.
• Human-in-the-loop. Reps can approve/edit outputs (that signal trains v2 prompts).
• Measure. Time saved, reply rate, meeting rate, pipeline created.

2) Build the Agent: Knock2 → Clay → Personalized Landing Page

Goal: When a qualified visitor hits your site, generate a 1:1 custom landing page that mirrors their interests and brand—then use it in thoughtful follow-up.

End-to-end flow

1. Knock2 Webhook fires on new identified visitor.
2. Clay table ingests the payload (person & account + pages viewed).
3. Agent – Researcher enriches with company/person context and extracts brand cues.
4. Agent – Assembler writes: headline, sub-headline, value bullets tied to the pages they viewed.
5. Dynamic page is created/updated at a unique URL (e.g., /p/{record_id}).
6. Distribution: Write the URL back to CRM; send to Slack; use in outreach after a short delay.

Step-by-step build

A) Wire up the Knock2 → Clay ingest

• In Knock2: create a Webhook for “New Identified Visitor.”
• In Clay: create a table Website_Visitors with columns:
  • record_id (string)
  • person_name, person_title, person_linkedin
  • company_name, company_domain, company_linkedin
  • pages_viewed (array or CSV)
  • visit_datetime, utm_source (optional)
  • priority (low/med/high)
• Map inbound webhook fields to these columns. (Adjust naming to your exact Knock2 payload.)

B) Agent — Researcher (Clay LLM step)

Purpose: Pull brand cues + use-case signals to personalize the page.

Input: Clay row + light web context (company site/LinkedIn).

Output (JSON):

{
 "logo_url": "",
 "brand_hex": "#000000",
 "industry": "",
 "use_case_focus": ["lead scoring", "routing"],
 "evidence": ["Visited /lead-scoring", "Case study: Fintech"],
 "tone": "confident, concise"
}


Prompt:

Research this company/person enough to personalize a GTM landing page.
Return STRICT JSON only. Infer brand color from site if obvious; else "#111111".
Extract up to 2 use_case_focus terms tied to the pages viewed. Include neutral tone.


C) Agent — Assembler (Clay LLM step)

Purpose: Generate the on-page copy from the research JSON.

{
 "h1": "",
 "subhead": "",
 "bullets": ["", "", ""],
 "cta": "",
 "seo_slug": ""
}


Prompt:

You write conversion copy for a 1:1 ABM landing page.
Using the Research JSON, produce: h1 (≤12 words), subhead (≤18 words), 3 value bullets tied to the visitor’s interest, and a concise CTA.
Style: specific, non-generic, no buzzwords. Mention our product only where relevant.
Return STRICT JSON per schema.


Pro tip: Splitting the jobs reliably beats one mega-prompt. Give the writer step a higher token limit. Keep temp low for on-brand consistency.

D) Build the dynamic landing page (WordPress/Webflow)

Template approach (simple + fast)

• Create a single page template with placeholders for h1, subhead, bullets, logo, and background color.
• Add a querystring param ?id={record_id}.
• A tiny script calls a serverless endpoint (or Clay View) to fetch the JSON for that record_id and injects the content.

Nice-to-have: a lightweight editor (small admin UI) that lets reps tweak the h1 or CTA; write edits back to Clay/DB to improve future prompts.

E) Write back & distribute

• CRM update: attach landing_page_url and use_case_focus to the contact/account. (ICP/fit/score live in Knock2.)
• Slack (or Knock2 workflow): post a card with page URL, reason for fit, and suggested outreach.
• Delay send 24–72 hours before first touch. (Polite, not creepy.)

F) Outreach snippet (steal this)

‍Subject: A quick page we built for {{Company}}

Hey {{First}},

Noticed interest in {{use_case_focus_1}}—so we mocked up a 1:1 page showing exactly how we’d help:
{{landing_page_url}}

If it’s useful, happy to compare notes and tailor it further. If not, I’ll get out of your hair.

– {{Rep}}


G) QA & metrics

• QA: Does logo render? Brand color legible? Headers specific (not generic buzzwords)?
• Metrics: Page CTR, reply rate, meetings booked, pipeline influenced. Compare personalized vs control.
• Iteration: Capture rep edits → feed back into the Assembler prompt; keep examples library.

3) Who is Mars Aziz & sixtysixten (and how to reach him)

Mars Aziz is a Revenue Systems Architect and the founder/CEO of sixtysixten. With a mechanical-engineering background and years as a business analyst, Mars built a practice around designing and implementing GTM systems that teams can run themselves—versus one-off campaign work.

What sixtysixten does:

• Deep-dive mapping of your sales/marketing processes.
• Design + implementation of AI agents and automations (Clay, Zapier/Make, custom code).
• Patterns they ship often: AI battle-cards, post-meeting analysis agents, ICP filters, and personalized experiences (like the landing-page agent above).
• Handoff, training, and support so your team can own it long-term.

Get in touch:

• LinkedIn: Marwan “Mars” Aziz
• Web: sixtysixten.com
• Open-door policy—he’s happy to trade notes with founders and GTM leaders.

Want our templates?

We’ll package the Clay prompts, JSON schemas, and a WordPress/Webflow starter. Ping us and we’ll connect you with Mars.