Work in detail
How the Salsabeel Real Estate marketing system is built: the tracking, the automations, the agents, the campaigns and the reporting that runs them, plus the creative produced in-house. Facts come from the system documentation; campaign names are hidden and customer data never appears.
One pipeline for every channel. A lead from the website, a Meta instant form, TikTok, Snapchat or an Instagram comment lands in the same CRM within seconds, is assigned to a rep, and reports back to the ad platforms as a server-side event with the same event ID the browser pixel used.
Website forms post to a Cloudflare Worker (Turnstile, honeypot, rate limit, event ID), then to n8n. Meta instant forms arrive through the native GoHighLevel integration. Instagram comments reach n8n through the Meta webhook.
n8n validates, maps attribution fields (platform, campaign, ad set, ad, form, first-touch UTMs), blocks duplicates, tags the project and consent source, opens an opportunity and round-robins the lead to an active rep with a WhatsApp alert.
Lead and browse events go server-side to Meta, TikTok, Snapchat and GA4 with a hashed visitor ID. Pipeline stage changes are polled every 10 minutes and sent as CRM conversion events, so the platforms optimise on qualified leads, not raw form fills.
Nurture suggestions, nudges, a conversation coach, a knowledge bot and an ops console sit on top. Nothing in the system messages a customer automatically; reps send, the system prepares.
Diagram Blue: acquisition. Amber: automation layer (Worker and n8n). Teal: sales and measurement. The dashed line is the feedback loop to the ad platforms.
The browser pixel and the server send the same event with the same event_id, so the platform counts it once. The visitor ID cookie is hashed and sent as external_id, which lifts the match quality. Stage changes in the CRM become conversion events (Lead → Contacted → Interested → Qualified) so the campaigns learn from what sales actually qualifies.
Diagram Browser and server events carry the same ID; the platform keeps one. CRM stage changes feed the same conversion dataset.
Where a lead is stored, routed and followed up. The CRM holds the record; n8n does the work between systems; Zapier and Apps Script fill the gaps.
Leads enter from forms, n8n validates and routes them into the CRM, and stage changes go back out. Silent 10-second loop. Highlights follow the order the work actually happens.
Names only; icons are neutral category marks, not product logos.
Around 45 n8n workflows and 50 data tables run on a self-hosted instance. The ones below are the backbone. Each has a rollback version recorded before any change, and none sends a message to a customer except the Instagram reply, which was an explicit decision.
router_log, intake_config
lead_capi_keys, intake_configcapi_event_log, lead_capi_keys, fact_lead
signal_events, signal_state
price_sheet
fact_ad_daily, fact_lead
budget_engine_log, plan_budgetig_comment_dm_rules, ig_comment_dm_log, ig_dm_suppressmarket_intelFrom the first portfolio. Account identifiers and campaign names are pixelated.
Used to research, build and review, never to talk to customers. Each model has a job it is best at, and a person approves what leaves the system.
Research feeds the build; agents run inside n8n; a person approves anything customer-facing. Silent 10-second loop. Highlights follow the order the work actually happens.
Names only; icons are neutral category marks, not product logos.
Four n8n agents sit above the workflows for questions and analysis. Sensitive paths (intake, router, CAPI, funnel events) stay deterministic workflows. No agent is built or published without approval, and none talks to customers.
Rule of thumb: workflows for anything that must be predictable; agents for questions with no fixed steps. Every agent version is recorded with a rollback ID.
Where the budget is spent. Each platform receives the same server-side events, so all of them learn from qualified leads rather than raw form fills.
Every platform gets the same events; the quality signal loops back into delivery. Silent 10-second loop. Highlights follow the order the work actually happens.
Names only; icons are neutral category marks, not product logos.
Lead-gen for off-plan units in Ajman, bought on a cost-per-qualified-lead basis. Naming is strict (PROJECT | OBJECTIVE | DESTINATION | GEO | MONTH) so reports and the budget engine can match campaigns to projects and targets automatically.
OTP-verified phone plus qualifying questions (unit type, payment method, purpose, timeline). Optimised for quality leads with CRM stage events feeding the dataset. Instagram-heavy placements.
Traffic to the Astro landing pages with pixel + server events, Turnstile, consent texts and first-touch UTMs. Used for the exhibition period and for retargeting.
Measured in the ops console as conversations → CRM → interested with first-reply time, so it is compared on the same Verified CPL basis.
Broad UAE ad sets with Advantage+ placements as the main set; emirate-split ad sets kept only where the data justified them.
Lead-gen forms and website-lead campaigns, with a CRM event set for lead quality and lead IDs synced to the CRM. Kept as a test channel at a small share of spend.
Pixel plus server-side SIGN_UP and a custom event for Qualified+. Measurement only, since Snap cannot optimise on CRM stages. Billed in USD.
Retargeting pools from lead-form openers who did not submit, video viewers and Instagram and page engagers; lookalikes (1–10%) seeded from the CRM lead list, form submitters, engagers and website conversion events. The same CRM-list approach is used on TikTok and Snapchat.
LinkedIn Ads, leasing campaigns and international investor and HNWI targeting come from earlier roles and freelance work, not from this account.
Diagram Both arms share creative and targeting; only the form differs. The tag written by the router is what the weekly report groups by.
A fixed monthly budget covers days 1–20; whatever remains keeps running until spent. The engine applies the same rules every morning, writes the decision to the ad account, and emails a plain-language log. The manager approved autonomous writes after a dry-run period.
Diagram Every morning the same inputs go through the same rules. The output is a budget write, a pause or nothing, and always an email.
Spend ÷ new CRM contacts after removing duplicates, invalid data and rent-only enquiries. The raw platform CPL is shown next to it but never used for decisions.
Qualified or beyond (site visit, proposal, booking, closed) ÷ verified leads, read as a snapshot 14 days after the wave.
First human touch within 15 minutes is the team rule; the console shows buckets and the live no-touch queue.
Invalid share, rent-only share, not-a-buyer share and the Meta↔CRM gap, so a cheap lead cannot hide a bad one.
How the site is built and deployed, how data is collected, and where it is read.
Tags and scrapers collect, the site and GA4 run, dashboards read. A GitHub push deploys the site. Silent 10-second loop. Highlights follow the order the work actually happens.
Names only; icons are neutral category marks, not product logos.
Reporting is built from the company’s own data, not agency benchmarks. The ops console reads a single summary endpoint (Worker → n8n) and renders marketing and sales end to end; the weekly email and per-rep reviews come from the same tables.
Spend and pacing, unique leads, blended and Verified CPL, interested and visits, speed-to-lead buckets with a live no-touch queue, funnel by channel, budget pacing per channel and project, campaign → ad set → ad explorer, click-to-WhatsApp tiles, rep scorecards, lost reasons, data quality and targets. Login through Cloudflare Access.
Monday 09:00: what moved and why (≥20% changes), what changed in the accounts (activity logs), pacing, week priorities, permit tracker and the A/B section.
Notes, stage changes, unanswered inbound, stuck leads and a score with tips, per salesperson, from CRM data only.
Budget close-out, cost per interested lead, best campaign, spend vs plan and next-month plan parameters.
An in-house marketing team still needs films, renders, voice and layouts every week. These tools cover that, with a conventional edit at the end.
A photo and a script become generated shots and voice, then one edit produces every format. Silent 10-second loop. Highlights follow the order the work actually happens.
Names only; icons are neutral category marks, not product logos.
Three projects to sell and no retained creative or ad agency. Films, renders, voiceover, brochures, print and the exhibition kit were produced in-house with generative tools and a conventional edit, under the approved facts file and the brand’s navy-and-champagne palette. The one exception was the Ajman real estate investment exhibition, where the company hired event photography firms and an exclusive influencer to cover the show days only.
Exterior renders for six company buildings from site photos and plans, generated and upscaled in rounds (corner, front, low-angle), then used on the website, brochures and the tender portfolio.
Area and project films from real footage plus Veo 3.1 (Google Flow) and Wan 2.1 (ComfyUI) shots; a construction time-lapse from a single site photo; Google Earth Studio map pins; an AI presenter pilot in Emirati dialect with ElevenLabs stock voices.
LED-wall films in 3:1 and 2400×800, a 6:5 version for a 3 × 2.5 m stand screen, a hologram loop, a printed 2 × 2 m stand graphic, an on-screen apartment-draw sequence with printed coupons and bilingual registration.
A 56-page bilingual brochure, a pre-qualification portfolio for a government tender, payment-plan PDFs and the generated BOF payment sheet.
Ad creatives, exhibition visuals and LED-wall material picked from the working archive. Phone numbers are pixelated; previews are short muted excerpts.
Carried over from the earlier portfolio. Films are 8-second muted previews that play only when tapped; full versions open on the old site.
Arabic exhibition brand voiceover, ElevenLabs, 31.6 seconds. Plays only when you press play.
Open the brochure PDF (5.4 MB)Enjaz Salsabeel project brochure, 56 bilingual pages. Archived design sample; prices and plans inside are not a current offer.
Open the study PDF (6.0 MB)Al Hamra corporate presentation: an independent personal design study, 34 pages. Not commissioned, endorsed or approved by Al Hamra, and not evidence of employment or a client relationship.
salsabeel-re.ae was rebuilt from WordPress to a static Astro site on Cloudflare Workers with a small Worker API, bilingual pages for every project, a blog for answer-engine visibility and the tracking described above.
Numbers below come from the ad accounts and the CRM for August–October 2026 at Salsabeel. Campaigns are labelled A–D and CPL is indexed (best week = 100) because the absolute figures are the company’s. Career totals are at the end.
| CPL index | |
|---|---|
| Campaign A · Aug · W1 | 192 |
| Campaign A · Aug · W2 | 266 |
| Campaign A · Aug · W3 | 370 |
| Campaign B · Sep–Oct · W1 | 165 |
| Campaign B · Sep–Oct · W2 | 100 |
| Campaign B · Sep–Oct · W3 | 115 |
| Campaign C · Oct · W1 | 207 |
| Campaign C · Oct · W2 | 716 |
| Campaign D · Oct · W1 | 377 |
| Campaign D · Oct · W2 | 1008 |
| CTR | |
|---|---|
| Campaign A · Aug · W1 | 1.21% |
| Campaign A · Aug · W2 | 1.05% |
| Campaign A · Aug · W3 | 0.89% |
| Campaign B · Sep–Oct · W1 | 1.32% |
| Campaign B · Sep–Oct · W2 | 1.23% |
| Campaign B · Sep–Oct · W3 | 1.03% |
| Campaign C · Oct · W1 | 0.62% |
| Campaign C · Oct · W2 | 0.53% |
| Campaign D · Oct · W1 | 0.69% |
| Campaign D · Oct · W2 | 0.53% |
| Share of leads | |
|---|---|
| Raw leads | 100% |
| Interested+ | 8.8% |
| Qualified+ | 4.7% |
| Site visit+ | 3.8% |
| Qualified+ rate | |
|---|---|
| Meta (78.8%) | 4.3% |
| TikTok (7.7%) | 5.9% |
| Snapchat (7%) | 3.2% |
| Direct / WhatsApp (4.1%) | 5.6% |
Career figures are cumulative across employers since 2018, reported from ad-platform and CRM data. Monthly lead volume at Salsabeel is lower and is governed by Verified CPL rather than volume. Agency and freelance (2023–2025): campaign work, including leasing, for a UAE real-estate marketing agency serving several developers and brokerages.
The whole stack, or one piece of it. WhatsApp is fastest.