Twelve selected projects

The work.
And the evidence.

Twelve pieces worth sharing, chosen from a longer list. Each one says what I did and what changed, and shows the evidence for it. Three parts, then how I work with AI and automation.

12 projects

Part one. The click

Fewer campaigns. More from each.

Five campaigns across Meta and Google. I set the budget, chose the channels and ran the accounts myself.

What the campaigns brought in.

leads
27,000+
conversations started
17,000+
calls
4,000+
website enquiries
4,000+

More than twice the leads, each at 64% lower cost, on a smaller budget, with 15 campaigns in place of 52.

Platform records · Rounded down
Meta Ads reporting totals: 27,027 leads, 17,367 messaging conversations started and 4,017 phone calls placed.
Redrawn from Meta Ads reporting: every campaign that had delivery, the longest range Meta offers (11 Oct 2023 – 11 Nov 2026). See the platform screenshot
Google Ads conversion actions, all time, with two thank-you-page form actions of 4,054 and 77.
Redrawn from Google Ads: all conversion actions, all time (12 Mar 2024 – 11 Nov 2026). The two thank-you-page actions are the website enquiries counted: 4,054 and 77. See the platform screenshot

01Meta lead campaigns

Twice the leads. Less than half the cost.

More than 27,000 leads. Across the Meta lead campaigns, leads more than doubled and the cost of each fell 64%, on a smaller budget.

Campaign planning, launch, management and optimisation

Platform-reported result Open the full case
Cost per instant-form lead for one campaign: AED 5.64 at launch and AED 2.43 in January–October 2026, 57% lower.
Redrawn from Meta Ads Manager’s date comparison for one campaign: cost per instant-form lead, 1 Jan – 31 Oct 2026 against 22 May – 9 Jul 2025. Change calculated by Meta. See the platform screenshot
Bar chart: cost per lead AED 10.65 in the first twelve months and AED 3.81 in the latest twelve; leads indexed 100 and 219.
My analysis of exported Meta data: all campaigns with the Leads objective, first twelve complete months on record and latest twelve. Lead counts and spend withheld. Not a platform screenshot.

02Meta WhatsApp campaign

72% lower cost per conversation.

The cost of starting a WhatsApp conversation fell from AED 5.35 to AED 1.52. Across the account, messaging conversations grew 3.7 times.

Campaign planning and management

Platform-reported result Open the full case
Cost per WhatsApp conversation started for one campaign: AED 5.35 in September 2024 and AED 1.52 in December 2024–January 2025, 72% lower.
Redrawn from Meta Ads Manager’s date comparison: cost per WhatsApp conversation started, 1 Dec 2024 – 31 Jan 2025 against 1–30 Sep 2024. Change calculated by Meta. See the platform screenshot
Weekly cost per conversation started, September 2024 to January 2025: about AED 11 to 13 in the first weeks, then under AED 5.
Traced from Meta Ads Manager’s weekly chart of cost per conversation started, same campaign, 1 Sep 2024 – 31 Jan 2025. Close to Meta’s line, not exact. See the platform screenshot

03Meta call campaigns

From AED 17.46 a call to AED 3.30.

Four call campaigns, one after another. Each cost less per call than the one before: AED 17.46, then 10.26, then 3.91, then 3.30.

Campaign planning and management

Platform-reported result Open the full case
Cost per call placed for four call campaigns in turn: AED 17.46, 10.26, 3.91 and 3.30.
Redrawn from the Meta Ads Manager campaign table: cost per call placed, four successive call campaigns, 1 Aug 2025 – 31 Oct 2026. See the platform screenshot

04Google Search campaigns

40% lower cost per enquiry in Abu Dhabi.

A separate Search campaign for each region. Two of the three became cheaper per website enquiry, Abu Dhabi by 40%.

Campaign setup, conversion goals and management

Platform-reported result · Mixed by region Open the full case
Cost per conversion and change for four Google campaigns: Performance Max 45.44% lower, Search Abu Dhabi 39.84% lower, Search other emirates 16.44% lower, Search Dubai 11.44% higher.
Redrawn from Google Ads’ date comparison: cost per conversion, 1 May – 31 Jul 2025 against 22 Mar – 23 Apr 2025. The three Search rows are Dubai, Abu Dhabi and the other emirates. See the platform screenshot
Line chart of monthly cost per website form for the Dubai, Abu Dhabi and other-emirates Search campaigns, March to July 2025.
My analysis of exported Google Ads data: monthly cost per website form by region. Not a platform screenshot.

05Google Performance Max

45% lower cost per website enquiry.

A paused Performance Max campaign, reactivated and restructured. The cost of a website enquiry fell 45%.

Campaign restructure, assets and management

Platform-reported result Open the full case
Cost per conversion and change for four Google campaigns: Performance Max 45.44% lower, Search Abu Dhabi 39.84% lower, Search other emirates 16.44% lower, Search Dubai 11.44% higher.
Redrawn from Google Ads’ date comparison: cost per conversion, 1 May – 31 Jul 2025 against 22 Mar – 23 Apr 2025. The first row is the Performance Max campaign. See the platform screenshot
Line chart of monthly cost per website form for the Performance Max campaign, with dated changes marked.
My analysis of exported Google Ads data: monthly cost per website form, with dated changes. Not a platform screenshot.

Part two. The enquiry

Know what a lead is worth.

Four tools, each built to answer one question. Where do enquiries go? What needs attention today? Which leads were worth their cost? Where is this shipment?

06CRM reporting

Where did the enquiries go?

Where do enquiries come from, where do they stall, and which are won? One reporting tool answers all three. I set the rules, built it with AI assistance and kept it in use.

Measurement rules, report design and AI-assisted implementation

Tool demonstration · Synthetic data Open the full case
CRM dashboard with a lead funnel and a chart comparing each source’s share of enquiries and wins.
Demonstration with synthetic data. The dashboard compares lead stages and shows how each source contributes to enquiries and wins.
A table of conversion rates by source and salesperson, followed by a chart of monthly rates.
Demonstration with synthetic data. The report groups conversion rates by channel and salesperson and leaves small samples blank.

07Odoo KPI dashboard

A clearer view of the day.

What needs attention today? One screen for the day’s follow-ups, pipeline, bookings and collections. I defined the KPIs and configured it in Odoo.

KPI definitions, dashboard configuration and filters

Reconstruction · Fictional values Open the full case
Recreated grid of KPI tiles for sales, follow-ups, pipeline stages and warehouse activity.
Reconstruction from the historical Odoo tile definitions. The layout is simplified, the counts are fictional and revenue values are withheld.

08Lead-quality analysis

Look past the cheapest lead.

The cheapest lead is not always the best one. I measured lead quality alongside cost and followed enquiries through to booking and collection.

Lead-quality measures, time-to-stage analysis and lost-reason grouping

Method illustration · Invented figures Open the full case
Illustrated lead-quality measures, weekly chart, time-to-booking examples and grouped lost reasons.
Methodology illustration with synthetic data. The example qualification rule and figures are illustrative.

09Inventory portal

Find it. Check it. Move it.

Where is this shipment? A portal for finding cargo across warehouses, checking mismatches and planning loads. I designed it and built it with AI assistance. Colleagues used it for real work.

Requirements, matching rules and AI-assisted implementation

Tool demonstration · Fictional bookings Open the full case
Inventory locator with booking tiles, warehouse labels and hold flags.
Demonstration with fictional bookings. The locator shows where each booking’s pieces are stored and flags holds and bookings split across sites.
Example loading plan with a hold alert and pick sheets grouped by warehouse.
Demonstration with fictional bookings. The tool prepares a pick sheet for each warehouse and flags bookings on hold.

Part three. The next step

Give interest somewhere to go.

Two websites and a nutrition app, each directed from the first idea to a working result.

10FitCalorie landing page

Start with something familiar.

A launch page that starts with food people already know, and gives them a clear route to early access.

Positioning, page narrative and early-access journey

Actual website capture · 11 Nov 2026 Open the full case
Actual FitCalorie landing page with the headline Built For How Indians Actually Eat and a sample app screen.
The FitCalorie website as it stood on 11 November 2026. This is the website case; the app is a separate product project.

11FitCalorie app

From a question to a working product.

A nutrition app taken from an idea to working Android and iOS builds and internal testing. I directed it; AI assisted the build.

Product vision, requirements, prioritisation and hands-on validation

Historical build milestone · Internal testing Open the full case
FitCalorie product screen showing food and serving information.
Product-development screen from the FitCalorie records. A historical working-build example, not a claim of current public availability.

12Chroma Creative website

A brand story you can move through.

An agency’s positioning turned into a website you move through, with a direct route to the work, the services and an enquiry.

Narrative, website direction and desktop/mobile review

Actual website capture · 11 Nov 2026 Open the full case
The Chroma Creative website, opening on its storefront scene.
The Chroma Creative website as it stood on 11 November 2026. Shown as project evidence.

How I work

If the tool doesn’t exist, I build it.

AI and automation do the building. The rules, and the final word, are mine. Three examples, each with its record.

Automation

Leads file themselves.

A lead from a Meta form, the website or a chatbot lands in a shared sheet within seconds. Nobody copies it across.

Automation design and setup, in Latenode

Redrawn from platform screenshots · 11 November 2026 · Runs estimated from a month of usage, at one credit a run
automations live
5
months running
16
runs in the last month
~120
Latenode scenario "Website to Google Sheets": a webhook trigger feeding a Google Sheets row, with seven successful runs on 8, 9 and 10 November 2026.
Redrawn from Latenode, “Website to Google Sheets”: a website form submission becomes a row in a sheet. Seven runs on 8–10 November 2026, all successful, each in two or three seconds. See the platform screenshot
Latenode scenario list showing seven scenarios, five active, including Meta Leads to Google sheets, Website to Google Sheets and Manychat to google sheet.
Redrawn from the Latenode scenario list, 11 November 2026: seven scenarios, five active, the oldest from July 2025. See the platform screenshot

AI-assisted building

Three tools. No development team.

A CRM report, a warehouse portal and a nutrition app. I wrote the rules and tested every result; AI did the building. This website was made the same way.

Requirements, rules, review and testing

Each tool has its own case and evidence label
tools built and put to use
3
report tabs from one export
14
app platforms
2

AI agents

A team of AI agents. I sign off everything.

AI agents work on FitCalorie’s marketing, each with one job and written rules. A checker reviews every post before I see it, and nothing goes out without my approval.

Roles, rules, skills and every approval

From the office’s own records · 13–21 October 2026 · Screens from a paused copy
agent runs, every one logged
161
drafts the checker sent back
28
tasks in nine days
21
The FitCalorie Office: a 3D office with a founder’s room, a marketing studio with five agents at their desks, a shared memory at the centre and a live activity feed.
The office, paused. The founder’s room, the marketing studio with its agents, the shared memory they all read, and a live feed of every hand-off. Captured 11 November 2026 from a paused copy of the office records.
An approved task in the FitCalorie Office: a three-slide dosa carousel, with a history showing the designer’s draft, the checker’s send-back and pass, and the founder’s approval.
One post and how it reached me: the designer’s draft, the checker’s send-back and pass, then my approval and scheduling. Usage cost hidden.
A decision in the FitCalorie Office inbox, with Approve, Ask for changes and Reject buttons above the checker’s open points and a summary of the work.
A decision waiting for me. The checker’s open points come first; then I approve, ask for changes or reject. Of 21 tasks I approved 8 and rejected 11. Usage cost hidden.

A closer look

Open full size