Four brands · 2026Ongoing

Ads & analytics

Meta ads for watch, apparel, real estate and EV brands, and the measurement behind them: reports that check their own numbers, leads that reach sales in seconds, and audits that follow the money past the click.

My role: Ran the ad accounts with the team and owned the numbers: structure, launches, reviews, reporting, pipelines and audits

1,603

Meta leads for a real estate developer, April to October, at about ₹183 ($2) each

10 to 12x

Verte's Shopify sales against Meta spend, June to August

698

leads delivered straight into sales sheets since 22 July

analytics reports delivered, each as HTML and a one-page PDF

The brief

The agency runs Meta ads for Verte and Alexandre Christie (watches), Season Earth (linen menswear), THS (real estate) and HarboCharge (EV charging), with no in-house performance marketer.

The job: run the accounts, and know what is really working. That means Shopify orders, leads the sales team can call and site visits they can book, not the purchases an ad platform claims for itself.

What I did

Account structure, launches, budget proposals and kill or keep reviews for every brand.
A reporting standard, and 36 reports built on it for Verte, Alexandre Christie, Season Earth, THS and prospects.
Lead and abandoned-cart pipelines that put every lead in front of the sales team within seconds.
Audits that follow results past the click: leads to site visits, failing funnels, and our own change logs.
Rebuilt the customer lists behind four brands' lookalike audiences.
Market research: a repeatable way to size competitors from the ad libraries, and a festive-season brief on the Indian ad market.
Tools: SparrowOS analytics, MCP servers that let Claude query SparrowOS and Reddit Ads, and AdPilot, an assistant where the engine proposes, a person confirms and code executes.

How I did it

A standard before a verdict

Ads are judged on the funnel the platform actually measures (reach, frequency, CTR, CPM, clicks, spend and days live), benchmarked per market, since US CPMs ran about 13 times India's. Attributed ROAS can confirm a call but never leads one. Ads are judged at the ad-set level, because an ad with no sales is often the first touch that makes the next one sell.

Structure that keeps learning

One ad set per product per market, with location as a hard rule. No verdict until days live and spend give enough evidence. Market tests are built as paused drafts for approval, so nothing goes live by accident.

Reports that check themselves

Orders come from Shopify, never from an ad platform. A script re-derives every figure in a report and checks it against the document before it ships, and every report goes out as HTML plus a single-page PDF.

Leads in seconds, not days

Meta lead ads arrive through a webhook whose signature is checked, get answered fast, then are written to the brand's sheet from a queue with retries. A daily sweep reconciles anything missed. 698 leads have gone through it since 22 July.

Follow the money past the click

For THS, 90 days of leads were traced to site visits campaign by campaign. For Season Earth, a one-click checkout widget skipped the cart entirely, so add-to-cart was the wrong thing to measure; most people saw an ad only once, and broken tracking had starved the retargeting pool. For Verte's US Reddit test, three weeks of spend had produced no sales by Shopify's count or Reddit's own, and the rebuild got pass marks set in advance.

Fix the audiences under everything

Every brand's lookalikes turned out to be built from one shared list that was almost entirely one brand's customers. I rebuilt each brand's seed list with a spend value for value-based lookalikes, plus a deduplicated pool of 16,919 people across brands.

The first page of a Verte monthly report with every number, name and finding blocked out in solid black. Only the period, a few labels and the title What changed in September remain. — One of the 36 reports, as it went to Verte. Every number, name and finding was taken out of the page before this picture was made, so nothing is left to recover. The titles that remain only say what was being looked at.

The end of the same report, also blocked out: five numbered recommendations and a section called How we measured, with only those titles left. — One of the 36 reports, as it went to Verte. Every number, name and finding was taken out of the page before this picture was made, so nothing is left to recover. The titles that remain only say what was being looked at.

What it changed

160 → 300+

Verte orders a month: best month before, then July to September 2026

~₹183

per lead ($2) across 1,603 THS leads, April to October

~₹126

per lead ($1.30) for Beyond Horizon's launch campaign, across 91 leads

154

pre-launch registrations for Alexandre Christie's Cosmic watch

THS leads since the pipeline went live, 22 July to 6 November

693 of 698 leads reached the sales sheet within 30 seconds. Half took 11 seconds or less.

6 Nov · 698 leads

In the sheet within 30 seconds: 693Took longer: 5

Of the 5 that took longer, 2 took a few minutes and the nightly sweep caught the other 3, hours later. One dot is one lead, and all it holds is a date. Names and numbers stay in the client's sheet. From the pipeline's own log, test leads removed. Meta's own count for 23 July to 6 November is 692.

Each brand's decisions now rest on numbers that reconcile with its own sales: Shopify orders for the stores, sheet rows and site visits for real estate. When the ads aren't the problem, the reports say so and point at what is.