Daniel ConstantGrowth for e-commerce, lead generation and digital products · performance media · tracking · AI automation · Warsaw, Poland

US$4.5M in paid media over the last 12 months, run with tools I built

Google US$1.90M + Meta US$2.56M + TikTok US$0.11M, read from the platforms on 17 Nov 2026 (Meta is a floor). I am a performance marketer with a developer's background (COBOL, now PHP/Laravel and JavaScript). I run paid media for e-commerce brands, lead-generation businesses and digital products, set up their tracking, and build the tools that run the work: a lifecycle CRM with AI-written emails, a measurement panel for online stores, and agent-run optimisation routines.

US$5.17M
Google Ads spend under my manager account since Dec 2015, 35 accounts with delivery E1
US$3.29M
Meta spend in 25 months across 30 ad accounts with delivery (floor) E3, E4
US$152k
net revenue in the first 12 months of a digital product, from zero, 8,506 sales E15
168
Tag Manager containers across two logins, plus 84 GA4 accounts E5
6.5×
organic clicks for a wellness brand after an SEO + GEO blog program I ran E6–E8
822
commits in 3 months building Skarb, a lifecycle CRM now live with merchants E9–E12

Verified spend vs. value returned (platform-reported)

US dollars. Bar length is to scale within this chart.

Ad spendConversion / purchase value

Sources: Google Ads manager account (standard columns), Meta Ads Manager, read 17 Nov 2026. Platform attribution, not incrementality. Google value stops at Aug 2026 (see Method).

Paid media at scale, across many accounts and channels

Everything below was read on 17 Nov 2026 straight from the ad platforms, with client names removed. Google Ads goes back to my first accounts in December 2015. Meta only lets reports go back about 25 months, so the Meta totals are a floor.

Google Ads · all time (Dec 2015 – Nov 2026)
US$5.17M

spend in 35 accounts with delivery (40 accounts, 8 managers). Through Aug 2026: US$26.1M conversion value on US$4.94M spend, ROAS 5.30.

E1
Google Ads · last 12 months
US$1.90M

spend. Nov 2025 – Aug 2026: US$9.39M conversion value on US$1.67M, ROAS 5.64.

E2
Meta Ads · all portfolios (floor)
US$2.56M

last 12 months (22 accounts with delivery); US$3.29M in 25 months (30 accounts). 69 ad accounts in 37 business portfolios.

E3, E4
TikTok Ads · lifetime
US$203k

7 ad accounts in 6 Business Centers; US$108k in the last 12 months, almost all in one fashion account.

E16
LinkedIn Ads · lifetime
US$8.6k

6 ad accounts, 38 campaigns: food & beverage, HR, IT and B2B consulting.

E17
Measured history, all channels
US$8.67M

Google since 2015 + Meta 25 months + TikTok + LinkedIn. Older Meta history exists but the platform no longer reports it.

Who the accounts belong to (anonymised)

Client typeChannelSpendValue / resultROASPeriod

Case: fashion e-commerce, scaled on Google, Meta and TikTok at once

An apparel brand selling online. I run all three platforms from my manager accounts: Search, Performance Max and Demand Gen on Google; Advantage+ prospecting, retargeting and seasonal campaigns on Meta; Smart+ campaigns on TikTok. Every day an agent-run routine reads the accounts and applies negatives, bids and budgets within guardrails I set.

Meta AdsLast 12 months25 months
SpendUS$2.26MUS$2.68M
Purchase value (platform)US$11.34MUS$13.58M
ROAS5.015.07
Purchases338,187407,952
Cost per purchaseUS$6.69US$6.57

Same brand on Google and TikTok

Google Ads, 2019–2025US$3.08M spend
Conversion value · ROASUS$16.98M · 5.52
Google Ads, 2025 aloneUS$1.40M · ROAS 5.34
TikTok, lifetimeUS$199k spend
TikTok, last 12 monthsUS$108k spend
Meta impressions, 25 months1.67B
Meta Ads account totals for the fashion e-commerce account, 25 months
E3 · Meta Ads Manager totals, Oct 2024 – Nov 2026, campaigns hidden: US$2.68M spend, 407,952 purchases, ROAS 5.07.
Meta Ads totals for 30 days
E3b · One 30-day window (9 Aug – 7 Sep 2026): US$284k spend, 47,796 purchases, ROAS 4.79.

Case: a digital product launched from zero

Two digital protocols in the natural-health niche, sold on Hotmart. Sales started from zero in December 2023, and I ran the launch and the growth.

Hotmart, approved + completed salesFirst 12 monthsAll time
PeriodDec 2023 – Nov 2024Dec 2023 – Nov 2025
Sales8,5069,408
Net revenue to the producerUS$152kUS$167k
Best monthmonth 2: 1,430 sales

How to read it

"Net revenue" is what Hotmart reports as the producer's net receipts, after platform fees and commissions; the price buyers paid was higher.

About 73% of the first year's revenue came in the first 6 months. The screenshot below is the totals card of the sales report; the buyer list was not captured.

Hotmart sales report totals
E15 · Hotmart sales report, all time: 9,408 sales; net revenue shown in the platform's original currency.

Lead generation

Not every account is judged on ROAS. Law firms, clinics, schools, restaurants, retail franchises and B2B services buy leads, and I run them on cost per lead and lead quality.

Partner Google Ads manager · 11 accounts
47,996 leads

on US$106k of spend: retail, furniture and local services.

Second agency manager · 6 accounts
22,674 conv.

on US$58k: law firms, clinics, water filters, a packaging store.

Meta · franchise chain (2 portfolios)
US$200k

in 25 months across franchise and consumer accounts, plus clinics, a school and restaurants in other portfolios.

Tracking that the numbers can stand on

I set up and maintain the measurement layer myself: Google Tag Manager (web and server-side), GA4, Google Ads and Meta conversions, Consent Mode v2 for EU sites, and the store platforms' data layers.

  • 168 GTM containers across two logins (134 accounts listed, 14 containers shared by both), at least 15 of them server-side (sGTM). 84 GA4 accounts.
  • Recovered lost purchase tracking on wBuy (data layer + GTM) and on Shopify with server-side GTM, and built daily alerts for when it breaks again.
  • Diagnosed GA4 "Unassigned" / payment-gateway self-referrals and fixed channel grouping in GA4 itself.
  • Audited my own reporting: custom columns in the manager account inflated conversion value, so every number here uses standard columns; and a tracking change in the fashion account since Sep 2026 inflates conversions, so Google value is shown only through Aug 2026.
Tag Manager account list with names blurred
E5 · Tag Manager, all accounts in one login (names and IDs blurred).

SEO and GEO: a blog that search engines and AI answers both pick up

A premium tea and herbs brand I work with had almost no organic traffic. From June 2026 I ran a long-form blog program written for two readers: Google, and the AI answers that quote sources. That meant 142 articles of about 3,000 words, each built on the questions people actually ask ("is it an infusion or a decoction?", "how long should it steep?"), with an answer first under every heading and an llms.txt at the site root.

Site clicks, Aug–Oct vs Mar–May 2026
793 → 5,150

6.5× clicks; impressions 277k → 2.07M (7.5×).

Blog share, last 16 months
65%

of all organic clicks (7.27k of 11.1k), 75% of impressions.

Google AI features (Search Console beta)
292k

impressions in AI answers, from about 300 a day to about 3,000 a day. Every top page is a blog article.

Revenue from the blog is not measured, so it is not claimed here.

Skarb: a lifecycle CRM where AI writes each email from the customer's purchase history

I designed and built Skarb, a multi-tenant SaaS for e-commerce that puts order tracking, CRM, email, WhatsApp and support in one place. The wellness brand I work with runs its whole post-purchase operation on it, and other merchants are live on it too.

How the automation works

  • Orders, customers and invoices flow in from the store platform and the ERP. 14 journeys fire on store events: payment, shipping, delivery, abandoned cart and win-back at 30, 60, 90 and 180 days.
  • Each journey sends email first and escalates to WhatsApp only when the email is not opened. That saved 225 paid messages in 90 days.
  • In post-delivery and win-back, AI writes each email for the individual customer from what they bought, when, and their order number. It recommends complements only from a closed list of the store's live catalog and never sees personal data.
  • Skarb builds a ready-to-buy cart in the store with the original items (out-of-stock items left out). It is valid for 210 days, the same link is reused across the four win-backs, and it switches off when the customer buys.
  • Safeguards: the AI only rewrites a marked area of a designed template, the engine re-inserts any action button it drops, fixed brand rules rank above every prompt, there are two AI providers with failover, and if the AI fails the base text goes out so a journey never stops.
Wellness brand, last 90 daysResult
Messages sent (email + WhatsApp)16,933
Opened / read · clicked28.4% · 3.5%
Attributed orders82
Share of the store's paid revenue12.2%
Return on tool + WhatsApp cost9.6×
Abandoned-cart conversion4.3% of sends
November so far: share of paid revenue21.6%

"Attributed" means a paid order within 7 days of a message. This is attribution, not incrementality.

Engineering

PHP 8.4 / Laravel 12, MySQL, Redis queues, 49 scheduled jobs, 159 migrations, about 207k lines of my own code. Fail-closed multi-tenant isolation. Adapters for 7 store platforms, 6 carriers, email (Amazon SES pool at 50,000 emails a day, Brevo, SendGrid), WhatsApp (YCloud) and AI (Anthropic, OpenAI).

Second Order: the store's real orders next to what each ad platform claims

A SaaS panel I designed and built for online stores in Poland. Meta and Google both take credit for the same order, and GA4 quietly loses sales to consent banners and payment redirects. Second Order starts from the store's real orders and puts every platform's claim beside them.

  • It shows real ad cost per order, MER, how much the platforms over-claim, GA4 coverage day by day, monthly goals with budget pacing, and a retention lift measured against a holdout group.
  • Every morning it compares yesterday's orders with GA4 and emails an alert when tracking breaks, MER drops or spend runs over budget.
  • 29 read-only integrations: Shopify, WooCommerce, IdoSell, Shoper, PrestaShop, BaseLinker, 8 ad platforms (including ChatGPT Ads), GA4, Search Console, social profiles and email tools.
  • Agency mode, PDF reports in Polish and English, and fully automated sales: a Stripe subscription, a 14-day trial and Polish VAT invoices through Fakturownia / KSeF.
  • It runs on Cloudflare Pages + Workers with KV and AES-GCM encrypted credentials, plus a separate cron worker. About 38k lines of code and 21 automated test suites.

Live demo with sample data: panel.secondorder.eu/demo

Second Order overview
E13 · Overview (public demo, fictional store).
Second Order tracking health
E14 · Tracking health: store orders vs GA4, Meta and Google, with automated checks.

Other things I have built and run

Client dashboards

At Tags & Pixel, where I work as a digital business strategist: a PHP portal per client with its own login. A server-side proxy calls the Reportei API, so the token never reaches the browser, and every client sees the same 7 shared report pages versioned on GitHub (overview, sales and products, Meta, Google, GA4, SEO, audience).

Instant-payment store for digital products

A storefront on Cloudflare Workers + D1 + KV with automatic instant payments through a Mercado Pago webhook, PDF delivery by the Gmail API, offers, a customer area and an Instagram/Facebook DM bot built to replace ManyChat.

Mobile idle games

Second Order Games, a TypeScript + Vite + Capacitor monorepo with a shared idle engine, AdMob with GDPR consent and RevenueCat, for Google Play and the App Store.

AI video clone

A HeyGen digital-twin pipeline run through MCP for Instagram videos in Portuguese, English and Polish.

Store operations

A wBuy domain and email migration to Cloudflare + Zoho, Customer Match uploads with hashing, Merchant Center promotions, and Black Friday launches across site, email and ads.

Code footprint

7 private repositories. Skarb alone has 822 commits from 19 Aug to 16 Nov 2026, 1,202 files and about 207k lines.

How I work: AI agents with guardrails, not chat

  • Scheduled optimisation agents. A daily Google Ads routine for the fashion account and a weekly one for the wellness brand apply negatives, bids and budgets within limits I set (for example, a target ROAS floor and step-wise budget changes). Meta changes go to me for approval.
  • Connected data. The agents read the accounts directly through connectors (Solomon, Reportei, Bling ERP, BigQuery, the Google and Meta ad UIs) instead of from pasted exports.
  • 67 written runbooks (skills). These cover PMax audits, ROAS-decline diagnosis, profit-based scaling with break-even ROAS and CAC, tracking repair, seasonal launches and app-store submissions, so every routine runs the same way each time.
  • Claude Code with subagents for software, with a reviewer agent that checks every plan and every finished stage.

Background

Before paid media I spent over ten years in software and banking technology (IBM, Itaú-Unibanco, NTT DATA, Accenture Poland), from COBOL development to leading delivery teams. I live in Warsaw and work in English and Portuguese.

Stack

Exhibits: the screenshots behind the numbers

Client names, account IDs and campaign names are blurred or cropped. Screenshots show each platform in the account's original currency; the text converts to US dollars. Exhibits E3, E5–E15 appear in the sections above.

How the numbers were taken