B2B SaaS, AI sales agent: 364 card-on-file trials, paying customer at $168

The task

The client is a B2B SaaS: an AI outbound sales agent. PLG model: the trial opens only with a card attached, usage-based billing via Stripe.

Starting point (tech audit 16.06.2026): the Meta pixel sat inside GTM and sent one event — PageView. Sign-up and payment lived in GA4 and on the app domain, not tied to fbq, no Conversions API. Meta saw only clicks — there was nothing to optimize for.

What we did

• Moved the pixel out of GTM natively into the <head> of both domains: pixel in GTM + events in code = an fbq load race, events get lost and duplicated.
• Tagged the funnel exactly along the real path: PageView → Lead (form with URL input) → StartTrial (email + card) → Purchase.
• Set up the Conversions API on the product’s API domain: Lead and StartTrial duplicated from the browser with a shared event_id, Purchase — from the Stripe webhook. Payment happens server-side, the browser pixel never sees it.
• Captured fbclid → _fbc, stored _fbp/_fbc/device_id in the DB, advanced matching on hashed email: payment comes days after the click, otherwise there is nothing to tie it to.
• Optimized campaigns for StartTrial — a deep-funnel event. From 24.07 we split geo into EU and US/UK and separately tested optimization for InitiateCheckout.

Results

For 24.06–31.07.2026:

• Ad spend — $17,310
• Leads (Lead) — 4,327 at $4.00
• Card-on-file trials (StartTrial) — 364 at $47.6
• Payments (Purchase) — 103
• Cost per paying customer — $168
• Reach — 132,287 accounts, frequency 4.27

Insights

• The top of the funnel is wide and cheap: 51% of landing page views reach the form, a lead costs $4.00. Then it narrows sharply — 8.4% of leads attach a card, 28.3% of trials make it to payment.
• Lead $4.00, card-on-file trial $47.6, paying customer $168 — a 42× spread. Had the campaign optimized for a click or a form, the algorithm would have pushed volume into the first number and never seen the last.
• Frequency 4.27 at 132K reach over 38 days of delivery: the B2B audience is narrow and burns out. The answer is not more budget on the same geo but a split: in the first 8 days US/UK delivered a trial at $53.45, EU — at $73.85.