← All posts

The First 90 Minutes: What Broke and What Worked on a New Customer's First Plyto Campaign

Nov 19, 2026

Why we're writing this up

Most "case studies" skip the embarrassing parts. We're not going to. This is a minute-by-minute log of the first campaign a new Plyto customer launched last month — a B2B services firm with a $60/day Google Search budget, one landing page, and a sales team of two.

The campaign worked. It also broke twice before it worked. Here's the timeline, so you know what to expect when you spin up your first one.

T+0: Account connection

The customer connected Google Ads, Google Analytics, their Stripe account, and their domain. Plyto pulled the last 90 days of ad history and conversion events on the first sync.

What went right: historical data meant the agent didn't start blind. It already saw which keywords had converted before and which had bled money.

What went wrong: the customer had two Google Ads accounts under the same MCC, and the agent connected the dormant one first. We caught it because spend data looked flat for the last 30 days. Lesson: if you have multiple accounts, label the active one clearly before connecting.

T+14 minutes: Conversion audit

Plyto ran a conversion audit before writing a single ad. It flagged three problems:

1. The "Contact" conversion was firing on page load, not form submit.

2. Phone call tracking wasn't installed.

3. The Stripe-to-Ads value import had never been turned on, so the account had been optimizing for raw lead count, not revenue.

The customer fixed #1 and #3 that afternoon. Phone tracking took longer because it needed a DNS change. This is the single most common delay we see — plan for it.

T+41 minutes: Keyword and audience draft

The agent proposed 23 keywords across four ad groups, plus a negative list of 180 terms pulled from the account's own search term history. It also proposed one broad match group gated by an audience signal, and one phrase match group with no audience gating.

The customer rejected two ad groups outright — both were adjacent service lines they don't actually sell. Good catch. The agent had inferred them from the website's "Services" page, which listed offerings they'd quietly discontinued.

Lesson: your website is the agent's source of truth. If it's out of date, so is your first campaign draft.

T+58 minutes: Ad copy and landing page

Plyto generated 12 responsive search ads and one landing page variant. The copy was fine. The landing page had a problem: it referenced a pricing tier the customer had sunsetted six months ago. Same root cause — the agent had scraped old content.

The customer edited the page in Plyto's editor, re-approved, and the agent regenerated the ads to match the corrected pricing. Total time on this step: 19 minutes.

T+1 hour, 17 minutes: Launch approval

The customer hit approve. The agent pushed the campaign live in Google Ads, set a $60 daily cap, and configured the first automated check-in for 48 hours out.

T+2 hours: First impressions

First impressions came in about 40 minutes after launch. First click came in at the 2-hour mark. The agent did not touch bids yet — it was waiting for a minimum sample before acting.

This is a point worth making: a good AI agent waits. If yours is reshuffling bids in the first hour on five clicks of data, it's guessing.

T+26 hours: First conversion, first problem

The campaign got its first form fill at the 26-hour mark. Cost per conversion: $31. Not bad for a cold start.

Then the agent flagged something: three of the first eight clicks came from a search term that looked commercial but wasn't — someone researching a competitor's layoff news. The agent added it to the negative list automatically and asked the customer to confirm. They confirmed.

T+48 hours: First optimization pass

The scheduled check-in ran. The agent:

  • Paused two keywords with four+ clicks and zero engaged sessions.
  • Shifted $12/day of budget from the broad match group to the phrase match group, which had a 3x better conversion rate.
  • Flagged that the landing page's hero headline was getting high bounce on mobile and proposed two alternates.

The customer approved the budget shift and one of the headline tests. They rejected the keyword pauses because one of those terms was a brand-adjacent phrase they wanted to hold position on, even unprofitably. Fair call. The agent noted the override and stopped suggesting it.

What this campaign taught us

A few things we'd tell any new customer before launch:

Fix conversion tracking first. If your conversions are wrong, every optimization an AI agent makes after that is wrong too. The 14-minute audit is not optional.

Your website is the brief. The agent writes ads and pages from what's publicly on your site. Outdated service pages produce outdated ads.

Expect one DNS-related delay. Phone call tracking, domain verification, or a conversion pixel will hold you up for an afternoon. Build that into your timeline.

Overrides are fine — and they teach the agent. When the customer kept the brand-adjacent keyword against the agent's advice, Plyto logged that preference and stopped flagging it. The feedback loop is the whole point.

The 30-day result

At the end of the first 30 days, the campaign had spent $1,780 and produced 47 form fills and 6 booked sales calls. Cost per booked call: $297. Their previous agency's number, on a similar budget, was roughly double that.

None of this is magic. It's the same work a sharp media buyer would do — compressed into one evening, with a tighter feedback loop after. The parts that went wrong went wrong because of stale inputs, not bad decisions. Fix the inputs before you launch, and your first 90 minutes will look a lot cleaner than this one did.

Plyto is your AI marketing assistant: it makes and runs your ads

It writes the ad, builds the landing page, captures the leads and shows you which dollar came back. Try it free, no credit card.

Start free