LinkedIn
The founder had the pricing page open on screen share when he told me the offer was final. Three tiers. A 14-day trial. That was the hill they were going to die on.
I took notes, built the Google Ads account around that promise, and spent the next two days writing exact-match terms by hand for every way a buyer might ask for it. I believed him for about a week.
That was quarter one. By quarter four, we had run four offers through the same account, rewritten the ads again and again, and rebuilt the negatives whenever the old intent stopped fitting the new pitch. At the time, I wanted the account to settle down. Now I think about the window we were using to judge it: a month was too short, and a year bundled four different promises into one meaningless average.
Quarter one: the offer that was definitely final
I built the first structure the way I built everything back then. One campaign for exact match, one for phrase. Ad groups split so tightly that each held five or six keywords meaning roughly the same thing. Trial pricing appeared in the headlines; trial language appeared on the landing page. I started with a negative list I had carried from client to client, then added to it at 1am when the search terms report showed me what I had missed.
Say we were spending around $20k a month. At that spend, a loose structure does not stay a tidy little mistake. It spends money while you are looking elsewhere. So I watched the terms, kept the promises aligned, and made sure the page answered the question the ad had raised.
For three weeks, it worked the way the textbooks say it should. Search terms lined up with ads. Ads lined up with the page. Trial starts came in at a cost I could defend on a call. I remember thinking this one might be stable.
That should have been my warning.
Quarter two started with a short email. The subject line was something like quick update. The trial was gone. The new offer was a demo, sold as a guided walkthrough and priced a tier above where we had been. Same budget. Much of the same search traffic. A completely different promise behind the click.
What the deck would have called a pivot, I called rebuilding the account on a Tuesday night with cold coffee. I had written the original ads to meet people looking for a trial. Now those people were arriving at a request for a sales conversation. Keeping the keywords did not keep the intent intact, and keeping the budget did not make the old conversion history a useful verdict on the new offer.
Google could still show our ads for searches that had made sense under the trial. Meanwhile, I was rewriting headlines for the demo, sorting search terms against a new definition of useful, and deciding which negatives belonged to the old account rather than this one. I threw out half the list. Cost per demo came in 40% higher than cost per trial start had. Those two actions were not worth the same thing to the business, but the difference was impossible to miss on a dashboard.
I used to tell clients a pivot was just new ad copy. I was wrong. A new offer resets what a useful click means. Quarter two needed to be read as month one again, even though the account itself was older.
The Thursday the landing page stopped matching
It was a Thursday when the numbers broke. Click-through held. Impressions held. But cost per click climbed day after day, and our average position slipped. I checked the search terms, the bids, and whether a competitor had entered. Then I opened the landing page from an ad preview and did not recognize it.
New headline. New hero image. New form with three extra fields. Someone on their side had swapped the page on Tuesday and told no one.
My ads promised the guided demo we had just rebuilt the account around. The new page talked about enterprise features. The person who clicked had one expectation; the page presented another. That mismatch is the sort of thing that damages Quality Score, and it gave me a reason to look beyond the bids as CPCs rose by roughly a third in a week.
I have since seen the split tested cleanly: generic or mismatched pages bouncing near 87% with a 4 out of 10 score, matched pages bouncing near 32% with close to an 8. That week, I did not need a tidy test result to tell me which side our page resembled. The ad and page no longer carried the same offer.
I keep a simple rule now because Quality Score is a check-engine light, not the engine. The visible 1-to-10 score is historical and aggregated, drawing on expected click-through, ad relevance, and landing page experience. A 7 is good. Even a 6 can be fine. I pay attention when it trends down or impression share starts slipping to rank. Ours was trending down fast, for a reason no bid adjustment could fix.
I called the founder, asked who had touched the site, and heard a pause that told me everything. If someone changes the page, the person buying the clicks needs to know before those clicks arrive. It sounds obvious until a Tuesday page swap spends the rest of your week proving it.
Quarter three: the report sales did not recognize
Quarter three was the strangest one. On my dashboard, it looked like the best quarter we had. Demos were up. Cost per demo was down from the quarter-two spike. The search terms report was clean for the first time in months. I had numbers I was ready to discuss on the monthly call.
Sales sounded like I had sent them nothing. Not fewer leads than they wanted. Nothing they could use. I remember looking at my report while the head of sales walked through his list and realizing we were describing two different accounts. Mine ended at the form. His began there.
The cause took me too long to see because the numbers looked healthy. I had set the primary conversion to form fill, so Google found people likely to fill forms. It was doing what I had asked. It could not tell from that signal alone which demo requests would become useful conversations, let alone customers.
The math is plain once you stop calling every fill a win. In one breakdown, 40 demos yield 6 qualified leads and 1 customer. The ad account reports 40 conversions; the business gets one customer. Another SaaS example shows 86 leads at $116 each, with only 5 becoming customers. Our account had the same shape: plenty of fills and very few second meetings. I had been defending the cost of an action sales could not do much with.
That is not an argument for ignoring forms. Early in a build, they can give bidding something to work with while later sales stages are still sparse. It is an argument for knowing when the form has stopped being an adequate definition of success. I should have captured the click IDs and connected sales stages to the account before the clean report gave me false comfort. Then we could have used fills as an early signal and shifted the primary conversion toward qualified opportunities once there were enough of them.
Count fills as the win, and you buy fills. I learned that one a quarter later than I should have. By the time sales and I were looking at the same list, the offer was about to change again.
Quarter four: one offer, one number, 90 days
Quarter four brought the fourth offer. The demo stayed, but now it sat behind a paid pilot: two weeks, a fixed fee, credited back on an annual signature. Another headline. Another page. Another definition of what counted as a win. I had stopped arguing about the pivot itself. Startups pivot. What I could ask for was a fair window in which to judge this version.
I told the founder we would hold the offer steady for 90 days, put a target tied to sales in writing, and stop grading the account every Monday as though a new B2B pilot should close at the speed of a form submission. Trial to paid can take 14 to 30 days. A same-month report can mark an unfinished cohort as a failure, or celebrate cheap fills before sales has had time to reject them. I wanted a rolling 90-day view of cost per opportunity, with the CRM deciding what qualified.
If I started that account again, I would put the measurement in place on day one: capture the click ID on every form, map sales stages to their own conversions in Google Ads, and keep the form fill as an early signal while later outcomes come in. That is the path from lead volume to pipeline signals the bidding can use. It also gives the person managing the account a way to answer sales with something better than a screenshot of a conversion column.
I sat with the founder in quarter four and did the math I should have forced in quarter one. If annual value was around $1,200 and a customer stayed three years, lifetime value was $3,600. We set an acquisition ceiling of $1,188, just under a third of that value. At a 20% pilot-to-paid rate, that allowed up to $237.60 per pilot. The bid-to-CAC math gave us a way to discuss a costly-looking pilot without retreating to the cheapest form fill on the page.
There was still a catch. We had plenty of fills but almost none of the qualified kind. Smart Bidding generally needs around 30 to 50 conversions a month to stabilize, and I was not going to pretend the account had that volume at the stage we actually cared about. So we stayed manual, ran one deliberate test at a time, and let sales mark the real outcomes. Switching the primary conversion to an imported sales stage was the next move once we had enough of those outcomes, not a button to press because the slide looked better.
I have seen the value of that restraint in other 90-day search fixes: hold the offer long enough for the cohort to develop, then read the account against the target you set at the start. That was all I wanted from quarter four. One offer, one target, written down. Cost per opportunity, not cost per fill.
The pilot offer held. By the end of the quarter, the page had stayed put by agreement, and cost per qualified opportunity had been moving down for six straight weeks. Not a victory story. A stable story, for once.
Coda: four tabs and a whiteboard
I left that account with a spreadsheet I still keep. Four tabs, one per quarter. Each has its own offer, its own cost per fill, and its own short list of opportunities sales actually worked. I do not compare the tabs as though they describe one uninterrupted experiment. The buyer arriving for a trial in quarter one was not being asked to make the same decision as the buyer arriving for a paid pilot in quarter four.
On my last visit, the whiteboard still had the pilot pricing on it. I picked up a marker and wrote the line I wanted any search partner, myself included, to be willing to write: 18 qualified opportunities in 90 days, cost per opportunity under $237, counted in the CRM. If a partner will not put a pipeline number beside a quarter of work, I do not know what they have committed to.
Some nights I still hear that rebuild. Cold coffee, the negatives list open, search terms scrolling past midnight while I tried to teach the account its third new language in six months.
And I still see that board in quarter four: pilot pricing underneath, my marker squeaking across the line I would sign again.
