Stop reading the conversion column as a scoreboard.
The tracking had not stopped working. This market simply produced few form fills a week, and some leads only showed up in the CRM. No cost-per-lead target was ever agreed, only a monthly spend plan.
So we steered on what the platform could measure. With a cycle this long, one ad rarely collects enough conversions to judge it on cost per conversion. We set ads to rotate evenly and compared them on click-through rate within the same ad group. Searches from people still getting to know the subject barely registered in the conversion column. We kept them running, because someone who is still reading up can become a lead later.
After we moved to manual bids, Google recommended switching back to conversion-based automated bidding. We declined. On a signal this thin and this unreliable, it would have optimised toward noise.
Spend was running behind plan, and the obvious fix was a bigger budget. The data pointed elsewhere. The generic campaign used only a small part of its budget, lost nothing to it, and was losing more and more auctions on ad rank.
We moved it off conversion-based automated bidding and onto manual bids, with a default bid per ad group. The week after, it was showing up in clearly more of the auctions it had been losing on rank, and spend moved toward plan without any change to the budget.
Then we raised the default bids on the main ad groups and checked what had actually changed. Far less than it looked. Most keywords still carried their own bids from earlier tuning, and a keyword’s own bid overrides the ad-group default. So we raised bids keyword by keyword, only where there was proven traffic. That moved spend. Within a few weeks we were holding it back to stay inside the monthly plan.
The one exception was the brand campaign. It took on more traffic once we routed brand searches out of the generic campaign, and competitors bidding on the brand name pushed up what each click cost. On busy days it ran out of budget. A small raise closed that gap the following week.
We would rather run behind a spend plan than hit it by buying more expensive clicks.
Every week we pulled competitor names out of the search terms. From mid-year on, each candidate was researched, then argued against, before it went on the exclusion list. Negatives match literally, so two spellings need two entries, and a short brand name sometimes needs an exact match so it does not block a genuine search.
Searches for its own brand name leaking into the generic campaign were excluded there, and we checked every week that it held. When Google later advised removing those negatives as “conflicting”, we kept them. Removing them would have reopened the leak. Searches that looked off-topic at first sight, but came from employers with a real hiring question, stayed in on purpose. Those are the buyer’s questions too.
Generic category phrases stayed off the list, even when a competitor carried one in its name. Excluding them would have blocked real buyers.
What was left was a quality problem.
Manual bids got the generic campaign into more of the auctions it had been losing on rank. Later in the year it ran behind plan again, and this time bids were not the limit. Nearly every keyword with impressions already bid above Google's estimate for the top of the page. A late recount still turned up a relevant keyword that a low bid was holding back. We raised it in one measured step. Google rated most ads relevant, but rated expected click-through and the landing page experience below average on almost every keyword it had enough data to score. No bid or budget reaches either of those. As the next step, we put concrete fixes for the page on the table. There is no clean before and after on cost per lead, because the leads came in through the CRM.
Read an empty column as a question.
Every engagement starts with a 30-minute audit — free, no slides, just a screen-share through your account.
Book the audit →