The campaign that never spent a cent

On July 5 we launched a Google Ads campaign to sell our own services. A modest budget, reviewed ads, keywords picked by hand.

Two months later it had spent MXN 103 (about USD 6). Not 103 thousand: one hundred and three.

We diagnosed it four times. The first three times we found something broken and fixed it, and all three times it still would not serve. The fourth time we found the real reason, and it was none of the previous ones.

This is the full autopsy, including the part where one of our own fixes is what killed it.

Autopsy 1: the bid cap

Ten days after launch: zero impressions. Not a few. Zero.

Everything said it was fine. Campaign "Eligible". All 18 keywords, "Eligible". The ads, approved. Balance available. Targeting correct.

The culprit was hidden in the bidding settings: the "Maximize clicks" strategy had a maximum CPC bid limit of MXN 25 (about USD 1.40) turned on. For B2B software and artificial intelligence keywords in Mexico, 25 pesos is not enough to get into the auction. The ad is eligible and simply never competes.

We removed the cap. And it worked: the campaign started serving and spent those 103 pesos.

What we learned: "Eligible" does not mean "showing". It means nothing is stopping it from showing. Those are different things, and Google gives them the same name.

Autopsy 2: the declaration nobody answers

From July 20 to August 19, zero again. A full month.

This time the trail showed up in the bid strategy report, a screen almost nobody opens. It said Inactive. The campaign was eligible, the ads approved, and the bid strategy switched off.

We found two causes and fixed both:

• An unanswered regulatory declaration in the account settings, about European Union political ads. Even if you have nothing to do with that, if you do not answer it Google can pause delivery. It does not send an urgent email: it just sits there.
• Google had turned on an automation feature on its own that we had deliberately switched off when we created the campaign. We switched it off again.

And while we were at it, we changed the bidding from "Maximize clicks" to "Maximize conversions", reasoning that with conversions set up the system would optimize better.

That change was a mistake, and it took us twelve days to notice.

Autopsy 3: bidding toward an impossible conversion

August closed at zero pesos. A full month, zero impressions.

We checked absolutely everything obvious, and all of it was healthy: campaign eligible, two ads rated "Good", 18 eligible keywords with none flagged as low search volume, billing up to date, bid strategy report showing Active with no recommendations, automation switched off.

The cause was in a place we had not looked: which conversions were assigned to the campaign.

The account's seven conversion actions were templates Google creates on its own, meant for local campaigns: phone calls, click-to-call from the map, "directions". All of them calls or map actions. None from the website.

A Search campaign that sends traffic to a website, with no call asset, can never produce those conversions. And an automated bid strategy optimizing toward an unreachable conversion does the only reasonable thing: it bids practically zero.

The campaign settings said so in plain text, in a line we had read without reading it: "your campaign is optimized to generate Contacts, Phone calls".

Here is the uncomfortable part: we broke that ourselves. The August 19 switch to "Maximize conversions", made as a fix, is what shut the campaign down for that stretch. We went back to "Maximize clicks" with no cap.

What we learned: eligible on everything, plus an active bid strategy, plus zero impressions, means you have to check which conversion the campaign is bidding toward. And you have to distrust your own fix as hard as you distrust the original problem.

Autopsy 4: the question we never asked

September 9. Thirty more days: zero impressions, zero pesos.

We verified everything again, one item at a time. The CPC cap checkbox, unchecked. Ad groups, eligible. All 18 keywords, eligible, none with low search volume. Both ads, eligible and rated well. Available balance: MXN 646 (about USD 36). Not a single policy, verification or suspension alert anywhere in the account.

There was nothing broken left to fix. So we finally asked the question we should have asked on July 5, before spending the first peso:

how many people are searching for this?

We pulled the real volumes from Keyword Planner, for Mexico, over the last twelve months:

• agencia de inteligencia artificial ("artificial intelligence agency"): 10 to 100 searches a month
• agentes de ia para empresas ("AI agents for business"): 10 to 100
• ia para empresas ("AI for business"): 10 to 100
• soluciones de inteligencia artificial para empresas ("artificial intelligence solutions for business"): 10 to 100
• desarrollo de software a la medida ("custom software development"): 10 to 100
• chatbot para empresa ("chatbot for a company"): 10 to 100
• automatización de procesos empresa ("company process automation"): not enough data
• automatización de procesos con inteligencia artificial ("process automation with artificial intelligence"): not enough data

Nine out of ten of our keywords get between 10 and 100 searches a month across all of Mexico. That is one to three searches a day, for the entire country, split among every competitor.

The only one with real volume, automatización de procesos ("process automation"), dropped 90% in three months and 90% year over year. And it is so generic that a good share of those searches are about industrial automation, not software.

On top of that, when an auction does come up, the top of page bid ranges from MXN 37 to MXN 2,000 per click (about USD 2 to USD 111).

The campaign was never broken. There is no market of that size. There are almost no auctions, and the ones there are cost more than the entire daily budget.

The right order to diagnose

If you have a campaign that will not spend, this is the sequence that would have saved us two months. It is inverted on purpose: the cheapest thing to check goes first.

1. Search volume. Before anything else. If your keywords add up to a few dozen searches a month, no setting will fix it. Five minutes in Keyword Planner.
2. The bid limit. If the strategy has a CPC cap, compare it against the top of page bid for your own keywords. A cap below that is an off switch.
3. The bid strategy report. If it says Inactive, that is your problem, even if everything else says "eligible".
4. Pending declarations in the account settings. They pause delivery silently.
5. Which conversion you are bidding toward. Open the campaign settings and read the line that says what it is optimized for. If that conversion cannot happen in your funnel, automated bidding drops to zero.
6. Billing and policies. Last, because those are the things Google does warn you about.

We went through it from 6 to 1.

Why we are publishing this

Because the industry's bias pushes the other way. An agency with a campaign that will not spend has every incentive to keep "optimizing": every review looks like work, every tweak looks like progress. We did exactly that three times.

The honest conclusion was not billable: this channel does not work for this service in this country.

And there is a contrast that makes it clear. The same week, with the same operator and the same tools, a Search campaign for an auto repair shop in Puebla had been hitting its daily budget for weeks and bringing in leads at a cost per conversion of MXN 31 (about USD 1.70).

The difference is not skill. It is that thousands of people a month search for "llantas en Puebla" (tires in Puebla), and a few dozen search for "agencia de inteligencia artificial". Google Ads is a channel for capturing demand, not creating it. If the demand does not exist, there is nothing to capture.

What we are doing now

Before shutting it off, we left one last, bounded two-week test running with AI-expanded search, to see whether there is adjacent demand our keywords are not catching.

We wrote the decision rule before seeing the results, which is the only way a test means anything:

• Zero impressions in fourteen days: there is no market, it gets shut off.
• Impressions from irrelevant searches: it is noise, it gets shut off.
• Clicks from real commercial searches: there is a signal, it gets evaluated.

In parallel, we measured the same service in the United States, in English. There, the same categories get between one thousand and ten thousand searches a month. That is another article, and probably another campaign.

What we are taking away

We spent two months of attention fixing something that was never going to work, because we skipped the cheapest question of all. Measuring demand before building the funnel takes five minutes; finding out at the end cost us a full quarter.

And the second one, less comfortable: when you fix something and the problem persists, the hypothesis to review includes your own fix. One of ours was to blame for an entire month.