Google Ads for B2B: long cycle, few applications, high cost of error

No, and this is a major B2B pitfall. Automatic strategies learn from conversions, and they need dozens in a short period - on ten bids, the system works almost at random and rearranges bids due to random fluctuations. We start with manual or portfolio management and move to automation when the volume of conversions supports it. If it never lasts, we leave the tame forever and say it directly. An intermediate step that often saves: to count as a conversion not only an application, but a wider set of actions with different value - downloading a price, requesting an account, registering a dealer. The signal volume grows, but the priority still remains on what is more expensive.

Three layers, and none of them work alone. Minus words at the campaign level remove "retail", "used", "one piece", household brands and repair requests. The text of the ads mentions the minimum batch and the format of work - some people simply do not click, and this is the cheapest cut-off, because it is per click. Posadkova shows the terms of cooperation instead of the "buy" button. Honestly about the limit: retail will not disappear completely, because some of the requests are the same for both audiences. Its share falls, and the expenses for it fall more - simply because expensive clicks stop going to the wrong ones.

In part, there is also a hard technical limit here, which is better to know before launching. The conversion window in Google Ads extends to a maximum of 90 days, and it will not work to return a click transaction older than this period to the account - the import of offline conversions will reject such a line. So we do two things at the same time. We expand the window to the actual cycle, as much as the system allows, and in parallel we keep our own accounting in CRM by the source label: the deal lives in it as long as it is needed, and no platform will cut it. Because of this, the numbers in the office and in your account are different - this is a normal state of affairs, not a configuration error, and in the monthly report we show both.

So that the system learns from qualified appeals, and not from all in a row. Without it, optimization goes to the number of forms, and the campaign conscientiously finds the cheapest applications - usually from those who are not suitable for you. Technically, this is either the import of offline conversions based on the gclid saved in the CRM card, or enhanced conversions for leads on hashed email and phone: the second option saves when the gclid did not reach the CRM, because the form was filled out after returning from the bookmarks. If neither works, we at least separate the events by action type: a price request and a bill request are signals of different value, and to count them equally is to deliberately spoil the learning.

The presence of a meter and working measurement are different things, and we saw this on our own projects during the technical measurement on July 31, 2026. In the 1,848-card B2B catalog of industrial sewing machine parts, the code is still running Universal Analytics config, disabled by Google on July 1, 2023, and the GA4 property is completely absent: for three years, the site looks measured and collects nothing. Everything is alive in the wholesale building materials store with 3,145 items — a tag manager, GA4, a remarketing tag — but the events that go there are product views, not appeals: you can gather an audience for return, but you can’t teach a campaign on deals. Therefore, the first step of parsing is not semantics, but reading the code of the pages.

A click is often more expensive, a transaction is almost always cheaper, because the average check is different. It makes no sense to compare B2B and retail by the cost of a click at all: a meaningful comparison is only based on the cost of a deal or at least a qualified referral. That is why we insist on returning statuses - without them, this number does not exist, and all that remains is to argue about the cost of a click, looking at two different reports.

Change the signal and filter at the input. We add a question about the volume to the form, divide the events by the size of the request and transfer to the advertising account only those requests that pass the threshold. The number of applications in the report after that drops, sometimes by half, and for exactly one month it looks like a deterioration. Then the budget stops going to requests, after which there are no shipments. The second half of the work is not ours: someone has to manually set the threshold in CRM, and without this person the scheme falls apart in two weeks.

Sometimes there is not, and we say so before the contract. If your potential buyers in the country are several hundred companies, search demand may not be collected even on a modest budget: the campaign receives twenty impressions per day, and statistics for decisions are not collected for years. In such a situation, it is more honest to advise industry platforms, direct contacts and technical content than to charge money for management. Analysis of the funnel is exactly what is needed to find out in a few days, not in a quarter of expenses.