Cost per lead goes up. The meeting that follows is about campaigns, keywords, bids and audiences, because those are the things the account gives you buttons for. Meanwhile the actual leak is on a page nobody in the room has opened this month, in a form that asks eleven questions before it asks for anything the business will use.
Answer first
Before you touch a bid, split the number in two. Cost per lead is cost per click divided by conversion rate, and those two halves have completely different owners. If cost per click is stable and the rate has fallen, the auction is innocent and the problem is somewhere between the click and the submission. Instrument that stretch properly, page view, form start, field-level abandonment, submission, and let it tell you where people stop. Then fix the worst step, which in a lead generation business is very often the form, by asking the smallest number of questions that let you decide whether to talk to someone and collecting the rest later.
The two halves of a cost per lead
Cost per lead is an average of an average, and averages hide the thing you need.
The first half is what you pay to get someone onto the page. That is competition, quality signals, match types, placements, the things the platform is genuinely responsible for. When this half moves, cost per click moves with it, and you can see it in the account without leaving the interface.
The second half is what proportion of those people do the thing you wanted. That half lives entirely on your side of the boundary. The platform reports it, but the platform did not cause it, and no bid adjustment will repair it.
So the first diagnostic is trivial and almost nobody runs it: put cost per click and conversion rate side by side over the same period. If cost per click is flat and the rate has dropped, stop reading the campaign report. The account is fine and the page is not.
One caution. A stable average can hide a changed mix. If a campaign quietly started serving in a different geography, on a different device or through a different placement, cost per click can sit still while the underlying audience changes completely. Break the same two numbers down by device and by campaign before concluding anything.
Instrument the stretch you own
Between a click and a lead there are more steps than most accounts measure, and each one is a different diagnosis.
Landing page views per click. If this is meaningfully below one, you are paying for clicks that never see the page. That is a speed, redirect or consent problem, and it is invisible in the campaign report because the platform counted the click and stopped watching.
Form starts per landing page view. People arrived and did not begin. That is a message match problem, an offer problem or a layout problem: the page did not say what the advertisement promised, or the form was below three screens of hero image.
Submissions per form start. People began and gave up. This is the most damning of the three, because these are the visitors who had already decided they wanted what you were selling. Whatever happens in this gap, you caused it after they said yes.
Field-level data is worth having for the third one. Most form and session tools can tell you which field a visitor last interacted with before they left, and that single view usually ends the argument in one afternoon. There is always one field, and it is usually the one that felt reasonable to whoever asked for it.
Every field is a chance to say no thanks
This is the principle to carry into the rebuild. A form is not a container for the information your business would enjoy having. It is a sequence of small requests, and the visitor gets a fresh opportunity to abandon at every one of them.
The requests are not equal. Some are cheap: a first name, an email. Some are expensive out of proportion to their size, because of what they imply rather than what they cost to type. A phone number implies a call you did not agree to. A date of birth implies a check. A file upload implies twenty minutes and a laptop, on a phone, on a bus. An open text box implies homework.
So the question for each field is not “would we like this?” but “would we decline to speak to this person without it?” Anything that fails that test can be asked on the call, inferred from what they already gave you, or collected on a second screen after the commitment has been made. Progressive disclosure is not a trick. It is the recognition that somebody who has already given you three answers is far more likely to give you a fourth than someone looking at a wall of them.
Two adjacent things are worth checking while you are in there. Whether the form works properly on a phone, since that is where the traffic is and where a badly behaved date picker or an aggressive validation rule does the most damage. And whether error handling loses what people typed, because a form that clears itself on a failed validation converts approximately nobody.
What happened when we ran this
On a staffing and gig work platform in India, the account looked like a bidding problem and was diagnosed as one for a while. Two months of work went into establishing what was actually true: which campaigns produced applications rather than clicks, where in the sequence people were leaving, and what the traffic was really made of behind a stable average.
The application form came out of that as the largest single loss point in the funnel. Not the keywords, not the creative, not the bidding strategy. The form.
It was rebuilt to ask three things first and everything else afterwards. Following the July 2026 restructure, cost per lead fell by 57 percent.
The order matters as much as the outcome. The diagnosis took two months and the change took a fraction of that, which is the usual proportion, and it is why the instinct to start adjusting bids in week one is so expensive. You cannot fix the thing you have not located.
What this does not fix
A shorter form raises submissions. Whether it raises the number of people you actually want depends entirely on which questions you dropped, and it is possible to cut a form down to a cost per lead you love and a sales team that is drowning in unusable applications. Remove friction, keep qualification, and judge the result on cost per qualified lead rather than cost per submission.
There is also a floor under this. If a large share of your buyers would rather ring you than fill anything in, form optimisation is fixing the minority route while the majority stays invisible, and the account will keep drifting towards the people who like forms. That is a different break in the same chain, and it usually needs solving first.
And a conversion rate is only as trustworthy as the tracking under it. If the submission fires an event that also fires on a page refresh, or the page sits on a different domain from the form, you are optimising against a number that is partly fictional. Confirm what the tag is actually counting before you celebrate a change, and confirm that the sessions survive any hand-off between domains.
Where this sits
Everything above is one stage of a longer sequence. Working out whether the problem is the auction, the page or the form is the tracing work, and it comes before any spend decision at all. That is the whole of the method we run: prove what you can, repair what is broken, then buy against it.
The uncomfortable version is that campaigns are easy to blame because they are easy to change. The form is somebody else’s ticket in somebody else’s backlog, which is exactly why it sat there costing you money for a year.
Go and fill in your own form on your own phone. Most people have never done it.