Every lead gen account I’ve audited that was “underperforming” had the same thing wrong with it, and it was almost never the bidding strategy. The account was optimising beautifully toward a conversion definition that had nothing to do with revenue.

Smart Bidding does exactly what you ask. If you tell it a conversion is a form submission, it will find you the cheapest possible form submissions — and the cheapest form submissions are, reliably, the worst ones. Your CPA falls, your sales team gets busier and closes less, and the report looks better every month while the business gets worse.

The fix isn’t a different bidding strategy. It’s better signal.

Why the algorithm optimises toward junk

Three mechanics stack up:

Cheap conversions are easy to find. There is a large, always-available population of people who will fill in a form without any intention of buying: comparison shoppers, students, job seekers, competitors, people who wanted the PDF. The model finds them because they’re the lowest friction path to the thing you asked for.

Bots and form spam count as conversions. Unprotected forms generate submissions that Google records as successful outcomes and then goes looking for more of.

There’s no feedback loop. Your CRM knows which leads closed. Your ad account doesn’t. Without a connection between them, the platform is optimising with half the information and no way to learn it was wrong.

The account isn’t broken. It’s obediently maximising a badly specified objective.

Step one: stop counting junk as a conversion

Before anything sophisticated, harden the form. Google’s own guidance on invalid leads points at three things, and they’re the right three:

  • CAPTCHA or an equivalent bot check. Invisible challenges are fine; the goal is that automated submissions don’t reach your conversion tag.
  • Double opt-in or email verification, where the funnel tolerates it. A confirmed email is a meaningfully better conversion event than a typed one.
  • Server-side validation. Validate the submission on your backend and fire the conversion from there rather than from a browser event on the thank-you page. This also removes a whole class of accidental double counting. Mechanics in server-side tracking.

While you’re in there, audit what’s actually firing. The most common findings: a form-load event counted as a submission, a thank-you page reachable by back button counting twice, and two conversion actions for the same event both marked primary. None of these are exotic and all of them corrupt the bidding.

Step two: send the outcome back

This is the step that changes accounts, and most accounts skip it.

Offline conversion imports. Capture the Google Click ID (GCLID) when the lead is created, store it on the record in your CRM, and import back the events that actually matter — qualified, opportunity, closed-won — with values attached. The bidding model then optimises toward deals instead of form fills.

The practical requirements are small and the failure points are consistent:

  • Your form must capture and persist the GCLID (and wbraid/gbraid for iOS app traffic).
  • Your CRM must have a field for it that doesn’t get wiped on lead merge.
  • The import must run on a schedule, not once as a proof of concept.
  • The events need to arrive inside the conversion window. Very long sales cycles need the intermediate stage — usually “sales qualified” — as the optimisation target, with closed-won imported for reporting.

Value, not just a flag. Importing “qualified: yes” is better than nothing. Importing “qualified, estimated value USD 4,200” lets value-based bidding distinguish a small deal from a large one. If you don’t have deal values at lead stage, use an average by segment — an imperfect value beats no value.

Multi-source conversions. If you’re bringing conversions in from more than one source, they become biddable after roughly 14 days. Check the event definitions and how duplication is handled before you switch bidding to them. More recorded events are only useful if they represent outcomes you actually want.

Step three: only then, more automation

Google is piloting Lead Intent Scoring and journey-aware bidding, aimed at exactly this problem: longer sales cycles where the conversion event and the revenue event are far apart. The pitch is that the model weighs lead quality signals rather than treating every form fill as equivalent.

Worth pursuing. Worth being clear-eyed about the prerequisite, though: these features consume the same signal you should already be sending. An account that can’t distinguish a qualified lead from a junk one in its own CRM export will not get a better answer from journey-aware bidding. Map your enquiry-to-close journey first — stages, timings, typical values — and you’ll both qualify for the pilot and know what to do with it.

The principle generalises: better lead signals come before more automation. Every automated bidding product is a multiplier on your conversion definition. Multiplying a bad definition harder is not an improvement.

What good looks like

A lead gen account in reasonable shape has:

  • One primary conversion action, defined at the stage where you can tell good from bad — not at form submission unless your form is genuinely high friction.
  • Server-side conversion firing with bot protection in front of it.
  • GCLID captured and stored, with a scheduled offline import running.
  • Values attached, even estimated ones.
  • A monthly reconciliation between leads the platform reports and leads that reached the CRM. The gap is always larger than people expect, and it’s where the problems live.
  • Someone looking at close rate by campaign, not just CPA by campaign.

That last one is the tell. If nobody in the business can say which campaign produces leads that close, the account is being managed on a proxy.

The short version

Your bidding isn’t bad. Your conversion definition is. Harden the form so junk stops counting, push CRM outcomes back into the platform so the model learns what a good lead looks like, attach values so it can tell a big one from a small one — and only then reach for the automation that promises to solve lead quality for you.

Related: what to ask before hiring a performance marketer and why last-touch attribution breaks on long cycles.

If your CPA is falling and your sales team is complaining, that gap is usually the first thing I look at.