Most paid media accounts are optimizing toward a form fill. The bidding algorithm sees a submission, calls it a conversion, and goes looking for more people like that person. Nobody ever tells it which of those submissions turned into a customer.
So it finds more form fills. Cheaper ones. Worse ones.
This is the single most common gap we see in accounts across the roughly $100M+ in managed spend we have worked in. The campaigns are fine. The creative is fine. The feedback loop is missing.
Why cost per lead keeps dropping while revenue stays flat
Smart Bidding and Meta's Advantage campaigns are prediction engines. They predict who is likely to do the thing you told them to count. If the thing you told them to count is "submitted a form," they will get very good at finding people who submit forms.
Some of those people are tire kickers. Some are competitors. Some are in the wrong service area. Some wanted a job, not a quote.
The platform cannot tell the difference. It has no idea what happened after the click. Your CRM knows. Your sales team knows. Your point of sale knows. That knowledge just never travels back.
The symptom is familiar. Cost per lead looks better every quarter. Sales volume does not move. The paid media report and the revenue report tell two different stories, which is a version of the attribution problem we wrote about here.
Define "qualified" before you automate anything
Do not start with tooling. Start with a definition, in writing, that a salesperson and a marketer would both sign.
Pick one primary outcome and one or two backup signals:
- Services business: an appointment that was actually held, not just booked.
- Multi-location retail: a lead that walked into a store, or a purchase matched back to an email.
- Hospitality: a completed stay or a booking that did not cancel.
- Considered B2B purchase: a lead your team marked as sales qualified after a discovery call.
The backup signal matters because your primary outcome may take too long. A ninety day sales cycle is too slow to train a bidding algorithm. If closed revenue lags badly, use the fastest reliable proxy you have. "Discovery call held" arrives in days and correlates well enough to steer bidding.
Write the definition down. Put an owner on it, the same way you would put an owner on any number in a marketing scorecard.
The four pieces of a working loop
Every version of this, on every platform, comes down to the same four parts.
1. A click identifier that survives the trip
Google appends a GCLID to the landing page URL. Meta appends an FBCLID. These are the only reliable way to tie a specific conversion back to a specific ad click.
Capture the parameter in a hidden form field. Pass it into the CRM as a field on the lead record. If your form tool strips URL parameters, fix that first, because nothing downstream works without it.
If you cannot capture click IDs, the fallback is hashed customer data. Email address and phone number, normalized and hashed, matched by the platform. It works, at a lower match rate, and it is the right path for in-store purchases where there was never a form at all.
While you are in the form, confirm your UTM parameters are consistent. The click ID drives the bidding. The UTMs drive your reporting. You want both.
2. A CRM field that records the outcome
One field. A short list of values. Something like: new, contacted, qualified, appointment held, won, lost, junk.
Resist the urge to build twelve statuses. Reps will not maintain twelve statuses, and a field nobody updates is worse than no field, because it looks like data.
3. A scheduled upload back to the platform
This is where most teams stall, and it is the easiest part to automate. Google Ads and Meta both accept scheduled imports. Google can pull a Google Sheet or an SFTP file on a recurring schedule. Meta takes offline event uploads through its API or a scheduled file.
Daily is ideal. Weekly is fine. Manual and occasional is not, because the algorithm needs a steady signal, not a quarterly dump.
This is exactly the kind of unglamorous plumbing that belongs in your marketing systems layer rather than on someone's task list.
4. A value, not just a yes or no
A qualified lead flag is good. A dollar value is better.
If you know actual revenue, send it. If you do not, send a modeled value. Average first order value by service line works. A roofing inspection and a full roof replacement should not carry the same number.
Once values are flowing, you can move from target cost per acquisition to target return on ad spend, and the math in a ROAS calculation starts describing your actual business instead of your form volume.
Google Ads specifics
Create a new conversion action for the qualified outcome. Import it under offline conversions. Set the new action as your primary and demote the raw form fill to secondary so it still reports but no longer steers bidding.
Give it two to four weeks before you judge anything. Volume drops at first. That is the point. You removed the junk the algorithm was chasing.
One caution. Smart Bidding needs enough conversions to learn. If your qualified count per campaign is very thin, consolidate campaigns before you switch the primary action, or bid to the faster proxy signal instead.
Meta specifics
Use the Conversions API with an offline or CRM dataset, and send the lead outcome as a custom event with a value. Match on click ID first, hashed email and phone as backup.
For lead form campaigns, Meta's conversion leads optimization needs that outcome data flowing before it can do anything useful. Turning it on without the feedback loop changes nothing.
Four ways this goes wrong
Reps stop updating the field. Build the status update into the workflow they already use. If it lives in a separate tool, it will decay.
Junk leads never get marked junk. The loop only works if bad outcomes flow back too. Marking losses is as valuable as marking wins.
The upload silently breaks. A sheet gets renamed, a token expires, nobody notices for six weeks. Put an alert on the daily row count, not on a dashboard nobody opens.
You judge it in week one. Learning periods are real. Set the review date when you launch and hold to it.
Where to start this week
One hour, three moves.
- Open your lead form and check whether GCLID and FBCLID are being captured. If not, that is your whole first sprint.
- Pull last quarter's leads and have sales mark them qualified or junk. Look at which campaigns produced which. That single view usually changes a budget decision on its own.
- Write your one-sentence definition of qualified and get sales to agree to it.
If you want a broader sweep of the account first, our paid media audit checklist covers the structural items to fix before you touch bidding, and our managed paid media service builds this loop as part of the engagement, in accounts you own.
Start with a Diagnostic Call
If your cost per lead keeps improving while revenue sits still, the feedback loop is almost always the reason. We can trace where the signal breaks in your setup and tell you what it takes to close it.
Start with a Diagnostic Call, or read more about what the diagnostic covers.