Layer 2 · after the lead
⭐ To Google Ads, all your leads are worth the same. They are not.
The form from someone who bought €8,000 and the one with a made-up phone number reach your account as the same conversion: same value, same bid, same weight in the learning. Lead Rating scores each lead, follows what happened next and sends it back to Google — so it stops looking for forms and starts looking for customers.
The change, in one sentence
Your account stops optimising towards the cheapest lead and starts optimising towards the lead that resembles the one who buys from you. It isn't a new report: it is a different instruction for the algorithm.
And the lead that was no good is withdrawn from the account, so it teaches nothing wrong.
The score
0 to 100, built from what that person did before contacting you
The score isn't guesswork: it comes from signals that already exist and that almost nobody collects, let alone cross-checks.
How they arrived and what they did
Where they came in, how many pages they saw, how long they stayed, whether they came back another day, whether they read before writing.
Where they come from
Which campaign, which network and which ad. And whether they arrived through a brand search or a discovery campaign, which are not the same thing.
What they wrote
Whether the message has substance or is four letters; whether the contact details look like they could be real.
And what matters in your business
The area you operate in, the service they asked about, the conversions that genuinely count for you.
No contact details, no lead. A form with no phone and no email isn't a hard-to-reach person: it's a bot. Those are flagged automatically, stay at 0 and never upload to Google Ads — not even as a zero-value conversion, because under "maximise conversions" they would count all the same.
The full funnel
Four moments, four conversions
Day one's score is a prediction. What confirms it is what happened afterwards — and that goes back to Google too.
Raw lead
The form arrives, gets its score and uploads that same night with its value. Your account is already learning from quality, not volume.
Qualified
Your team — or your CRM — confirms this is someone worth working with. Uploads as its own conversion.
Opportunity
A quote sent, a visit booked, or whatever "this is serious" means in your business.
Sale, with its amount
The conversion that actually matters, with real money. And if the sale falls through, it is withdrawn from Google Ads: it doesn't stay there teaching something that never happened.
Each stage has its own job. Some teach the algorithm and others teach you — and it pays to know which is which. That is what the next section explains.
The most expensive misunderstanding
90 days to upload. 7 days for the algorithm to see it.
Most of the market still counts on 90 days, and that is only half true. Understanding this difference changes how you set up your measurement — and it is the reason everything else exists.
| When the conversion is uploaded | Does it show in your reports? | Does Smart Bidding learn from it? |
|---|---|---|
| Within 7 days of the click | Yes | Yes |
| Between 7 and 90 days | Yes | No |
| More than 90 days | No | No |
The 90 days to upload haven't changed. What changed is that attribution modelling only takes into account what arrives within the first 7 days — and that modelling is what feeds automated bidding. This is why a sale imported at month end looks perfect in your reports and still teaches the algorithm nothing.
What arrives in time to teach the algorithm
The raw lead always: the form comes in and uploads that same night, carrying its score. And the qualified stage nearly always, if your team works leads in the first few days — which is what you should be demanding of your sales process anyway.
Opportunity and sale, on the other hand, arrive late in almost any business with a normal sales cycle. And that's fine: those stages aren't there to bid with.
And why you should still mark them
A sale not entering the learning doesn't mean marking it is pointless. It is exactly the opposite: it is what lets you know which campaigns produce customers rather than forms, with the money next to it.
And above all, it is what calibrates the rating: without knowing which leads ended in a sale, the score can neither learn nor be audited. The calibration curve is built precisely from those late marks.
And this is why the rating exists. If the information that really matters — whether the lead is worth anything — arrives late by nature, the only way the algorithm can use it is to predict it on day one. That is why the raw lead doesn't upload as "one more conversion": it uploads with its score. That places inside the 7-day window a quality signal that would otherwise arrive two months late, or never.
What you won't have to do
All the offline-conversion plumbing, solved
This is where most people's measurement breaks: sheets to share, schedules that expire, date formats, misconfigured actions. None of those two dozen manual steps exist here.
The actions create themselves
The four conversions appear in your account configured through the API, with a dry run first so nothing is touched by mistake. No setting them up by hand and no guessing the right options.
And they upload every night
No file in Drive, no import schedule, no service accounts to share and none of the time-zone mess. Automated for real.
Enhanced conversions
Hashed email and phone so Google matches the conversion even when the click identifier was lost. Only with consent, and never in the clear.
Worthless ones are deleted from Google
Marking a lead as bad removes its conversion from the account. The algorithm stops learning from it — and if you were wrong, unmark it and it returns.
Duplicates grouped
The same person writing three times is not three leads. They are grouped with the window you choose, and what you mark on a duplicate goes to the original lead, the one carrying the click.
Nothing uploads twice
Every conversion carries a stable reference, so even if it is re-sent one day, Google recognises it and doesn't double-count.
The practical effect: an onboarding that by hand means two dozen steps in Google's interface — with a silent point of failure at each one — here is configuration that comes done.
Cleaning the signal
And whatever wasn't a customer, out of Google Ads
A lead that was worthless isn't just an annoyance for your sales rep: it is a wrong instruction for the algorithm, which will go looking for more people just like them. The system tells two cases apart and treats them differently.
The one that never gets uploaded
A form with no phone and no email isn't a hard-to-reach person: it's a bot or a broken submission. It gets a 0 and is never sent to Google Ads at all — not even as a zero-value conversion, because under "maximise conversions" it would count just like a good one.
Your account never learns it existed. It is the cheapest defence there is, and it runs on its own.
The one that goes up and is then withdrawn
The one that did carry details but wasn't for you: asking for something you don't do, outside your area, made-up details, or outright fraud. Your team ticks one box and its conversion is withdrawn from your account, so the algorithm stops learning from it.
And it is reversible: if someone marked it by mistake, untick it and the conversion comes back. Same with a sale that falls through: it withdraws itself.
How hard to push is your call, not ours: you can withdraw them from Google Ads entirely, or let them keep counting but worth less. It is one row in your configuration, and you can change your mind at any time.
📊 And they tell you two very different things
Which is why the report doesn't lump them into a single column, the usual mistake:
- Lots of no-contact ones → you have a traffic problem: bots, a form that doesn't validate properly, junk inventory. That's what the Shield fixes.
- Lots discarded by your team → those are real people who don't fit: the problem is in targeting or messaging, not fraud.
Each with its own denominator: the junk over everything that came in, and the discarded ones only over the leads you could actually have contacted.
Not an off-the-shelf formula
Calibrated to your business, recalibrated every month
A good lead doesn't mean the same thing in a clinic, a driving school and a machinery firm. That's why the score doesn't arrive ready-made:
- Your project's metrics are added: the conversions that count, the area you operate in, what separates your good customer from a browser.
- It recalibrates itself monthly with your real results: the ranking doesn't stay frozen on installation day.
- And it can keep being tuned: if a new signal turns out to predict well in your business, it goes into the algorithm.
And it audits itself
A scoring system that never publishes its own accuracy is an opinion with decimals. The report includes the calibration curve: for each score band, what share of those leads turned out useful.
| Band | Leads | % that worked |
|---|---|---|
| 90-100 | — | high |
| 70-89 | — | high |
| 40-69 | — | medium |
| 0-39 | — | low |
With your data and the cumulative cut-off: "working from 70 up, you capture X % of your good ones". If the curve doesn't discriminate, you see it — and it gets recalibrated.
Your team
It connects to the CRM you already use. Or you don't need one.
What costs a salesperson effort is changing tools. Here they don't change anything.
If you have a CRM
Your CRM notifies when a lead changes status and the system does the rest. It works with the usual ones — HubSpot, Zoho, Pipedrive, Salesforce, Odoo, Bitrix24 — and with an in-house one, if your developer can send a notification.
Including the operation almost nobody has: a cancelled sale unmarks the sale and withdraws it from Google Ads.
If you don't
The sheet works as your inbox: every day, leads sorted by quality, with what is known about each one and four checkboxes to mark what happened. What arrives first is what is worth most.
And the forms that aren't leads — account logins, cart, search, newsletter — are discarded before they arrive.
Personal data
Consent isn't an add-on: it decides what travels
For every lead the system knows whether they accepted advertising, and that governs what leaves your systems. If they didn't, you choose between two reasonable options:
Send nothing about that person — or send only their click identifier: the conversion counts for your campaigns and neither their email nor their phone travels, and it is declared as denied so it never feeds audiences.
The default is always the most conservative one. How data is handled →
Towards ChatGPT Ads too
If you advertise on ChatGPT, the same system sends it your quality stages through its conversions channel. With one difference worth knowing: there value-based bidding doesn't exist, so what really changes its behaviour is choosing which event counts — and for that you need the stages measured, which is exactly what this does.
Next step
How many of your leads are actually any good?
If the answer is "I don't know with numbers", that is exactly the problem — and also the easy part to fix. Tell us your volume and how your measurement is set up.