An ad platform optimises toward whatever you tell it a conversion is. Tell it a form submission counts and it will find you the cheapest possible form submissions, which is a genuinely different objective from finding customers. Most accounts that plateau have not run out of demand. They have taught the bidding algorithm to chase the wrong outcome very efficiently.
Lead quality scoring is the correction. It means deciding what a good lead actually is, measuring it, and sending that judgement back to the platform so the model learns from it.
Why cost per lead stops being useful
Cost per lead is a fine metric while every lead is roughly equivalent. It stops being useful the moment your leads vary in value, which for most service businesses is immediately. A campaign delivering leads at half the cost can easily be the least profitable one in the account if those leads are out of area, out of budget, or looking for something you do not sell.
The symptom is familiar: the paid reports look healthy, the sales team says the leads are poor, and neither side can prove it. That argument is a measurement gap rather than a disagreement.
Defining the score before you build anything
- Agree what disqualifies a lead outright. Wrong area, wrong service, no budget, a supplier or a job seeker.
- Agree the two or three attributes that predict value. Usually service type, urgency and job size.
- Decide the point in your process where quality becomes knowable, which is normally after first contact rather than at submission.
- Keep the scale short. Three or four tiers are actionable; a hundred-point model is a project nobody finishes.
The temptation is to build something sophisticated. A simple tier that everyone applies consistently beats an elaborate model that only one person understands.
Every account we take over is optimising toward something the business would not recognise as a sale. The fix is rarely clever targeting. It is telling the platform the truth about which leads were worth having.
Aisha Karim, Analytics & CRO Manager, Media @ Marsons
Feeding the score back to the platform
Scoring leads changes nothing until the score reaches the bidding model. Offline conversion import is the mechanism: your CRM sends back which leads became qualified opportunities and which became revenue, tied to the original click. The platform then optimises toward the outcome you actually care about instead of the proxy you set up on day one.
This requires the click identifier to be captured at form submission and carried through your CRM. It is a small piece of plumbing and it is the difference between a bidding strategy that improves over time and one that gets very good at buying noise.
Give the model enough to learn from
Smart bidding needs a reasonable volume of the event it optimises toward. If qualified leads are rare, optimising directly toward closed revenue will starve the model and performance gets worse rather than better. In that case optimise toward the highest-volume reliable signal, and use revenue data to judge the account rather than to steer it.
What the scoring will expose
Usually two things. That a campaign everyone assumed was the best performer is delivering the cheapest and least valuable leads, and that a campaign nobody liked is quietly producing the revenue. Both are uncomfortable and both are worth knowing before the next budget conversation.
It also frequently reveals that the problem is not the ads at all. If good leads arrive and never get called back, no amount of bidding work will help. For how this differs by market and sector, the PPC hub indexes the pages covering each one.
Starting small
You do not need a data warehouse. A weekly review where sales tags each lead as junk, valid or won, applied consistently for six weeks, produces enough signal to reallocate budget with confidence. Build the automated version once the manual one has proved which attributes actually matter.
Optimising toward the wrong lead?
We will map your CRM outcomes back into your ad account so bidding learns from booked revenue rather than from form fills.
