Pardot and Account Engagement · lead scoring and grading
Your leads are scored.
They are not qualified.
A score built on clicks rewards the people studying the problem and hides the people preparing to buy. Rebuilding it means separating interest from fit, making inactivity subtract, and setting the qualification line against deals that actually closed rather than against an opinion.
€2,250 for 45 hours, the same rate as everything else here.
The price, and what is in it
Forty five hours, two weeks, and sales signs the threshold.
Discovery with both teams, intent mapping against your last ninety days of closed deals, the model itself, negative scoring and decay, validation before handover, and the routing that delivers the result. Documentation and sixty days of support after handover are in the price rather than added to it.
€2,250 · 45 hours · about 2 weeks elapsed. The same work reaches this page from the other direction at sales ignores marketing leads, and it is the same price, because it is the same work.
That number sits below what a partner would quote, and it is deliberate rather than a discount. There is no delivery manager to fund, no solution architect writing a document that somebody junior then configures, and no partner margin. You are paying for the person doing the work. The market rates are published here so you can place it yourself. Above roughly 300 seats, or where procurement requires a partner tier on the invoice, a partner is the better answer and I will say so.
Why models break
Six architectural reasons, and none of them are a tooling problem.
A scoring model is not a set of fields, it is the logic that decides who gets sales attention and when. When that logic is wrong, every system downstream reports confidently on the wrong thing.
One model for every buyer
Every prospect is evaluated the same way, although deal sizes, buying journeys and sales motions are not the same. A self serve trial and a six month enterprise evaluation produce completely different behaviour, and one threshold cannot describe both.
Nothing ever subtracts
Inactive or low intent behaviour never reduces the score, so a lead that went quiet in March is still marked hot in August. Sales works it, gets nothing, and starts discounting the score on every other lead too.
Interest is read as readiness
Content engagement is treated as buying intent. Somebody who downloads three guides outranks somebody who opened the pricing page twice, and the model has no way to tell the difference between studying a problem and preparing to buy.
Sales never signed the threshold
The number that defines a marketing qualified lead was chosen by marketing alone and never checked against deals that actually closed. Two teams then argue about lead quality using two different definitions of quality.
The model does not know the stage
Prospects move through the funnel but scoring logic does not adapt to where they are. The same click means one thing before a demo and something else entirely after one.
Built once, never recalibrated
Buyer behaviour keeps moving and the model does not. Drift is invisible month to month and obvious after two years, by which point nobody remembers which rule was supposed to do what.
What a working model does
Four things separate a model that works from one that looks sophisticated.
Scoring and grading stay separate
Engagement shows motivation. Grading shows whether the company can buy at all. Both have to be true before sales is involved, and mixing them into one number hides which one is missing.
Intent signals outweigh engagement noise
A pricing page view, a demo request and repeated product research carry weight. A newsletter click does not. The list of what counts is agreed in discovery, not inherited from a template.
Scoring knows the lifecycle stage
The same behaviour scores differently depending on where the prospect already is, so the model stops rewarding people for repeating what they did last month.
The threshold is back tested
The qualification line is set against your own closed deals from the last ninety days, then agreed with sales in writing. If top quartile leads do not convert better than mid quartile, the model is not finished.
How the work runs
Six steps, and the first one is the one that decides the rest.
The order matters. Agreeing what qualified means, with the people who close deals, comes before any rule is written. Every engagement that skipped that step produced a model marketing believed and sales ignored.
Discovery with both teams
What counts as qualified, in the words of the people who close deals. Sales agreement here is the prerequisite for everything after it, not a formality at the end.
Intent mapping against closed deals
Behavioural events mapped to actions that preceded real purchases, using your last ninety days of closed opportunity data rather than a generic list of engagement types.
The model itself
Engagement scoring, fit grading and lifecycle logic built separately, within the constraints of the edition you actually have.
Negative scoring and decay
Rules that remove priority from leads that went quiet, with decay tuned to the length of your real sales cycle rather than a default.
Validation before handover
The model is run against historical opportunities. Top quartile has to outperform mid quartile in conversion, and if it does not, the model changes.
Operational handover
Routing, alerts and ownership so sales receives a lead with the context attached. Written documentation and sixty days of support are in the price.
When this becomes urgent
Six conditions that turn a tolerable model into an expensive one.
Scoring drift is invisible month to month. It becomes visible when volume, motion or product mix changes, and by then the model has been quietly wrong for a while.
Paid or inbound volume is growing
More leads arrive and quality drops, because the model cannot separate intent from noise at any volume above what it was built for.
You are moving to account based motion
Scoring works at contact level only, and account level context does not exist anywhere in the model.
Sales has stopped using the score
Reps re qualify everything by hand. Response time stretches and the two teams stop agreeing on what happened.
Attribution is not defensible
Marketing cannot explain what drove revenue because the signals were never structured to answer that question.
The setup was done quickly
Automation conflicts, scores that do not match reality, and logic nobody documented at the time.
The business got more complicated
New products, regions or buyer roles arrived and the original single model never accounted for any of them.
Asked often enough to answer here
Questions
How much does a Pardot lead scoring rebuild cost?
45 hours at 50 euro an hour, so 2,250 euro, about two weeks elapsed. That covers discovery with both teams, the model, negative scoring and decay, validation against your closed deals, and the routing that delivers the result to sales. Documentation and sixty days of support after handover are in the price rather than added to it.
Custom scoring, Einstein, or Agentforce: which one is right?
It is decided by edition, data volume and whether an AI roadmap actually exists, not by which is newest. Custom rule based scoring works on any edition and is auditable line by line. Einstein needs Plus and enough historical conversions to learn from. Agentforce based scoring needs Advanced or Account Engagement Growth on the newer platform. Most teams asking about the AI options have a rule problem, not a model problem.
How long does it take?
About two weeks of elapsed time for 45 hours of work. The part that sets the pace is not the build, it is getting sales to agree what qualified means and getting access to the closed deal data the thresholds are tested against.
Do you rebuild the model or fix the one we have?
Whichever is cheaper for you, and that is decided after looking, not before. A model with sound structure and wrong weights is a repair. A model with one track for every buyer and no negative scoring is a rebuild, because the structure is the thing that is wrong.
What happens if the new model does not convert better?
It is tested against your historical opportunities before handover, so that question is answered before you rely on it. If top quartile leads do not outperform mid quartile in that test, the model is changed until they do or you are told it cannot be done with the data available.
Serhii Skrypnyk · Senior Salesforce Administrator and developer · 7 Salesforce certifications · on the platform since 2018. Reviewed 26 August 2026.