TL;DR
- Duplicates split one buyer's activity across records, so each record's score reflects only part.
- A known buyer's new sign-up under another email can become a false MQL.
- A lead score that crosses the threshold late sends a ready buyer to sales late.
- Merge the obvious duplicates before tuning scoring rules or training an AI scoring model.
- Dedupely finds and merges duplicate leads and contacts, so scoring reads one record per buyer.
Lead scoring is supposed to tell sales who to call first. Marketing sets the points or trains a model, sales trusts the number, and the threshold decides when a lead becomes an MQL.
The score is only as good as the record it sits on. When one buyer lives in two or three records, each one holds part of the story, and the score reflects that part. Some ready buyers never reach sales, and some records reach sales that shouldn't.
In this post, you'll see how duplicates distort rule-based and AI scoring, and the approach that fixes it. It's a way of thinking about scoring and duplicates, so it doesn't walk through a scoring setup.
What lead scoring reads
Many scoring models look at two things. Fit is who the person is: job title, company size, industry, country. Engagement is what they do: form fills, email clicks, page views, meetings.
Rule-based scoring
With rule-based scoring, your team decides the points. A demo request might be worth 20 points, a pricing page visit 10, and a job title in your target list another 15. When a record reaches the threshold, it becomes an MQL and goes to sales.
AI or predictive scoring
AI scoring learns from past records. It looks at which leads became customers and which didn't, then scores new leads by how much they resemble the winners.
Both approaches score records. Neither one knows that two records belong to the same person unless something in the data links them.
Three ways duplicates distort scores
How much each one matters depends on your sources, your threshold and how your team routes leads. These are the patterns to look for.
Engagement split across records
A buyer downloads a guide with a personal email, then books a webinar with a work email. Each record gets part of the activity, and neither crosses the threshold. The buyer is ready, and nobody calls.
Fit gets split too: one record has the job title from a form, the other the company size from an import. Neither looks like a strong fit on its own. Leads are especially prone to this, since duplicate leads form faster than contacts.
False MQLs
Duplicates can push scores the wrong way too. A customer or an open opportunity's contact signs up for a webinar with a new email. The CRM creates a new record, the webinar adds points, and the record becomes an MQL.
Sales now gets a "new" lead who already talks to an account manager. The rep calls, the buyer is confused, and sales starts to doubt the MQLs that follow.
Duplicates also inflate MQL counts. One person counted twice makes volume look higher and conversion rates look lower than they are.
Routing that comes too late
Scores decide when a lead goes to sales. When engagement is split, the record that finally crosses the threshold does it days or weeks later than the buyer did.
By then, the moment that made the buyer reach out may have passed. Two records can also cross the threshold separately and land with two different reps.
Why AI scoring learns the wrong patterns
AI scoring adds a second problem. It learns from your history, so duplicates in past records shape what the model thinks a good lead looks like.
A customer whose engagement was split across two records looks less engaged than they were. The leftover duplicate that never converted counts as a lost lead. The model learns that the signals your best buyers showed matter less than they do.
The same split data affects other AI tools in your CRM, which duplicate records can break in similar ways. For scoring, the fix starts before training or retraining a model.
The approach: fix the records before you tune the model
When scores look wrong, the first reaction is often to change the points or the threshold. If duplicates are the cause, new rules only move the problem. Start with the records.
Agree on what counts as a duplicate for scoring
What counts as a duplicate is case by case, and what one team considers a duplicate, another may not. A prospect who comes back a year later might be the same lead for marketing and a new opportunity for sales.
Agree on the cases before merging. Common ones are a person with a personal and a work email, one person asking about two products, and a returning prospect.
Merge the obvious duplicates first
Start with exact matches on email. These are usually the safest to merge, and they're a quick first step.
Then look at people with two emails. Match on first name and last name plus phone or company name, and choose the fields to match on from what your records hold. Review these before merging, since two people can share a name.
Decide what the merged record keeps
Scoring depends on a few fields more than others. Decide which value wins for lifecycle stage, lead status, owner and MQL date before you merge.
Check how your scoring handles a merge. It might recalculate from the combined history, or it might keep one record's score until new activity arrives.
Compare scores and MQL counts after merging
Look at your MQL list before and after a cleanup. Count how many records crossed the threshold once their history was combined, and how many false MQLs disappeared.
That comparison tells you whether your scoring rules were wrong, or whether the records were. Tune the points and the threshold only after that.
Catch new duplicates before they reach the threshold
Forms, imports and integrations keep adding records. Checking for new duplicates on a regular schedule keeps scores from drifting again. Merging new duplicates before routing runs means the record that reaches sales has the full history.
Summary: duplicates and lead scoring
| What happens | Effect on scoring | What to do |
|---|---|---|
| Engagement and fit split across records | Scores too low; ready buyers miss the threshold | Merge exact email matches first, then review name matches |
| A known buyer signs up with a new email | False MQL sent to sales | Match on name plus phone or company to catch second emails |
| Split records cross the threshold late | Routing comes days or weeks late | Merge new duplicates before routing runs |
| Duplicates in past records | AI scoring learns the wrong signals | Clean history before training or retraining |
| One person counted twice | MQL counts inflated, conversion rates too low | Compare MQL counts before and after merging |
How Dedupely brings each buyer's engagement onto one record
With Dedupely, search for possible duplicate leads or contacts, confirm identity and choose the values that survive. Review the activities supported by your connector and preview each merge. Afterward, check the scoring tool's recalculation and routing behavior before enabling automatic merging for a tested rule.
Scores you can trust start with one record per buyer
Lead scoring works on records, so duplicates change what it sees. Split engagement hides ready buyers, and new sign-ups from known buyers become false MQLs. Late scores send leads to sales after the moment has passed, and AI scoring learns from the same split history.
The fix is to agree on what counts as a duplicate, merge the obvious ones and check scores again. Once each buyer has one record, you can tell whether your scoring rules need work or whether the data did.
Frequently asked questions
Do duplicate contacts affect lead scoring?
Yes: lead scoring adds points to a record, so when one buyer has two records, each holds part of their activity and fit. Both records can score below the threshold while the buyer is ready. Duplicates can also create false MQLs and inflate MQL counts.
Why are my lead scores too low?
One possible reason is engagement split across duplicate records. A buyer using personal and work emails can have activity on both. Confirm the match, merge with the right field rules, and check how your scoring tool recalculates. A merge does not necessarily add the two scores together or raise the result.
Should I clean my CRM before setting up AI lead scoring?
Yes, ideally: AI scoring learns from past records, so duplicates can teach it that engaged buyers were less engaged than they were. Merging the obvious duplicates before training or retraining gives the model a cleaner history. Keep checking for new duplicates afterward, since forms and imports keep adding records.
How do duplicates create false MQLs?
A known buyer, such as a customer or someone already talking to sales, signs up with a new email. The CRM creates a new record, the activity adds points, and the record becomes an MQL. Sales then gets a "new" lead who is already in conversation with the team.
What happens to a lead score when two records merge?
It depends on the scoring tool. Some may recalculate the score from the combined activity, while others may keep one record's value until new activity arrives. Check how yours behaves, and decide beforehand which lifecycle stage, lead status, owner and MQL date the merged record should keep.
Or sign up with Google
Salesforce AppExchange 
