How to Reduce False Positives in Dedupe Matching (Without Losing Real Duplicates)

You set up your deduplication job, hit run, and then the noise starts. Records that are clearly different get flagged as duplicates, and your team spends the afternoon clicking Not a Duplicate over and over.

False positives are one of the most common complaints in data quality work, and they carry a real cost. Once people stop trusting the review queue, they skip it, and the whole effort quietly falls apart.

False positives aren't a sign your matching engine is broken. They're a sign your matching configuration needs tuning. This is how to do that.

Start Tight, Not Loose

The most common mistake is starting with broad fuzzy matching across every field. It feels thorough, but it creates enormous noise. A record for "John Smith at Acme Corp" and "Jonathan Smithfield at Acme Construction" might score as a probable duplicate when they clearly aren't the same person.

A better approach: start with exact-match filters on your most reliable fields. Email address is the gold standard — if two leads share the same email, they are almost certainly duplicates. Phone number is a solid second. Company domain works well for account deduplication.

Once you've caught the obvious matches, introduce fuzzy logic to handle variations like typos and nickname differences. Build from certainty outward. Don't start fuzzy and try to tighten things up after the fact.

Fix the Empty Field Rule First

This one setting is responsible for more false positives than anything else, and it's almost always overlooked.

By default, the Empty Field Rule in a Plauti scenario is set to Disregard. That sounds harmless, but it means a field gets ignored completely when calculating the matching score if it's empty. So if a contact has no phone number on file, that blank field doesn't lower their score at all.

The result: a record with just a first name and a postal code can match at 100% against another record with the same first name and postal code, even if they're completely different people.

The fix is simple. Go into your scenario settings and change the Empty Field Rule from Disregard to Score 0%. Now empty fields actively bring the score down. That single change often cuts your false positive count significantly on the very next job run.

Choose the Right Matching Method for Each Field

Not every field should use the same matching method. This is where a lot of configurations go wrong.

Email addresses should use Exact matching. Apply fuzzy logic to an email field and you'll start flagging records that merely share a domain as potential duplicates — technically "similar," but not a meaningful match signal.

The same logic applies to external IDs, EINs, or any unique identifier your organization uses. Use Exact matching there and assign those fields a high weight. If two records carry different values in a field that's meant to be unique, they're almost certainly not the same record.

For names, fuzzy matching makes sense. A First Name field benefits from synonym matching so that Bob and Robert get caught. A Last Name field works well with a partial fuzzy approach to handle minor typos. Company names are trickier: "IBM" and "International Business Machines" are the same company, but a loose fuzzy match might also connect "IBM" to "IB Medical." The Company Name matching method in Plauti is built specifically for this kind of comparison and handles it far better than plain fuzzy text.

Match the algorithm to what the field actually contains. That single habit removes a large share of false positives before you even touch your threshold settings.

Plauti's matching methods guide covers which methods apply to which field types in detail.

Raise Your Threshold

If your scenario threshold sits at 60% or 70%, you'll see a lot of noise. That range is broad enough to catch real duplicates, but also broad enough to pull in records that just happen to share a few common values.

Tightening your threshold to 80–85% is a good starting point for most scenarios — it requires a closer overall match before a pair gets surfaced for review. Combine a tighter threshold with a corrected Empty Field Rule, and you'll see a meaningful drop in false positives right away.

For automated merging without manual review, set your threshold higher still: 99–100% is recommended. Only auto-merge records when you're very confident. A wrong merge is much harder to undo than a missed duplicate.

Use Secondary Fields to Break Ties

A scenario built on two or three fields is fragile. Two records matching on name and record type alone isn't enough signal to call them duplicates with confidence.

Add more discriminating fields: postcode, email, phone, city, street, job title. The more concrete identifiers you include, the harder it becomes for two unrelated records to score above your threshold. When primary fields are close but not conclusive, secondary fields like state or department can be the deciding factor.

Weight your most unique fields the highest. Email and phone should outweigh first name. Postcode should outweigh city. The hierarchy of your field weights matters as much as which fields you include.

Add a Frequent Words List

Some words show up in nearly every record of a certain type: legal entities like "Inc," "LLC," and "Ltd.," or common industry terms like "Consulting," "Restaurant," or "Services." These words don't help distinguish one record from another, but if they're included in your matching logic, they push scores up and create false matches.

In Plauti, you can configure a Frequent Words list per field. Words on that list get excluded from the matching calculation, so "Willson Architects" compares against "Willson & Son" as "Willson" vs. "Willson & Son" — a much more honest score.

One thing to verify: check that Activate Improved Frequent Words Handling is switched on. You'll find it under Setup > Settings > Critical Updates. Without it, frequent words may not be excluded correctly in every matching scenario, even if you've already set up the list.

Use Match Scores as a Gate, Not a Guarantee

Most deduplication tools give you a similarity score for each record pair. The temptation is to auto-merge everything above a certain threshold — that's where false positives turn into false merges, a much bigger problem.

A tiered workflow works far better:

• Auto-merge only records that hit a very high confidence threshold on high-certainty fields, like an identical email address.
• Queue for review everything in the middle range, so a human confirms before any merge happens.
• Auto-dismiss pairs that fall below your minimum threshold.

The records your team reviews should be genuinely ambiguous, not obvious non-duplicates. If your review queue is full of clearly different records, your lower threshold is set too low.

Let AI Handle the Gray Zone

Even with well-tuned scenarios, some record pairs sit in an uncertain middle range — not obvious duplicates, not obviously distinct either. That gray zone is where review time piles up fast.

AI Match Recommendation is built specifically for this problem. After running a job, you can apply it to the results: it analyzes each duplicate pair across all fields in context and labels it as Duplicate, Not Duplicate, or Uncertain, with a confidence score and a reason.

Your team then focuses only on the pairs flagged as genuinely uncertain. Everything called clearly duplicate or clearly not a duplicate gets handled automatically — it doesn't replace human judgment, it focuses it on where it actually matters.

AI Match Recommendation is available on Plauti's higher-tier plans. Used alongside auto-merge, it lets you safely lower your auto-merge threshold slightly, to around 97–98%, because the AI recommendation adds an extra layer of confidence.

When You Still Get False Positives, Use the Discard Feature

Even after tuning your scenario, some false positives will come through. That's normal.

When you come across a pair that clearly isn't a duplicate, mark it as False Positive in the job results view or the live matching view. That specific pair won't appear in any future job results. Each discard is specific to that pair, so both records remain eligible to match with other records going forward.

If your matching criteria change later and you want to reconsider previously discarded pairs, remove them from the Discards tab. They'll re-enter the matching pool on the next job run.

Review, Adjust, Repeat

Deduplication configuration is not a one-time project. Your data changes, your processes change, and what worked six months ago might generate more noise today.

Build a habit of reviewing your false positive rate after each job run and adjusting one variable at a time. Change one field's matching method, or raise the threshold by five points, then run the job again and compare. Change too many things at once and you won't know what actually helped.

Plauti's scenario setup guide is a useful reference to keep open while you're tuning. If you're starting from scratch, do a record analysis first: list your objects, identify your most discriminating fields, and plan your weight hierarchy before you build anything. It saves a lot of back-and-forth later.

Quick Reference: Common Causes and Fixes