Research question
What combination of evidence is strong enough to flag two QuickBooks Online customer records as a probable collision while stopping short of a merge decision? This is narrower than asking how to clean a customer list. It treats duplicate detection as an uncertain record-linkage problem. Names can match while organizations differ, and names can differ while the underlying customer is the same. An administrative bookkeeping support role needs a defensible way to distinguish a useful lead from a conclusion that could alter transaction history.
Intuit documents a merge function for duplicate customer profiles and warns that merging cannot be undone. That product behavior makes false positives consequential. The relevant research target is therefore not the number of similar names. It is the quality of the evidence connecting each proposed pair and the reasons a reviewer might reject the connection.
Competing explanations for similarity
A repeated display name can arise from several mechanisms. One profile may have been imported and another entered manually. A customer may trade under a business name that differs from its legal or billing name. Two branches may share a parent name but require separate records. Spelling, punctuation, abbreviations, and contact changes can also create near matches. Conversely, a common surname or generic company name can connect unrelated customers.
The analysis should preserve at least two competing explanations for every candidate pair: same customer represented twice, or distinct customers with overlapping attributes. A third state, insufficient evidence, is important because uncertainty is not a failure. It prevents a support analyst from converting a similarity score into an unauthorized master-data change.
Methodology: pairwise record linkage
This study uses a documentary and design methodology rather than private file testing. Product behavior is drawn from Intuit's published customer-merge guidance. Record reliability and control concepts are compared with the GAO Standards for Internal Control, while retention and support context comes from the IRS recordkeeping page. Small-business record use is framed by the SBA finance management guide. The product-specific reference is Intuit guidance on merging duplicate customers.
The proposed unit of analysis is a pair of customer profiles, not an entire customer list. Each pair is compared on normalized display name, company name, email domain, telephone digits, billing address, shipping address, and the presence of linked activity. Normalization may remove case and punctuation for comparison, but the original values remain visible. No single field is treated as decisive. The pair receives an evidence narrative describing agreements, disagreements, missing values, and the source date of each observation.
Candidates can be selected through two passes. An exact pass identifies pairs that agree on a high-specificity field such as a complete email address. A fuzzy pass identifies close names only when another field also agrees. This is a blocking strategy, not a verdict. It reduces the number of pairs requiring review while retaining an explicit record of why each pair entered the candidate set.
Evidence hierarchy
Evidence should be weighted by discriminating power and authority. A complete, current billing email supplied through an approved customer record usually distinguishes profiles better than a first name. A full address may be informative, but shared offices and branch locations weaken it. Telephone numbers can be reassigned or shared. Transaction descriptions are contextual clues, not identity proof. A prior staff note is useful only when its source and date are known.
Conflicting fields deserve more attention than a simple count of matches. Two profiles with the same company name but different tax identifiers or independently confirmed billing contacts should not be collapsed by arithmetic. The support record should state which fields conflict and route the pair to the person authorized to interpret the business relationship. Sensitive identifiers need not be copied into the comparison table. A controlled reference to the approved source can establish that a conflict exists without duplicating the value.
Error costs and decision thresholds
Pairwise linkage has two error types. A false positive treats distinct customers as one candidate, creating pressure toward an improper merge. A false negative leaves a duplicate undetected. These errors are not symmetric because an irreversible merge can combine histories, while an unreviewed duplicate may remain available for later investigation. Intuit's warning about irreversibility supports a conservative review threshold.
A three-band interpretation is more transparent than a single score. A strong candidate has multiple independent agreements and no material conflict. A review candidate has some agreement plus missing or stale fields. A rejected candidate has a clear conflict or only a weak name resemblance. The bands describe evidence strength. They do not authorize a merge, choose which profile survives, or determine how open activity should be handled.
Administrative support boundary
A QBO administrative specialist can extract candidate fields, normalize comparison copies, preserve originals, attach source references, and prepare rejected as well as accepted candidates for review. Including rejected pairs matters because it reveals whether blocking rules repeatedly surface the same harmless pattern. The specialist can also record the reviewer, review date, decision, and reason without performing the irreversible action.
The accountable owner or qualified reviewer decides whether the records represent the same customer, whether separate branches should remain separate, and whether a merge is appropriate. Any decision involving legal identity, tax records, disputed balances, or retention requirements belongs outside an administrative inference. This separation keeps the research useful without presenting identity resolution as bookkeeping or legal advice.
What findings would be persuasive?
A persuasive finding would show that candidate pairs with agreement across independent, current fields are confirmed more often than name-only pairs. Another useful finding would identify a recurring source of collisions, such as imports that omit an email address or inconsistent abbreviations from one intake channel. Reviewer disagreement is also evidence. If two authorized reviewers reach different conclusions from the same pair, the definitions or source authority may be unclear.
Counts require denominators. Reporting twelve flagged pairs says little without the number of profiles screened, the selection rules, and the number rejected. Confirmation rates should be reported by evidence band rather than pooled. Missing fields should remain a separate category because absence of evidence cannot be treated as disagreement.
Limitations
This article proposes an analytical design and does not report results from a customer dataset. Public product documentation may change, and available fields vary by subscription, configuration, import history, and connected systems. Similarity does not establish legal identity, ownership, collectibility, or correctness of prior entries. A reviewer-confirmed match can still require additional handling before any product action.
The approach may under-detect duplicates when records contain little contact data. It may over-select families, franchises, property entities, or branches that share details. Normalization can erase meaningful punctuation or suffixes. Transaction history can provide context but can also carry earlier data errors. Results from one customer population should not be generalized to another without recalibrating the candidate rules and reviewing local privacy requirements.
Conclusion
The evidence supports treating duplicate-customer detection as pairwise record linkage with competing explanations, not as a name-cleaning exercise. Strong candidates combine independent, current agreements and disclose conflicts, missing fields, and source dates. Conservative bands are appropriate because Intuit describes merging as irreversible. QBOAssistant can prepare the comparison and preserve the decision trail, while an authorized reviewer determines identity and any merge action. The most informative measure is not how many names look alike, but how consistently documented evidence separates confirmed collisions from distinct customers.