Data Validation vs Verification

The email address has an @ sign and a valid domain. It belongs to no active mailbox. The postcode exists and the street line follows the expected format. The customer moved two years ago. The professional license number contains the right number of characters. It was revoked last month.

Each value can pass a system rule and still be wrong for the decision in front of you. Formatting checks create false confidence when a green tick is interpreted as proof of truth.

Validation asks whether a value follows a rule. Verification asks whether suitable evidence supports the claim. Cleansing makes records structurally consistent and removes or resolves known defects. The words are not interchangeable. Once the distinction is clear, the business can stop spending equal effort on every field and verify the records where a mistake has a real financial, legal, safety or customer consequence.

The record was clean, and it was still wrong

Consider an ecommerce order. The checkout validates that the postcode matches a permitted pattern, the country is present and required address lines are not blank. A cleanup routine standardizes capitalization and removes extra spaces. The package still goes to an old residence because neither operation checked whether the address is current for that customer or represented in the delivery network.

The system did not fail. It answered a narrower question than the fulfilment team assumed. “Valid” meant structurally acceptable, not confirmed current and deliverable.

The same gap appears in claims, contact databases, supplier onboarding, professional credentials and customer-service records. The remedy is not to label every field “verified.” Define the claim, acceptable source, date, method and level of confidence for the use case.

Validation, verification and cleansing are different

Validation: does it look right?

Validation tests a value against a rule or allowed domain. An email contains the required components. A date can be parsed. A country code belongs to the approved list. An invoice total equals its line items. Validation may be performed at capture or before import. It can reject impossible structure quickly, but it does not establish that the value describes reality.

Verification: does evidence support it?

Verification compares the claim with a source or process suited to that claim. A company number is checked in the relevant official register. A license is checked with the issuing body. A customer confirms a changed address through an authenticated account flow. Verification can return confirmed, contradicted, not found, inconclusive or unavailable; forcing every result into yes or no hides uncertainty.

Cleansing: is it consistent and usable?

Cleansing standardizes representation, resolves known duplicates, removes unintended characters and repairs defects under agreed rules. It may convert “United States,” “US” and “U.S.” to one stored value while preserving the original source. A clean record is easier to match and analyze. It is not necessarily true.

Audit your current systems using the three test questions. A required-field rule and regular expression are validation. A deduplication job and format normalization are cleansing. A dated comparison with an acceptable external or first-party source is verification. Most CRMs perform the first, can assist with the second and perform little of the third unless a specific service or workflow is added.

What can be verified, and against what?

Postal addresses

Verify the aspect needed. In the United States, USPS Delivery Point Validation can indicate whether a ZIP+4-coded address is represented as a known address in the USPS delivery file. USPS’s DPV information does not prove that a named individual currently lives there. CASS certifies address-matching software under USPS tests; USPS expressly notes that CASS processing does not measure the accuracy of codes in a mailer’s address file.

For UK postal addressing, use an appropriately licensed Royal Mail Postcode Address File product or service when that source fits the purpose. Delivery representation, residence and legal location are not identical claims. A customer-authenticated confirmation may be needed for current residence or delivery instruction.

Email addresses

Syntax and domain checks are validation. Mail-exchange records indicate whether a domain is configured for email. A provider may attempt a server-level or historical deliverability check, but servers can accept all recipients, defer decisions, block probes or change behavior. A successful test does not guarantee future delivery or permission to contact the person. Treat deliverability status, verification date, source and marketing permission separately.

Phone numbers

Parsing a number into a plausible country format is validation. Carrier or line-status data may corroborate whether a line is assigned or reachable, subject to the provider, country and allowed use. It does not prove the owner’s identity, consent or continued control. An authenticated one-time-code flow can confirm control at that moment, but even that expires as numbers are reassigned.

Company details

Use the register for the jurisdiction. The UK Companies House service provides company data and filings, including registered address and status. Its own service warns that Companies House does not check the accuracy of all information filed, so record what the register shows rather than convert a filing into an unlimited factual guarantee. US entities are generally checked through the relevant state Secretary of State or equivalent registry, not one universal federal company register.

Professional and licence details

Use the issuing regulator, licensing board or official verification service for the profession and jurisdiction. Define whether you need identity, license class, current status, conditions, expiry or disciplinary action. A copied certificate is evidence supplied by the person; current register status is a different check.

Corroboration when no register exists

Some claims have no authoritative source. Two genuinely independent, recent sources agreeing can raise confidence without proving truth. Record “corroborated by sources A and B,” not “verified” without qualification. Rate confidence using prewritten rules such as high, medium, low and inconclusive, and keep the evidence dates.

Sources are not independent when one syndicates from the other. Ten websites repeating the same commercial database may represent one origin. Trace provenance where consequence is high.

Not every record deserves verification

Verification consumes provider fees, staff time, access and personal data. Apply it top-down according to the consequence of being wrong.

1. Money and authority: bank details, payee identity, invoice approval, tax status and authority to transact.
2. Legal, regulatory or contractual duty: identity, eligibility, sanctions, registration, license or records needed for a required process.
3. Safety and care: allergies, care instructions, contact identity or other facts that can affect a person’s wellbeing.
4. Physical fulfilment: deliverable address, recipient and service restriction.
5. Operational dependency: supplier status, product identifier, account relationship or system owner.
6. Marketing contact: current role, route, objection status and applicable permission or lawful basis.

Within each category, assess likelihood of change, source reliability, value, recency and whether a later control would catch the error. Verify until the expected cost and risk of checking exceed the consequence the business is prepared to accept. This is a policy decision, not a claim that lower-ranked records do not matter.

A wrong allergy record and a stale marketing job title are both inaccurate, but they do not justify the same method, frequency or spend. The highest-consequence fields may require independent confirmation and dual review. Low-consequence fields may be updated when the individual next interacts.

What it costs to act on a wrong record

A failed delivery creates outbound handling, carrier cost, return handling, customer contact, correction, repacking and redelivery. The commercial cost varies by parcel, route, labor and customer agreement, so calculate it from your actual workflow instead of inserting an industry average.

A mismatched claim or account identifier can trigger rejection, investigation, resubmission and delayed cash. A returned mailing consumes print and postage, but it also signals that the database should stop treating the address as current. A wrong license or company status can invalidate approval and expose the business to contractual or regulatory consequences.

Customer trust is harder to price but easy to recognize. Sending confidential information to an old address, contacting a deceased person or challenging a customer using a record they already corrected makes the company appear careless. Record complaints and repeat corrections so quality reporting does not reduce these events to postage or handling cost.

Build a cost-of-error worksheet: incident volume × direct handling cost, plus staff time, delay, customer remedy and any separately assessed legal or risk exposure. Segment by error type. Do not reuse one dramatic ratio for all records.

Build verification into capture

Check high-consequence information when it enters or changes. Use allowed-value validation first, then verification suited to the claim. Present ambiguous matches for correction. Where possible, let the person confirm their own contact information through an authenticated channel. Preserve an audit trail without storing unnecessary evidence.

Scheduled re-verification remains necessary for fields that decay. Set cadence by volatility and consequence: immediately before money moves, at renewal for a license, before a physical dispatch for an address, or when a dormant contact re-enters an outreach workflow. Event triggers are often more efficient than checking every record on a calendar.

Returned mail, bounced messages, customer corrections, supplier notices and register changes should update status or create a review task. A periodic mass cleanup keeps recurring when the intake and feedback loop remain unchanged. Repair the point of capture and the process that ignores new evidence.

Adding missing firmographic or contact fields is data enrichment for B2B sales, not proof that existing values are correct. Bulk mapping and transformation belong in data processing services. Spreadsheet formatting and structural repair belong in Excel cleanup.

Manual review, automated checks or both?

Automated checks are fast, repeatable and inexpensive per record at volume. They can validate patterns, query supported sources, compare exact identifiers and route changed statuses. They are rigid: an upstream source outage, transliteration difference, name collision or complex exception may return a false match or no useful result.

Human reviewers can compare context, follow alternative official paths, interpret conflicting evidence and explain an inconclusive result. They are slower, costlier and can make inconsistent judgments. Use written source priorities, decision rules, examples, named accounts, sampling and reviewer calibration.

Combine them. Automate deterministic checks and unchanged low-risk records. Send conflicts, non-matches, high-consequence changes and confidence below the approved threshold to people. Require second review for defined critical actions. Measure false accepts, false rejects, inconclusive outcomes and downstream corrections, not only “records processed.”

Record the result so it means something later

A verification result should store:

• the exact field or claim checked;
• the source or source category and version where relevant;
• the verification date and time;
• the responsible named person or system;
• the method, match key and evidence reference;
• the result: confirmed, contradicted, not found, inconclusive or unavailable;
• a confidence level and reason;
• the permitted purpose or workflow where needed;
• the next review or expiry trigger; and
• any correction, notification or suppression action.

A “verified” flag with no date, source or claim becomes misleading as soon as the underlying fact can change. Do not make verification status permanent by default. Historical status can remain accurate if it is clearly labeled “confirmed as of” rather than presented as current.

Accuracy, consent and compliance are separate

For personal data governed by UK GDPR, the ICO’s accuracy principle guidance says information must be accurate and, where necessary, kept up to date, with reasonable steps taken to erase or rectify inaccurate data according to its purpose. It also states that the more serious the consequence, the greater the verification effort that may be justified. This is general information, not legal advice.

Accurate data is not automatically lawful to collect or use. A verified email does not create consent, legitimate interest or permission under channel rules. Keep accuracy status separate from marketing consent, lawful basis, suppression, contract status and identity assurance. An objection or unsubscribe must survive data refresh even if the underlying address remains technically current.

Apply the law and sector rules for the actual market and purpose. Document source, transparency, data minimization, access, retention, processing roles and rights handling. Do not use verification as a reason to collect more personal data than the decision requires.

Run a 200-record verification audit this week

1. Select the highest-consequence record category with enough volume to matter.
2. Define the exact claims, accepted sources, results and verification date.
3. Draw a random sample of 200 records from a stated population and period. Preserve the selection method.
4. Validate structure first, then verify each claim against the named source. Separate source unavailable from record contradicted.
5. Use a second reviewer for critical and uncertain cases.
6. Calculate confirmed, contradicted, not found, inconclusive and unavailable rates with their denominators. Report by field and consequence.
7. Review causes: old capture, source change, missed customer correction, duplicate identity, integration overwrite or reviewer error.
8. Estimate operational cost from actual affected workflows and prioritize capture or refresh changes.

A 200-record sample is a practical starting point, not automatically statistically conclusive for every population or subgroup. Rare critical errors may require a larger or risk-targeted review. Do not let the headline error rate hide a severe issue in a small field category.

You can run the method internally without buying a service. If you need source design, controlled review and a documented result, request a verification audit on a 200-record sample.

Bring the highest-consequence record category, the current “valid” rule and the action the business takes on it. We will define the claim, source, result and audit before checking the full population. You can also book a free consultation or email support@ovelit.com.