A spreadsheet can change a business identifier before anyone starts analyzing it. The risky moment is often earlier: someone pastes values into a sheet, opens a CSV file, or…
Read articleMerged cells are not always a mistake. A title centered over a report can make a sheet easier to read. The trouble starts when a visual layout device is asked to carry business…
A comma can be either data or structure. In `Northwest Supplies, Inc.`, the comma belongs to the company name. In a row such as `company,city`, the comma separates one field from…
A spreadsheet arrives with four columns: `date`, `status`, `amount`, and `id`. Nothing appears to be missing. Yet each header leaves a decision to the recipient. Is `date` the…
A spreadsheet shows the same order number on three rows. The instinct is understandable: select the column, click “Remove Duplicates,” and make the file look cleaner.
`2026-10-04T09:00` looks precise. But precision in the characters does not guarantee precision in the meaning.
A rounded number is a communication choice, not proof that the original number was the same. Consider two conversion rates: 18.46% and 18.54%. Displayed to one decimal place,…
A procurement sheet contains `Northstar Labs` in one column and what appears to be `Northstar Labs` in another. An exact lookup returns no result.
A cell that looks empty can still make a strong claim. “0 open safety incidents” says the measure applied and the count was zero. “Not applicable” says the measure did not apply…
A category field can look harmless: industry, customer segment, territory, product type, account status. The values are short, familiar, and easy to count.
A dataset is not clean or dirty in the abstract. It is fit, questionable, or unsafe in relation to what someone intends to do with it.
A $250,000 order appears in a weekly sales file where most orders are under $5,000. One warehouse’s delivery time doubles for a month. A department’s travel expense lands far…
A spreadsheet can look self-explanatory right up to the moment two people use it to answer the same question and get different results.
“Normalization” means two different things in data work, and confusing them can lead to quiet mistakes.
Adding a column from a second public source often feels like a straightforward upgrade. A directory gains corporate status. A list of locations gains industry information. A…
Consider a supplier-review field: **Certificate on file?** Four records appear nearly identical at a glance:
A polished record earns confidence quickly. The company name is consistently capitalized. The address parses cleanly into street, city, region, and postal code. The phone number…
Before calculating on a field, ask one question: **What is the unit?** Revenue has dollars. Delivery time has days. Quantity has items. A percentage has a defined numerator and…
A spreadsheet with columns called “Price” and “USD price” can look ready for analysis. Yet those headings often conceal the information that determines whether a comparison is…
`125` is a clean number. It has no stray symbols, no malformed decimal, no missing cell, and no obvious spreadsheet error.
A phone field often looks like the easiest column in a business dataset. It contains digits, perhaps punctuation, and sometimes an extension. That appearance invites a familiar…
A schema tells a receiving team which fields exist and how they are formatted. A data contract should answer the harder questions: what the dataset represents, what it may safely…
A percentage can look like a clean verdict: conversion rose, retention improved, on-time delivery reached a new high. But a percentage is not self-explanatory. It is a…
Automation is often described as a way to remove manual work. That is only partly true. The better goal is to remove manual handling from decisions that are genuinely routine,…
A pattern found in historical data may describe the past accurately and still be a poor guide for the next decision. A holdout period is a practical way to test that distinction.
A benchmark can look precise while answering a different question than the one a decision-maker has in mind. That often happens with public business data. A table may omit…
Consider these two fictional rows in a product file: - **Harbor Lime Sparkling Water — 12 × 12 fl oz** - **Harbor Sparkling Water, Lime — 144 fl oz**
Two records land in a review queue. - **Harbor North Consulting LLC** — 500 Market Street, Suite 410 - **Harbor North Holdings Inc.** — 500 Market Street
A final dataset shows what the data looks like now. It rarely explains how it got that way. A standardized region column does not reveal whether “N.E.,” “Northeast,” and “NE”…
A supplier sends a CSV. The columns line up, the row count looks right, and every comma is where it belongs. Then someone opens it and sees `José` instead of `José`, a currency…
A team asks a reasonable question: Which customer segment is most valuable? The tempting response is a ranked list. Segment A first, Segment B second, Segment C third. It looks…
A sales leader asks, “Which companies should we target?” It sounds like a research assignment. It is not—at least not yet.
A product file can look tidy and still describe several different things in the same row. Take “Trail Runner Shoe.” That may be a style family. “Trail Runner Shoe, blue, size 9”…
A data file can look familiar, load without complaint, and still stop meaning what your process assumes it means.
A public-web dataset is not a permanent statement about the world. It is a record of what a source made publicly available when it was observed.
Research can be careful, relevant, and well reasoned—and still fail at the handoff. That failure is usually quiet. Someone receives a spreadsheet, slide deck, or folder of links.…
A team opens a large file, reviews 25 records, and immediately finds oddities: inconsistent dates, unclear categories, duplicate-looking entities, or values that do not match the…
“Refresh every hour” sounds like a requirement. Usually, it is an unfinished question. A refresh schedule is not valuable because it makes a dashboard look current. It is…
A dataset can earn an impressive completeness score and still fail the decision it was meant to support.
A business metric can be calculated correctly and still be a poor basis for action. Consider a retention rate just above the threshold for a save offer, or a…
A spreadsheet opens a file, and a product reference such as `00048391` becomes `48391`. Or a long record ID appears as `1.23457E+19`. The instinct is understandable: put the…
One customer account is ten times larger than the rest. A monthly expense is negative. One branch appears to have impossible productivity.
Changing “Street” to “St” can make an address file look cleaner. That can be useful. It is not the same as knowing what the address means, whether it can receive mail, or whether…
“Average” sounds like a plain fact. In a business report, it is often a choice disguised as a fact. A single number can be useful. It can also answer a different question from…
A label such as “fiscal 2025 revenue” looks precise until two companies use it for different stretches of time.
A team with 240,000 supplier records cannot inspect every row before making a decision. That does not mean it should check 50 records, find nothing alarming, and call the file…
A number can look precise and still answer the wrong question. Consider a monthly employment figure, a company’s quarterly revenue, or an operations dashboard labeled “April…
Six months after a vendor shortlist, market map, or partnership screen is approved, someone asks a reasonable question: Why did this company make the cut while another did not?
A repeated name, email address, street address, or identifier is a reason to look closer. It is not, by itself, a reason to delete a row.
A join is often treated as a mechanical step: match two tables, add columns, build a chart. But every joined row makes a claim that deserves its own scrutiny.
A category is often treated as a permanent fact about a record: this company is in that industry, this customer belongs to that segment, this job has that occupation code.
An analyst finds the same customer, vendor, account, filing, or transaction ID on two rows. The column is called `ID`, perhaps even `Unique ID`. The natural impulse is to fix it…
A column full of digits can look like a numbers problem. Often, it is an identity problem instead. Consider the difference between `12.5%` and customer ID `00125`. The percentage…
A URL is evidence of a web address. It is not, on its own, evidence of a separate business. That distinction matters whenever a list of websites becomes a list of companies,…
A customer enters a German delivery address into a form built around U.S. expectations. The form requires a state, expects a numeric ZIP code, and places a street number before…
A mixed-case column can look untidy. That often prompts a quick instruction: make everything uppercase, lowercase, or Proper Case.
A familiar analysis begins with a click: sort a sales table from highest to lowest, rank locations by service time, or put campaigns in order by conversion rate. The first few…
A benchmark needs a job description before it needs a number. Many performance reviews start with a familiar move: pull an industry average, name a few recognizable competitors,…
Many data problems begin before anyone opens a cleanup tool. They begin when two people use the same field differently.
An empty field can describe several different situations, and those situations do not mean the same thing.
Every dataset has an invisible timestamp, even when there is no date column on the screen. Businesses move. Websites change. Products disappear. Prices shift. People change…
Three different kinds of updates stood out this week because they show how much useful context sits outside the spreadsheet itself.
A surprising number of data problems begin with a question nobody asked at the start: what is one row?
Here are two completely fictional records: At first glance, these look like the same company written two different ways.
A bad value does not stay politely inside its own cell. Once data starts moving through reports, joins, formulas, filters, exports, and other systems, a small inconsistency can…
A spreadsheet does not need to be wildly broken to produce a misleading result. Sometimes one small mistake is enough.
“Verify this” sounds clear until the project has to decide what verification actually means. A phone number appearing on three directory pages may look convincing. But if all…
Not all public sources deserve the same amount of trust, and “public” does not automatically mean “authoritative.”
Two credible sources can disagree without either one being careless. That is uncomfortable because research feels easier when there is one clean answer. Real data is often less…
One public dataset worth knowing is the Bureau of Labor Statistics Quarterly Census of Employment and Wages, usually shortened to QCEW.
A dataset becomes much easier to trust when a record can answer one simple question: where did this come from?
A company list can contain a lot of rows and still be a poor research product. Row count is easy to notice. Relevance is harder.
A date column can look clean and still be difficult to trust. Some problems are obvious. One row says “March 4” while another says “03/04.” Other problems are quieter: two…
A company name feels like one of the simplest fields in a dataset. Then two records show up that are obviously related but refuse to match cleanly.
A spreadsheet sees a row of digits and often assumes it is looking at a number.
A blank cell looks simple. It is empty, so the instinct is often to replace it, delete the row, or treat it as missing data.
Duplicates are rarely as simple as two identical rows. A quick review can miss them when names vary, fields overlap, or the important clue is somewhere else.
Why the goal of data cleaning is dependable decisions, not merely tidy formatting.