The industry column was wrong on two-thirds of the list. I checked every one.

A sales list landed on my desk this week — just under five hundred companies, already tagged by industry, ready to call. Someone had done the sensible thing and put a label on every row. Cybersecurity here, recruiting there, a bank, a manufacturer. All I had to do was write the pitch.

So before I did, I pulled a dozen names I happened to already know and checked the labels against reality. Several were wrong. Not close-but-off wrong — different-planet wrong. A company tagged as a drinks brand turned out to be a cybersecurity firm. A "recruiting" business was actually industrial engineering software. That's when I stopped trusting the column and decided to check all five hundred.

The label is the easy lie

Here's the uncomfortable part. A pitch is only ever as good as the tag it's built on. If I greet a cybersecurity founder as though he sells fizzy drinks, the email is dead on arrival and I've burned the one impression I get. A wrong label doesn't feel like a risk when you're staring at a tidy spreadsheet. It feels like a head start. It isn't.

This is the same trap I keep coming back to — a number or a category that looks authoritative right up until you audit it. The neat field lulls you. The work is in refusing to believe it.

Five hundred checks used to be a no

Ten years ago, verifying what five hundred companies actually do would have been a week of someone's life, so nobody did it. You pitched the label, accepted the miss rate, and moved on. That's the bit that has genuinely changed.

I ran the domains through a company-data lookup to pull the firmographics, then sent a research agent after the ones with no website and the ones where the first answer smelled wrong. Bulk classification by machine, targeted second looks by hand. The whole pass cost a rounding error in credits and an afternoon — not a week.

But — and this is the entire point — the machine got the confident-looking ones wrong too. Industry codes lump a boutique consultancy in with the giant it once served. A name with "cyber" in it turns out to be an industry association, not a vendor. The tool gets you eighty percent of the way at a hundred times the speed; the last stretch is judgement about which answers to distrust. That judgement is the actual leverage, not the tool.

What the recount showed

Of the rows that arrived with a confident industry tag, only about a third survived. Two-thirds were reclassified. Whole categories the original tags never even had — training providers, public-sector bodies, data and AI shops — surfaced once I looked properly.

The spreadsheet wasn't lying on purpose. It was just never checked. Most spreadsheets aren't.

None of this is exotic. Bad records are the quiet tax on every business — Thomas Redman's much-cited estimate is that poor data quality costs the US economy around three trillion dollars a year, and his follow-up work found that only three percent of company data meets basic quality standards. B2B contact data rots especially fast because people change jobs; a meaningful chunk of any list is stale within a year. The label being wrong isn't the exception. It's the base rate.

I've made this mistake the expensive way before — trusting a tidy summary that turned out to be missing the one thing that mattered. The reflex I've built since is boring and it works: when a field decides how I'll spend my time, I verify the field before I spend the time.

The list is in good shape now. The pitches are built on what these companies actually do, not on what a column claimed they do. And the part I'd underline for anyone leaning on AI to move faster: speed makes the checking cheap. It doesn't make it optional.

Trust the reps, not the label.

Sources & further reading

External
Thomas C. Redman, "Bad Data Costs the U.S. $3 Trillion Per Year," Harvard Business Review
"Only 3% of Companies' Data Meets Basic Quality Standards," Harvard Business Review

Related posts
The Death of Generic AI: Why Deep Domain Expertise Is the Only Real Leverage Left
The AI handed me a number that was wrong in the most convincing way
Claude Cowork said nobody replied. The reply was message 28.
Why your direct traffic is a lie (and how to audit it)