Two in three launches, partnerships and executive changes we read are not on Crunchbase’s cards
In June, John Jumper, the Nobel laureate who led AlphaFold, left Google DeepMind for Anthropic. Ambarella and ZEDEDA announced a partnership for AI on edge devices in September. Ataccama named Systemation its distributor for the Benelux the same month (BigDATAwire, 4 Sep). None of the three is on the companies’ Crunchbase event cards, checked on 31 October 2026.
That is not a gap in a corner. Of the real launches, partnerships and executive changes I tested, 6 to 7 in 10 were not on the company’s Crunchbase event cards (33 of 50). So is Crunchbase enough to follow a company? For its deals, my other test says yes. For the launches, partnerships and people moves in between, not on its own, and a news search on the name helps less than you would think: almost half of the real events I read did not come up in one.
What I wanted to know
Crunchbase lists more than funding these days. Through its new AI connector, a company profile has event cards for partnership announcements, product launches, key employee changes and layoffs, each with the date Crunchbase logged it. Its own home page now says “Before It’s News, It’s In Crunchbase.”
I build ins7ghts, which reads what practitioners publish about data and AI every day and keeps each company’s launches, partnerships, executive changes and contracts with the article they came from. If Crunchbase has event cards now, what does reading the field still add? I wanted to know before somebody asked me on a call.
I wrote the test down before looking, with a bar: at least 3 in 10 of the real events had to be missing from the cards, or the claim was off the table.
Step one: are the events even real?
Before comparing anything with Crunchbase, I had to know whether the events we read are worth comparing. We, here and below, means me and my AI team. Machines reading articles get things wrong. Most often they take something old, mentioned in passing, for news.
So I drew 150 events from July to September 2026 articles, 25 of each type, and had three AI models check each one against its article: is this a real, recent event, and is it about the company we name? A stronger model settled the splits.
About 7 in 10 were (109 of 150). By type:
| Event type | Real, recent, the right company |
|---|---|
| Contracts | 84 in 100 |
| Launches, partnerships, executive changes | about 3 in 4 each |
| Restructurings and shutdowns | about 6 in 10 |
The misses were instructive. A 2022 contract named in a list of older contracts. “The government” appointing a commissioner, which is not a company. A pricing page read as a launch. Restructurings and shutdowns came out weakest, because they are mostly named in commentary written later, so I don’t lean on those two.
Step two: what do the event cards hold?
Then the real comparison. For the real events of a type Crunchbase has a card for, and a company Crunchbase has a profile for, 50 in total, I looked at the company’s cards in a window around the article.
33 of the 50 were not there. 6 to 7 in 10, well over the bar.
33 of 50 real events were not on Crunchbase’s event cards
Still missing on the day I write this, after a fresh check: Jumper’s move to Anthropic, Ambarella and ZEDEDA, Ataccama and Systemation, and Snowflake’s Grok 4.6 preview on Cortex AI. Held, and held well: Visa’s 2,600 job cuts, and Island’s Enterprise Vibe Publishing.
One honest footnote. Adobe naming Anil Chakravarthy CEO looked like a miss in the test. It is on Crunchbase, dated 1 December, the day he starts, not the day it was announced in July. A card dated at the start date is a choice, not an error. But if you scan the cards by date, it sits in December, not in July, when the news was.
Is Crunchbase late, then?
No, and I checked because I half expected it. Where Crunchbase has the event, it logged it the same day as our article, in the median. In 8 of the 17 it was there before our article. Layoffs it holds well: 3 of 3 in the sample.
The gap is in what makes it onto a card at all. Reminds me of the official charts when I was DJing: what was on them, was on them the week it sold. Half of what filled the floor on a Saturday never made it onto them.
What about a news search on the name?
The obvious alternative is a news alert on a company’s name. I tested that too, on the 109 real events: a three-week Google News search on the company’s name, first 20 results.
50 of the 109 did not show up. Almost half.
Not because big names drown the results. The events sit in articles about something else: a market overview, a customer story, a list of who integrates with whom. The company is in the article, just not in the headline, and a name search ranks the headlines. That is where the Ataccama distributor deal lives, and the Ambarella partnership: in trade press where they are one paragraph among many.
How I follow a company now
Three sources, each for what it does well:
| What you want to know | Where it is |
|---|---|
| Who raised, who bought whom, who cut jobs | Crunchbase: complete and same-day |
| What the company announced itself | its newsroom |
| What the company did that others wrote about, where it was not the headline | the field, read every day, each event with its article |
Watching a list of companies, competitors or a portfolio? That third row is what Portfolio Watch does: your list, read every month, every event with the article that reported it.
The next test shows up first in Ins7ghts Daily, my daily newsletter on data and AI.
If what is on your desk is a pitch deck rather than a watch list: with the Deck Screen, every checkable claim in a pitch deck gets one verdict, Confirmed, Contradicted or Unverifiable from the public record, with the source attached.
The small print
Articles of July to September 2026. 150 events drawn for the first test, 109 real; 50 of those had a company profile and a card type on Crunchbase for the second. AI models did the reading and the first grading and checked each other’s work; the test was written down before the result. Crunchbase read through my Pro account on 31 October 2026, cited here as counts and a few names; the named examples re-checked the same day. The news search covered past dates on 31 October; a live alert may deliver more. One quarter of articles, about data and AI. More of these tests are on the 7wData research pages.
