In Korean on Brunch, in Japanese on note.
"We've started partnering with companies by paying $100K-$4M to license non-public data."
That landed in my LinkedIn inbox in September. The sender introduced himself as the CEO of Handshake. That one line was the whole offer.
Not a product. Not shares. The data that piles up inside a company while people work.
The half-finished idea someone typed to the team at 11pm. The customer's pricing a rep pasted into a channel. The thread where a deal nearly died.
Read it from the investor side of the table and two questions come first. If a founder in our portfolio got the same message tomorrow, who would decide? And would we even hear about it?
What Handshake says it buys
Handshake is best known as a hiring network for students and early-career talent. It now has an AI business, and its data partnership page is public. Here is what it says, in its own terms.
Payouts range from $100K to $4M, with a plus sign after the 4. It says it has paid out $250M to partners and works with more than 1,000 partner companies.
It is looking for companies with 20 or more full-time employees and at least 3 years in operation.
The sources it lists include Slack, Gmail, Google Drive, Outlook, Notion, Figma, NetSuite and QuickBooks, "+1K more."
It says the company determines exactly what is shared, and that all data is "scoped, de-identified, and confidential."
The estimate tool adds that its number is "an indicative estimate only and is not an offer to purchase your data."
Handshake's blog post announcing the program is more direct about the purpose.
"We purchase data directly from companies, data that reflects how real organizations actually operate. We use that data to build real-world environments where AI models can learn to execute tasks." The results, it says, "feed directly into the largest AI models in the world." For employers, it calls this "a chance to get paid for institutional knowledge that would otherwise go unused." It also offers $10K to anyone who introduces a company that becomes a partner.
None of this is hidden. It is a public program with public terms.
And it is not alone in putting a price on human work. Earlier this month Sequoia led a $60 million Series B in Mecka, a company that pays people to record everyday tasks so robots can learn from them. Different data, same direction.
What is actually for sale
The usual way to think about company data is defensive. Customer lists, contracts, source code. Things to lock up.
That is not what Handshake describes. Read its own words again: how organizations actually operate, how decisions get made, how teams use tools. That is a description of a Slack workspace and an inbox. Who asks whom. How a quote turns into a deal. Which problems get escalated and which ones die in a thread.
Here is my view. In a 3-year-old company with 20 people, those workflows are a big part of what founders mean when they say "our edge is how we work." The pitch deck calls it execution. The board calls it the team. In the data room it has no line at all, because until recently nobody was paying for it.
I am not saying Handshake is doing anything wrong by buying it. I am saying the seller should know what makes the data worth up to $4 million. It is the workflows.
Why "yes" comes easily
Put yourself in that CEO's chair.
The data is already there. The offer says you choose what to share. It will be de-identified. It is cash, possibly a lot of it, for something sitting unused on a server. For a company of 20 or 30 people, even the bottom of that range is real money.
I understand why that looks like a simple operating decision. No new shares. No change of control. Nothing that obviously needs a board vote.
That is why I think it can slip past a board. Nobody has to hide anything. It just doesn't look like the kind of thing boards get asked. A board with a broad IP licensing consent may already be covered. Many boards won't know until someone reads the clause.
What "de-identified" doesn't settle
De-identification answers one question: can a person be identified from this data? It doesn't answer several others that a board should care about.
Whose data is it? Picture a sales channel. A rep pastes in a customer's pricing from an email. The customer sent it under an NDA that limits what it can be used for. Strip every name out and it is still the customer's pricing.
Employees wrote the messages. Customers sent the emails. Removing names does not automatically make any of it the company's to license.
What does "de-identified" mean where the company operates? Take Korea. Under the Personal Information Protection Act, pseudonymized information is still personal information (Article 2, item 1(c)). The Act stops applying only to information that can no longer identify anyone even when combined with other data, judged by reasonable time, cost and technology (Article 58-2). "De-identified" is not a term the Act uses, so a founder has to ask which of the two it actually means. The Act does let pseudonymized data be used without consent for statistics, scientific research and public-interest archiving (Article 28-2). Whether licensing it to train a commercial model fits there is for a lawyer to answer before anyone signs. The US, Europe and Japan each draw their own lines.
Can it be undone? Once data has gone into training, I would not assume a deletion clause gets it back out. That is a reason to decide slowly, not a reason to decide never.
If you sit on a board, or invest in people who do
This section is for the people who sit on boards, or back the people who do: directors, CVC teams, GPs, and their LPs.
In the investment agreements I have seen, the consent list names things like issuing shares, selling major assets, or licensing intellectual property. Read yours with this offer in mind. These are the three questions I would put on the next board agenda.
Has the company been offered this, and who would decide? Not "did you sell." Start with whether the offer came in. If the answer is "I'd just decide," that is the answer you need to hear now, not after the wire.
Does our consent list cover it? Does licensing internal data count as disposing of a major asset, or as licensing IP? If nobody in the room can say yes or no, that is the gap. Write it in.
Whose data would be in it? Customer emails, partner documents under NDA, employee messages. Ask who checked, and what the contract says about scope and deletion.
If you are an LP, ask your GP the same thing at portfolio level: have any of our companies been approached, and is it on your board checklist? Then ask one more. A fund manager with 20 people and 3 years behind it fits Handshake's stated criteria too. Its inbox holds LP details and deal flow. Who at the firm would decide about that?
On October 12 I wrote that the word "AI" on a deck is a claim, not an answer. Same habit here. "Do we protect our data?" gets a yes from everyone. "Who signs a data license, and has anyone offered us one?" gets a real answer.
Selling can be the right call
I want to be fair to the other side.
Some companies will look at this and decide to sell. A business with a narrow, well-scoped dataset, clean consent, and a real need for cash may be making a perfectly rational trade. Handshake says the company chooses what goes in. A careful seller can use that.
My view is not "don't sell." It is that this is not one person's call. The price on that page is a starting point. The board's job is to ask what the same data is worth to the company that created it, and whose permission was never asked.
One question
If this offer reached one of your companies tomorrow, who would make the call: the CEO alone, or the board?
In this newsletter I write down the things I got wrong, or was on the wrong end of, in how capital picks people across borders. If you sit on either side of that table, it's written for you. Subscribe and the next issue comes straight to your inbox.