Transcript
Before any NDA, before any mandate, a company holding data needs a list of what it has. That's the inventory. It's the first working document in the process. It stays internal. It's built before any buyer sees a sample, and it can end in a decision not to license anything.
Chapter one. The data inventory.
Per source, the inventory answers three questions. What is it? Where does it live? How much is there? Then it adds who administers it, which contracts cover it, and whether people appear in the records. Those extras decide whether the source can go any further.
Don't confuse it with its cousins. A catalog helps staff search and understand data. A data map shows how data flows between systems, for compliance. A dictionary defines the fields in one dataset. The inventory here serves a licensing decision.
Here's the row. One row per source, ten fields. First, the source itself, one system or archive per row. Then the data type. Then the administrator and the plan, because the plan decides what exports. Then the date range. And the volume, with its unit.
The other five. People in the records: staff, customers, patients or none. Contracts over the records, such as a customer NDA that bars reuse. The export route and its deadline. Exclusions, meaning sub-sources kept out. And the status of the row.
The status field takes one of three values: include, exclude or hold. A hold row stays out of any offer until its open question is answered. So payroll lines, say, wait for review, and nothing leaks into a buyer conversation by accident.
A volume means nothing until it's in the unit a buyer counts. Chat is counted in messages and accounts. Code in repositories, commits and lines. Recordings in hours. Structured data in records or gigabytes. Always write the unit beside the number.
There's a public example. A bankrupt US airline filed its data schedule in court. It counts a hundred million emails and five hundred million chat items. One line per source, a unit, a count. And the sale isn't approved as of November ninth, 2026.
Same schedule, one more lesson. Ninety-seven and a half million passenger profiles are marked Not Included. The count that matters is what's left after exclusions, so each row carries two numbers: gross and licensable, with the gap explained.
Retention can shorten the range. One team chat tool deletes history older than one year in its free workspaces. So the history that exists can be shorter than the company's history. Find that out early, while the records still exist.
How do you build it? From admin consoles, billing records and the people who run each system. List systems from billing, retired ones included. Name each administrator. Record dates and counts, marking estimates. No record needs to leave its system.
And you'll be asked for it. Most programs want an inventory in all but name within the first call, before any sample or contract. Systems, date ranges, volumes, exclusions: each request maps to a field of your row.
Chapter two. The NDA.
An NDA before a sample or data room binds the AI buyer in two ways. It must keep what it sees confidential. And it may use it only to decide whether to license. It's usually the first signed document in processes that publish their steps.
Know its limits. The NDA grants no right to train on anything shared. But it also can't reverse training once data has reached a model. A breach after training isn't cured by deleting the sample. So keep real records out of the room until it's signed.
Five clauses decide whether it protects a sample. Purpose. No training. Return or destruction. Residuals. And term. Walk through them in that order, and you'll catch the weak draft before it's signed.
Purpose first. Name it in one sentence: use only to decide whether to license. Cover affiliates, contractors and advisers. Then say plainly that nothing shared may train, fine-tune or benchmark a model, or generate synthetic data. A purpose clause alone may be read narrowly.
Return or destruction: require it when talks end, with written confirmation, and answer backups expressly. Return and deletion are different promises. Residuals let the recipient use what its people retain in memory. With AI buyers, what's retained can be a model. Ask to strike it.
Then the term. It should outlast the period in which the data has value. One data program's published master terms keep confidentiality for three years after termination, and indefinitely for trade secrets. No published norm for NDA length in data deals exists.
Who signs decides who's bound. A mutual form protects both sides' information, not only the data. It also binds the seller to the counterparty's pricing and process, which matters when you compare several counterparties. Read it before you accept it.
Two traps. First, an intermediary may hold the NDA with the lab itself, leaving the seller outside that agreement. Ask who each NDA is with, and whether its terms pass through. Second, some platforms show the NDA as a pre-checked box at signup. Read each document before upload.
What do you hold back until it's signed? Records. Before the NDA, share descriptions: which systems, years and volumes exist, without content. The description layer is the data manifest. Inspecting records comes later, in a data room.
Chapter three. The mandate.
A mandate is the contract that authorizes an intermediary to approach buyers for the company. It sets the fee, the exclusivity, the notice and the tail period. And for the seller, it decides whether they can still go direct.
Here's one published example, and it's a useful baseline. The mandate is non-exclusive. Thirty days' notice. And a twelve-month tail, limited to listed buyers. That means after it ends, only buyers on an agreed list stay covered.
Exclusivity has a price. One data buyer's listing for medical practices names data already licensed exclusively through a broker as not a fit. So an exclusive mandate can close a direct door. Always say so to the seller, in plain words, before they sign.
Control matters too. One intermediary states that the holder can exclude named buyers, competitors, customers or uses. And some never receive or hold the data at all. Know which kind you are, and write it into the mandate.
Then the fee. A seller-paid success fee reduces proceeds, and published rates run from six percent to twenty percent, as of November ninth. Disclose yours before anything is signed. And if anyone else gets a referral reward, say that too.
Seven written answers separate one intermediary from another. One: who pays you, and how much, by buyer? Two: is the mandate exclusive, and what notice and tail apply? Three: who signs the license? Four: do you receive or hold the data?
Five: which buyers do you approach, and may the company exclude any? Six: have you closed a data license, and for whom, if you can say? Seven: is any fee or reward paid to anyone else? A good broker can answer all seven.
Chapter four. Introducing a client.
Suppose you know a company holding data. The introducer form recognizes four groups: advisers, such as accountants, lawyers and M&A advisers; insolvency office holders; investors winding down a portfolio company; and others. You must be free to say the company exists.
Before you name anyone, check four points. You may share the name. For an insolvent company, the person with authority to sell knows of it. The notes describe only the kind and age of the data. And they hold no personal data of customers or staff.
Two rules are worth knowing. In the United States, a bankruptcy code section bars sharing certain compensation, so any fee arrangement around an estate is checked first. And under GDPR, naming a person triggers an information duty. Talk to a lawyer about both.
And promise nothing. The introduction form says it promises no buyer, price or outcome. The company, or the office holder acting for it, decides whether to license any data. Deciding not to is a valid outcome.
So here's the whole track in one line. Understand the market, find sellers, qualify the data, disclose every fee, run the process, respect privacy, and put it all on paper with an inventory, an NDA and a mandate. That's the end of the broker track. Everything in it, with sources, is at datapacts dot com.
Read and watch the full lesson free. Enter your email to unlock every lesson of the academy.