DataPacts

Broker track · Lesson 5 of 7 · 9:16

Running a sell-side process

What you'll learn

  • Running several buyers at once
  • Clauses that block a second deal
  • How to read an offer
  • Levers to trade in a negotiation
  • How long a deal takes

Free lesson

Enter your email to watch this lesson and the rest of the course.

Transcript

Running several buyers against each other is the move people expect from a broker. So does it work? The reference page says there is no public evidence that parallel processes raise prices for company data. It can still be right for a seller. You just never promise it.

Chapter one. What parallel really means.

Parallel ends in one of two ways. In a competitive process, several bidders go for one grant, and the losing bidders receive nothing. In non-exclusive stacking, several buyers each get a license to the same data, and each pays separately.

The only fully documented competitive process for company data comes from a bankruptcy court. A bankrupt US airline's data drew a winning bid of ten million dollars and an alternate bid of seven and a half million. The sale still hadn't been approved as of November ninth, twenty twenty-six.

A later bid can displace an auction winner. In one bankruptcy sale, one buyer's auction result of two hundred fifty-six million dollars was superseded by a research institute's bid of three hundred five million. That was an asset sale, not an AI license, but the lesson carries.

Outside a court, no operating company has published how it ran competing bids. So this is practical guidance, and it isn't a market norm. Give every bidder the same data description. Sign a separate NDA with each. Set a sample policy, with deletion if there's no deal. Then ask for one written offer per bidder, by a common date.

Exits take time too. One program states that withdrawing after presentation needs thirty days' notice, and its processes may continue for ninety days. Completed licenses survive. So warn a seller early: leaving a program isn't instant.

Some programs take a non-exclusive license and place the same data with several buyers themselves, so the stacking happens inside one contract. One states the seller approves each buyer and can block competitors permanently. Another is paid on delivery and on each relicense. These are licensees that sublicense. They aren't brokers.

Chapter two. Windows and warranties.

Six kinds of published clause decide whether a second buyer can be signed now, later or never. An exclusive grant by default. Access that survives termination. A no-prior-grant warranty. Non-circumvention. A sole-seller window. And withdrawal notice with run-off.

A window is a fixed period when one party alone may sell the data. One program states it's the only seller for the first ninety days, then the company can renew or opt out. A clinical data side letter ran eighteen months. Until a window ends, parallel selling stops.

Now the order of signing. A no-prior-grant warranty makes you promise each new buyer that you haven't already granted the same rights to someone else. Sign a similar license elsewhere first, and you can't give that warranty truthfully. So the order you sign in decides which promises stay true.

Keep a register of every grant, so you can answer any no-prior-grant question from one record instead of memory. List the licensee, the data covered, any window end date, the field of use and term, the warranty wording, and the payment trigger.

Be straight with sellers about the downside. Parallel can break a no-prior-grant warranty. An early window or non-circumvention clause can block later deals. More parties hold samples. And you'll need more NDAs and more administration.

Chapter three. Reading an offer.

An offer is a buyer's written proposal at a stated price, and it's worth what its terms let the company keep. Four things decide that. Price. Payment trigger. Review window. And the grant of rights.

Watch the difference between an estimate and an offer. A calculator range or an illustration on a buyer's page is marketing. Only a written offer tied to an order form states a price the buyer proposes to pay. One estimator even calls itself indicative.

Offers price the data four ways. A fixed sum. A fee per delivered record. An upfront fee plus a revenue share. Or a revenue share alone, with nothing upfront. No program we read publishes its formula, and the share percentages stay unpublished too.

When does the money actually arrive? Buyers publish six different triggers. At signing. On delivery. On acceptance. After invoicing. On approval plus net terms. Or on each resale.

Then there's the review window. One buyer's master terms give thirty business days after delivery to reject, and silence counts as acceptance. Its own program page says thirty days. The two aren't reconciled, so ask which one applies.

And the headline price isn't what you keep. Extraction, de-identification or an intermediary's fee can come out of it. Some programs say they cover extraction themselves. One broker states that the buyer pays it when a license closes. Ask who pays what, in writing.

Eight questions turn an offer into terms you can compare. Which unit and quantity is the price based on? Who pays for extraction? What triggers payment? How long is the review, and what does silence mean? What survives acceptance? And what happens to the trained models when the grant ends?

Be careful with an unsolicited offer. One that names a large sum before anyone has seen the data is a reason to check who is offering before you reply. A competing intermediary's chief executive reported ads promising founders one, three and five million dollars.

Chapter four. Negotiating.

Before the first call, prepare four things. A walk-away point. The levers, ranked by what they're worth to the company. The terms it won't give. And a fallback if no deal closes, including keeping the data.

Seven levers move a deal, and each can be traded against price. First, price structure. Second, scope and field of use. Third, exclusivity. Fourth, retraining and models already trained.

Then deletion, audit, and the seller's protections: a liability cap, indemnity and exit. And walking away is a lever too.

Exclusivity is something buyers say they pay for. Yet no published figure says how much more, for company data. One program says only that exclusive licenses pay more up front.

Trade levers in pairs. A concession is worth asking for only when the other side values it more than you do. Exclusivity against price and term. A narrower field of use against a perpetual grant. An upfront fee against a revenue share. A shorter deletion deadline against model retention.

One more thing for brokers. Some intermediaries negotiate for the company while being paid by the AI side. One states it's paid by AI labs. Another states the buyer pays a fee when a license closes. Say so before the seller relies on you.

Chapter five. How long it takes.

How long does a deal take? One program says first call to first wire in under twenty-one days. Another says about three months from first conversation. No law, filing or seller report sets a standard.

Be careful comparing figures, because they measure different spans. One counts from first call to first wire. Another counts from the signed agreement to the wire. One counts to the first accepted contribution, which isn't payment. A short figure and a long one can describe the same deal.

A deal moves through ten steps, and for two of them nobody publishes any duration: the rights and consent check, and internal approval. Plan for those early. The published claims put sample to signed agreement at one to six weeks.

A court-supervised sale is slower. In the bankrupt airline's case, four hundred forty-five days passed from the petition to the scheduled sale hearing, with no approval.

So, to recap. Parallel is one tool, and nobody can promise it lifts the price. Watch the clauses that order your deals. Reduce every offer to the same points. Trade levers in pairs, and disclose who pays you.

Next up is lesson B6, Privacy, consent and ethics.

Read and watch the full lesson free. Enter your email to unlock every lesson of the academy.

Sources on DataPacts