Research
Notes on company data and AI
Practical guides for founders, operators and AI labs: what company data is worth, how it's anonymized, how licences work, and how we source it.
Latest · AI labs
A buyer's checklist for licensed company data
Before licensing company data for training or evaluation, buyers should check provenance, anonymization, licence terms, data quality and fit. A practical checklist for AI teams.
November 5, 2026 · 4 min readRead →
Why connected data beats big data
A million chat messages are worth less than a few thousand customer issues you can follow from first email to shipped fix. Here is why buyers pay for connection, not volume.
Nov 3, 2026
Our anonymization pipeline, step by step
From read-only access to a deletion certificate, here is the order in which company data is scoped, cleaned, tokenized, reviewed and delivered, and what the seller sees along the way.
Oct 29, 2026
Inside a sourcing campaign: from brief to signed companies
A walk through a sourcing campaign for company data, from the brief an AI lab sends to the companies that sign, and how the work moves between targeting, outreach, evaluation and delivery.
Oct 27, 2026
A data licensing FAQ for general counsel
Plain-English answers to the questions in-house counsel most often raise about data licences, from scope and anonymization to confidentiality clauses, deletion and signing authority.
Oct 22, 2026
What happens to a company's data when it shuts down
When a company closes, its operational records usually disappear on vendor deletion schedules. What gets lost, what must be kept, and how the history can instead return money to the company.
Oct 20, 2026
The economics of company data for AI
Company data for AI has unusual economics: low cost to the seller, real value to buyers, and prices set by connection depth rather than volume. Here is how the pieces fit together.
Oct 15, 2026
How AI labs evaluate a new data source
Before a lab commits to a new source of company data, it moves through a series of checks, from rights and provenance to structure, anonymization and fit. Here is how that usually unfolds.
Oct 13, 2026
Common reasons a dataset is worth less than expected
When an evaluation comes in lower than a founder hoped, the reasons are usually predictable. The common ones, from short history to thin links between tools, and what can still be done about them.
Oct 8, 2026
Consistent pseudonyms across years of history
Keeping a pseudonym stable inside one tool is easy. Keeping it stable across years of renamed accounts, changed emails, departures and tool migrations is the harder problem.
Oct 6, 2026
Building a target list from technographic data
A sourcing brief becomes a call list in stages: firmographic filters, technographic evidence, history checks, tiering and signer mapping. Here is how each layer narrows the field.
Oct 1, 2026
Representations and warranties in data licences, explained simply
Representations and warranties are the promises each side makes in a data licence. Here is what they mean in plain English, which ones usually appear, and how scope and exclusions keep them honest.
Sep 29, 2026
A founder's guide to telling employees about a data licence
How to tell current and former employees that your company has licensed its anonymized data: what to include, which questions to expect and how to handle pushback well.
Sep 24, 2026
What makes a dataset agent-ready
A large dataset is not automatically useful for training agents. The properties that make company data agent-ready, from connected tools and history to outcomes, privacy and clear rights.
Sep 22, 2026
Enterprise AI teams and private evaluation sets
Enterprise AI teams need to know whether an agent can handle their kind of work. Private evaluation sets built from licensed operational data answer that without exposing their own records.
Sep 17, 2026
Finance workflows as data: month-end close and audits
The month-end close and the annual audit are repeated, multi-step workflows with clear right answers. For finance leaders, that record of work is a licensable asset in its own right.
Sep 15, 2026
Reviewing an anonymized sample before delivery: what to look for
Before any data is delivered, you review an anonymized sample. Where identifiers tend to hide, what should be missing entirely, and how to read the sample like someone who knew the company.
Sep 10, 2026
Why outbound beats waiting for data to come to you
Inbound supply reflects whoever happened to find you. Outbound lets an AI lab decide which companies, sectors and tool stacks end up in its data, and reach them while their records still exist.
Sep 8, 2026
Revenue share vs. one-time licence fees
Some data deals promise a share of future revenue; others pay a fixed fee up front. Here is how the two models differ and why a one-time fee usually fits company data licensed for AI.
Sep 3, 2026