How to brief a data sourcing request for your model team
What to put in a request for licensed enterprise data: data type, industry, company size, department, model and terms, so the match comes back faster.
Written by Jimmy Lin
, 2 min read
A good sourcing request names what your model needs to learn, not just which tools the data comes from. Give us the data type, the industry, the company size, the department and the kind of model, and we come back with what exists and what we can source. The more specific the brief, the better the match.
Start with the task
Tell us what you are training and what it should do. A support agent, a model that drafts proposals, an evaluation set for finance work, a classifier. Language models and non-language models both work, and the task decides which data fits.
This matters more than the file format. Two companies can hold the same ticket system and very different tickets. The task tells us which one you want.
Choose the data type
Enterprise data comes in a few families: support tickets with resolutions, team chat where a decision got made, CRM history with deals won and lost, email and calendar, project and field records, code with issues and reviews, finance and ERP records, and documents, SOPs and knowledge bases.
Say if you need outcomes. Tickets closed with a resolution, deals marked won or lost, and changes approved or rejected can be rebuilt into tasks, trajectories and graders. Data without outcomes is worth less for that use.
Pick the industry, size and department
Sellers span many industries, company sizes and departments. Naming them narrows the search.
Industry. Logistics, construction and trades, professional services, accounting, IT services, software, manufacturing, insurance, property management and healthcare administration are all on the list. If yours is not, say so and we tell you what we can source.
Company size. Sellers start at around 20 people and go up. A small firm in a rare trade can beat a large generic one, so size is a preference, not a rule.
Department. Support, sales, finance and operations run different processes. If you want workflows, ask for them by department.
Decide on workflows
Beyond records, some sellers license how their teams work: documented processes, recorded task runs and AI-assisted tasks. These show the steps and the tools, not just the end state. If you build environments for agents, tell us and we filter for outcome-bearing, multi-system data and workflows.
Set the terms
Say whether you want exclusivity and for how long. Two years is common, a shorter term is possible, and licenses are never perpetual. Say whether you want a one-off corpus or a refresh later, and whether the scope is training, evaluation, environment construction or all three. Fees are discussed with you directly, with a long-term relationship in mind. Delivery format and schedule are agreed with you too.
What you get back
A one-page summary of the match: what the data is, its size, years, systems, outcomes, rights status and exclusions. A de-identified sample the seller has approved. A provenance memo with the package, covering ownership, contract clauses and exclusions. Personal information is removed by automated tooling and checked by trained human reviewers, and the seller reviews a sample before anything is accepted. You never see raw files.
The button below takes you to the buyer page, where you can send us your brief.
Questions people ask
Do you need to know the exact volume?
No. A rough range is enough. We say what exists and what we can source, and you refine from there.
Can I ask for data for a model that is not a language model?
Yes. Tell us the task and the data type, and we match sellers to it.
Do I choose the seller?
We bring you matches with the rights checked. You see a summary and a de-identified sample before you commit.
What if nothing on the list fits?
Tell us what you need. We say what exists and what we can source.