How to write a data brief for an AI sourcing request
A practical guide for AI teams: what to include in a data requirement so that potential sources can be identified, qualified and approached efficiently.
For AI teams ·
When the data a model needs isn't available publicly or from existing vendors, it has to be found: in the systems and archives of organisations that generate it as part of their work. The quality of that search depends heavily on the quality of the brief that starts it.
A good data brief doesn't need to be long. It needs to be specific about the things that determine whether a given source is worth pursuing.
Start with the model task
Describe what the model should be able to do, not just the data you think you want. “Diagnose likely causes of pump failure from technician notes and sensor history” tells a sourcing partner far more than “maintenance data”. It points to which records matter, which fields need to be present and what a useful example looks like.
Define domain, modality and coverage
Be explicit about the industry and sub-sector, the data types involved and the variety you need. Coverage often matters as much as volume: a smaller set spanning many sites, equipment types or time periods may be more valuable than a large set from a single source.
- Industry and sub-sector
- Modalities: structured records, documents, images, video, expert examples
- Geography and language
- Time period and recency
- Edge cases or rare events that must be represented
State the intended use
Pre-training, fine-tuning, evaluation and retrieval each imply different requirements and different licence terms. Data owners will reasonably want to know how their data will be used before they agree to anything, and some will permit certain uses but not others. Stating the intended use up front saves time on both sides.
Set quality and format thresholds
Note what would make data unusable: missing fields, inconsistent labelling, unreadable scans, insufficient detail in free text. If you need data in a particular structure, or are willing to handle preparation yourself, say so. Preparation effort is often a deciding factor in whether a source is viable.
Give an indication of scope, budget and timing
Approximate volume, budget range and deadlines help prioritise which sources are realistic. They needn't be precise. Even broad ranges make it easier to tell early whether a route is worth pursuing.
Keep the first brief high-level
A first brief should describe the requirement, not include sensitive material. Detailed specifications, samples and terms come later, once a potential source has been identified and both sides are ready to go further.
If you're ready to start, you can submit a data brief through our enquiry form. We'll follow up to understand the detail and outline a potential sourcing route.