Resources
Buy vs build vs commission training data
Buy, build, or commission your training data. Most comparisons stop at cost and speed. The lens that decides it is legal defensibility.
Published 2026-07-22 · 6 min read
Key takeaways
- Buy, build, and commission trade off cost, speed, and fit in the usual way.
- The lens most comparisons skip is legal defensibility: can you show where each asset came from and the right to use it.
- A bought dataset is only as defensible as the seller’s provenance, so verify it before sampling.
- Building in-house is defensible only if you hold consent and source records at a marketplace standard.
- Commissioning through a rights-first marketplace is defensible by construction, because consent and provenance are captured at the source.
Every data team faces the same three paths. Buy an existing licensed dataset. Build your own collection in-house. Or commission new material recorded to your brief. Most comparisons weigh cost, speed, and quality, then stop.
They miss the lens that matters most once a model ships: legal defensibility. How well does each path hold up when someone asks where the data came from? That question decides more than the others combined.
Buy: licensed datasets off the shelf
Buying a ready dataset is the fastest path. The material exists, the format is known, and delivery is quick. The tradeoff is fit. An off-the-shelf set was built for someone else’s brief, so it may not match your gaps.
Defensibility depends entirely on the seller. A licensed dataset with a signed per-asset licence, documented chain of title, and consent artifacts is strong. A cheap dataset of unknown origin is a liability wearing a price tag. Ask for provenance before you ask for a sample.
Build: collect it yourself
Building in-house gives the most control. You choose the sources, the format, and the standard. The cost is time and overhead. Recruiting contributors, capturing consent, storing rights records, and maintaining all of it is real work that competes with model work.
Done well, building is highly defensible, because you hold every consent form and every source record. Done casually, it is the most dangerous path, because teams often scrape to save time and inherit every exposure that comes with it.
Commission: recorded to your brief
Commissioning sits between the other two. You describe what the model needs, and suppliers record it. You get material shaped to your gaps without running the collection yourself. It is slower than buying and faster than building from zero.
Defensibility is strong when the commission runs through a marketplace that captures rights at the source. The material is new, so it cannot already be in a crawl, and consent is gathered as it is recorded rather than reconstructed later.
The lens competitors skip: legal defensibility
Cost and speed are easy to compare. Defensibility is not, so most guides avoid it. It is also the lens that survives a model release. When a claim or an audit arrives, the question is not how cheap the data was. It is whether you can name where each asset came from and show the right to use it.
Rank the three paths by that question. Buying is only as defensible as the seller’s provenance. Building is as defensible as your own discipline. Commissioning through a rights-first marketplace is defensible by construction, because provenance and consent are captured before delivery.
How to choose
Start from the gap, then the risk. If a clean licensed set already fits your brief, buy it and verify provenance. If your need is specific and no existing set matches, commission it. Build in-house only when you have the operational appetite to hold rights records at the same standard a marketplace would.
Most teams end up blending paths. The constant across all of them is the same: whatever you choose, keep provenance attached to each asset.
Frequently asked questions
Is building in-house always the safest option?
Only if you hold rights records at the same standard a marketplace would. Teams that build casually often scrape to save time and inherit contract, privacy, and biometric exposure. Discipline, not ownership of the pipeline, is what makes building safe.
When does commissioning beat buying?
When your need is specific and no existing dataset matches it. Commissioning records to your brief, so the material is new and cannot already sit in a crawl, and consent is captured as it is recorded.
Related resources
Want data that clears this in diligence?
Whether you're building a model or sitting on an archive, the first conversation is short and specific.
Send a brief