Data work is easy to buy and hard to buy well. The difference shows up in the first deliverable, which is why the evaluation should happen before the project rather than after.
Ask for a sample — and read it carefully
Ten rows tell you more than any portfolio. Check accuracy first, then formatting, then how they handled ambiguity. If two of ten rows have a formatting inconsistency, two hundred will have twenty.
A supplier who will not send a sample before a paid engagement is telling you something useful.
Listen to the questions they ask
A good partner asks about definitions: what counts as verified, how to handle a missing field, which country format wins, whether a partial match should be merged or kept separate. Those questions mean they have done this before.
A supplier who simply starts without asking anything will make every one of those decisions for you, and usually not the way you would have.
Look for process, not promises
- Source tracking per row so any error can be traced
- A defined verification method, stated precisely
- Deliverable format agreed in writing before the work starts
- A stated revisions policy with a clear definition of an error
- Regular progress updates for anything longer than a few days
Warning signs
- No sample, or a sample with obvious errors
- Prices far below the market, which usually means automated output with no verification
- Unwillingness to name the sources used
- Deliverables in screenshots instead of actual data
- Vague answers about how the data will be verified
The real calculation
Compare the cheap quote plus your own cleanup time against the careful quote that needs no rework. In practice the second option is almost always cheaper, because cleanup is slower than original work and far more frustrating.
The cheapest list is the one you do not have to fix.
Start with a small paid batch
A short checklist before you commit
- A sample of ten to twenty rows, reviewed by hand
- Sources named for every field, not just the email
- The deliverable format agreed in writing, including column order
- A definition of an error and how corrections are handled
- An agreed turnaround for the first batch
- Two hundred rows delivered before any larger order
Read the sample for how ambiguity was handled, not only for whether the rows are correct. The decisions made without asking are the ones that cost time later.
What to measure on the first batch
Record the error rate against your own spot check, the time from order to delivery, and the number of questions the supplier asked. A low error rate with few questions is fine. A low error rate with no questions at all usually means the definitions were guessed.
Keep the cleaning standard agreed with the supplier in mind. The steps a good cleaner follows are set out in cleaning a messy lead list in Excel, and the checks that matter most for outreach are in how to verify email addresses.
Rather than committing to a large volume, commission two hundred rows and review them properly against your own spot checks. That single test predicts the quality of the next twenty thousand far better than any conversation about capability.