Zovda
Data budgets are compared per unit: per row, per hour, per image. The unit that matters is the one that survived your checks, and nobody quotes that number.
A team that buys on the sticker price usually discovers the real one during the first training run.
Four costs hide behind the unit price
- The share of rows your own checks reject
- The hours spent reconciling two sources
- The rework when a field turns out to be unusable
- The delays when a batch arrives late in a format that needs a day of reshaping
None of these appear in a quote and all of them appear in a month.
Measure the usable rate
Take the last batch, run it through your own acceptance checks, and count what passed. That rate is the divisor for the unit price, and it is different for every vendor and every task.
Do it once per vendor per task type and keep the number. It is the only input that makes quotes comparable.
Price the ambiguity
A task with a clear guideline needs fewer review hours than one where the vendor had to invent the rules. If the quote is the same, one of the two is mispriced.
Pay for the guideline work separately and explicitly, or it will be done badly and for free.
Keep the reconciliation cheap
Two sources with different schemas cost days. Two sources with the same schema cost hours. Sometimes the cheaper quote is the one that ships a mapping.
Zovda ships each collection through Datuva with a field contract, so the acceptance check is a script rather than a week. The corpus catalogue shows what each family includes.
Tell us what your acceptance check looks like. We reply within one business day.