The question behind this one is usually a hiring question. You have one to three people, a warehouse, and a backlog, and someone has asked whether data quality means adding a QA engineer. For the mechanical checks, it doesn’t. For the business rules, it never did, because a QA engineer who doesn’t know your business can’t write them either.
What one afternoon looks like
These figures come from DataKitchen’s own published runs, not from a customer.
| Step | Time | Source |
|---|---|---|
| Install with Docker, open the UI, first quality score | About 15 minutes | Note to Great Expectations customers, May 2026 |
| Download, load, profile, and test a month of NYC taxi trips | About five minutes end to end | NYC taxi run, August 2026 |
| Profile 2,964,624 rows | 1 minute 48 seconds | Same run |
| Run 69 tests | 6 seconds | Same run |
| Freshness, volume, and schema monitors on every bronze table | About an hour | Medallion testing post, June 2026 |
The taxi run was one table on a laptop. A real schema with dozens of tables takes longer to profile. It doesn’t take longer to set up, because you don’t write a test per table.
What comes out the other end: a profile of every column on 55 characteristics, findings from 32 hygiene detectors, and tests from 34 generated test types. DataKitchen’s estimate is that generation covers about 80 percent of the tests you need. The same post prices the alternative. Hand-writing 2,500 tests at 30 minutes each is 1,250 hours.
That’s about 7.2 months of a senior engineer’s typing, per the same post, and you skip it.
The 20 percent nobody can generate
Generated tests learn the data in front of them. In the NYC taxi run, all 62 generated tests passed on a 300,000-row sample, because the minimum-fare test set its floor at negative $367. On the full month, 11 of 69 tests failed, including 4 generated minimum-value tests and all 7 business rules an AI wrote from the profile. A fare is never negative. No profile tells you that. Somebody on your team does, or an AI drafts it and somebody checks it.
That somebody is the analyst who reads the report, not a QA hire. Ask them to write rules from a blank page and you get four rows back. Show them a draft rule to correct and they fix it in seconds.
The job that’s really left: triage
A QA engineer on a big team spends most of the week deciding which failures matter. On a small team, that falls to you, and the fastest way to fail at it is to page someone for every statistical wobble. DataKitchen’s advice in How Do I Get My Data Team More Efficient? is to split on day one. Hard constraints fail loudly. Volume ranges and distribution checks run warn-only for a couple of weeks, then get promoted. Skip that split and someone mutes the channel.
What it costs at one, two, and three people
From the pricing page, October 2026. Every supported database works in both editions.
| Team | Edition | Monthly cost on one database |
|---|---|---|
| One person | Open source | $0 |
| Two people | Enterprise | $300 |
| Three people | Enterprise | $400 |
Open source is the full engine with every feature, limited to one user, one connection, and one project. A second person is where Enterprise starts.
For the first week’s plan, table by table, see how a two-person team should start. For keeping monitoring overhead low past month one, see the startup monitoring answer. And if your team has no engineer at all, testing pipelines without a data engineer covers that case.