Feature
August 10, 2026
Questions for every data quality demo
A bake-off script that surfaces noise, access, and price before anyone shares a service account.
REDWOOD CITY, CALIFORNIA — Most data-quality evaluations fail for a quiet reason. The room agrees the problem is real. The vendor shows a beautiful interface. A proof of concept starts. Six weeks later the channel is muted, the invoice is a second incident, and nobody can say whether the tool caught the break that actually mattered.
Vectornosis built VectorData for the morning when a warehouse looks fine and is not. That is a different job from inventorying every table in a catalog. It is closer to an on-call product than a documentation product. The questions below are the ones we ask teams to take into every demo — including ours.
They are not gotchas. They are how you find out whether you are buying operations or buying theater.
How many notifications does one upstream failure create?
Rename a column in staging and watch what happens. In a lot of stacks you do not get one page. You get a skipped model, eight failed tests, three freshness flips, and a Slack thread that reads like a stack trace. People mute the channel. Then leadership asks why revenue looks off.
Ask the vendor to simulate a single upstream break and count the messages. If the answer is “it depends” or “you can tune that,” you are buying a part-time job. VectorData is opinionated here: one calm alert per root cause, with blast radius and a fix-first ranking in Watchdog. The scarce resource is not another monitor. It is attention at 7:12am.
Bake-offs go better when the room watches a broken run together instead of debating screenshots.
Do you need our rows, or is metadata enough?
Row sampling sounds helpful until security review starts. Then the deal pauses while someone explains why a reliability tool needs order lines and event payloads. VectorData reads table timestamps, dbt artifacts, and job statistics. Customer rows stay in the warehouse. If a vendor cannot tell you whether a table is stale without sampling it, ask why INFORMATION_SCHEMA was not enough.
This is not a slogan. It is how VectorData is built, and it is why a security questionnaire should take a meeting instead of a quarter.
What did last quarter’s invoice look like for a team our size?
Per-host and usage-based pricing made sense for infrastructure. For warehouse health, a bill that grows with “how carefully you monitor” creates a perverse incentive: turn monitors off to save money, right when you need them. VectorData prices by plan. Managed starts at $999. The Data Stack Health Audit is $1,500–$2,500 and credits toward Managed. Ask every other vendor for invoice variance. Surprise is a feature of some models. Predictability is a feature of ours.
“If I cannot see a broken run in a browser before I share credentials, I am not in a product evaluation. I am in a pitch.”
Ricky Jiménez Sparks, CEO of Vectornosis
Can I see a broken run before I connect a warehouse?
The VectorData command center is live. You can watch a production-style run fail, read the blast radius, and decide whether the triage feels like work you would actually do on a Monday. Upload your own run_results.json if you want the story to be yours. That is the fourth question, and it is the one that ends most theater.
How to run the hour
Give the vendor a real failure, not a screenshot of a healthy board. A renamed column in staging is enough. Watch the clock. Count the messages. Ask who is supposed to act, and whether that person can tell the children from the parent without opening a second product. If the hour is a tour of integrations, you are not in an evaluation. You are in onboarding for a tool you have not chosen.
Bring the person who actually gets paged — not only the person who signs. Bring someone from security if row access is on the table. Write the four answers in the notes while the vendor is still in the room. Memory is generous to decks. Notes are not.
What good looks like in VectorData
Open /app. A run is already broken. Watchdog has a rank. The blast radius is a family, not a firehose. You can upload run_results.json if you want the story to be yours. Nobody has asked for a service account yet. That sequence is the product: see the failure, then talk about access, then talk about the plan.
Managed starts at $999. The Data Stack Health Audit is $1,500–$2,500 and credits toward Managed when you want a working session on your graph. Those numbers belong in the bake-off next to the four questions. A tool that cannot say the price in the same hour it shows the break is asking you to buy a mystery.
If a vendor needs a week to “tune noise” before you can see a single upstream failure, the noise is the product. VectorData is built so you do not spend the first month teaching the software to be quiet.
Take those four into the next bake-off. Try them on us at vectornosis.com/app. If the answers are crisp, you are looking at an operations product. If they are a slide, keep walking.
