Skip to content
World Model Readiness
Engraved team-functions instrument

For your team

Teams by Function

Module · Whether your data team can carry the load

The Data Team AI Check

Every AI initiative in the company eventually lands on your data team: it needs a clean pipeline, a defined metric, a dataset someone owns. If those are shaky, the AI built on top inherits the shakiness and calls it a hallucination. This module checks the five things that decide whether your team can carry that load: pipeline reliability, dataset ownership, shared definitions, an intake that matches your capacity, and quality standards the AI systems downstream can actually depend on.

Question 1 of 5 · Pipelines are reliable

Do your pipelines run reliably enough that people trust the numbers?

A pipeline that breaks silently is worse than one that breaks loudly: it feeds wrong data to everything downstream while everyone assumes it is fine. AI systems consuming that data cannot tell the difference. Reliability means failures are caught, surfaced, and fixed before they mislead anyone.

Question 2 of 5 · Datasets have owners

Does every important dataset have a named owner who answers for it?

A dataset with no owner is one nobody validates, documents, or fixes when it drifts. When an AI feature trains on it or queries it, the errors are silent and the accountability is nowhere. Ownership means a named person who can vouch for what the data means and how current it is.

Question 3 of 5 · Definitions are shared

Does your team share one definition for the metrics that matter?

When 'active user' or 'revenue' means three different things in three dashboards, an AI system asked about them will confidently pick one and be wrong for two audiences. A semantic layer or a shared definitions doc is what stops the same word from meaning different things.

Question 4 of 5 · Intake matches capacity

Is there an intake for data requests that matches what your team can actually deliver?

AI ambitions generate data requests faster than any team can absorb them. Without a real intake, the queue becomes whoever shouts loudest, and your team burns out serving noise. Intake means requests are triaged against capacity, not accepted by volume or seniority.

Question 5 of 5 · Quality SLAs exist

Do AI consumers of your data get a stated quality standard they can rely on?

An AI feature built on your data needs to know its freshness, completeness, and accuracy guarantees, or it will assume the best and fail on the worst. A data quality SLA states what the consumer can depend on, so a stale table becomes a known breach rather than a mystery bug in someone else's product.

For the statistics · one click each

Three questions for the public picture

These do not affect your score. They feed the anonymised, aggregated statistics; groups under 8 respondents are never shown.

How many people are on your data team?

What share of your important datasets have a named owner?

How does incoming demand compare to your team's capacity right now?

Your context

Used to calibrate the report. Company size and sector remain in the anonymized dataset; your email does not.

What the five levels look like

Every dimension in this assessment is scored 1 to 5. This is what the levels mean, dimension by dimension. The graded report diagnoses where your own answers land and what to do about it.

Pipelines are reliable

  1. 1Break silently
  2. 2Break often
  3. 3Mostly hold
  4. 4Monitored, alerting
  5. 5Reliable, observable

At the low end: A pipeline that fails without telling anyone poisons every decision downstream. Add failure alerting this month so a broken run stops the data instead of shipping it. What good looks like: Monitored, observable pipelines are the foundation everything else your team does sits on. Keep the alerting tuned; a monitor that cries wolf gets muted right before the real failure.

Datasets have owners

  1. 1No owners
  2. 2Owner in theory
  3. 3Some owned
  4. 4Most owned
  5. 5All owned, accountable

At the low end: Unowned datasets rot in place and take downstream models with them. Assign a named owner to your most-used datasets before anything new is built on them. What good looks like: Accountable ownership across your datasets means AI consumers inherit data someone stands behind. Keep the ownership map current through team changes; a departed owner leaves an orphan.

Definitions are shared

  1. 1Everyone differs
  2. 2Argued each time
  3. 3Some agreed
  4. 4Documented layer
  5. 5Enforced semantic layer

At the low end: When core metrics have no agreed definition, every report quietly disagrees with every other. Write down the definitions of your top metrics and get the team to sign off on them. What good looks like: An enforced semantic layer means humans and AI systems ask the same question and get the same answer. Govern changes to it deliberately; a quiet redefinition ripples through everything downstream.

Intake matches capacity

  1. 1No intake
  2. 2Loudest wins
  3. 3Informal queue
  4. 4Triaged intake
  5. 5Triaged against capacity

At the low end: With no intake, your team's roadmap is set by whoever interrupts them last. Put a single front door on data requests so work is chosen, not absorbed. What good looks like: Intake triaged against real capacity protects your team from the flood every AI initiative sends its way. Revisit the capacity assumption as demand grows; a stale number quietly recreates the overload.

Quality SLAs exist

  1. 1No standard
  2. 2Best effort
  3. 3Informal promise
  4. 4Stated SLA
  5. 5SLA, monitored, enforced

At the low end: Without a stated quality standard, every AI consumer of your data guesses at its reliability and blames the model when it is wrong. Define freshness and accuracy targets for your key feeds. What good looks like: Monitored, enforced quality SLAs let AI teams build on your data with their eyes open. Keep the thresholds honest; an SLA nobody measures against is decoration.