Home/Readiness assessment

Readiness assessment

Is your data ready for analytics agents?

Fifteen questions across the five things that decide whether an agent gives right answers: definitions, access, data quality, interfaces, and the team that runs it. Pick the answer closest to how things work today, not how they are meant to work. It takes about ten minutes.

Scored in your browser. Your answers are not sent anywhere; the result link keeps them in the address after the # sign.

Area 1 of 5

Semantic layer and definitions

An agent can only be as consistent as the definitions it reads. If "revenue" means three things, it will pick one without telling you.

1. How are your core business metrics (revenue, active customer, churn) defined today?

Inside each report or spreadsheet. The same metric is calculated differently in different places.Written up in a wiki or glossary, but each tool has its own SQL for it.Defined once in a semantic or metrics layer for some domains. Other domains still define their own.Defined once in code, under version control, and read by dashboards, notebooks, and agents alike.

2. When two teams use the same word (customer, order, region) with different meanings, what happens?

Nobody notices until two numbers disagree in a meeting.Analysts who know the history sort it out informally.The conflicts are written down, and each definition has an owner.Each meaning has its own named definition, and anyone or anything querying the semantic layer sees which one it is using.

3. Could a new analyst answer a routine question from the written definitions alone, without asking a colleague?

No. Most questions need knowledge that is not written down.For a handful of well-used datasets.For most questions in at least one business domain.Yes, across the domains we would put an agent in front of first.

Area 2 of 5

Governance and access control

Agents act faster and more often than people, and nobody watches each action. Access has to be enforced where the data is read, as the person asking.

4. How does a query tool or service get access to data today?

Shared service accounts or copied credentials with broad read access.Each tool has its own service account with role-based grants, reviewed now and then.Grants are enforced in one catalog or policy layer, but tools connect as themselves, not as the user.The user's identity is passed through the tool, and access is checked in one place with short-lived, scoped credentials.

5. Are row-level and column-level rules (by region, personal data, salary) enforced where the data is read, for every engine?

No. We protect sensitive data by keeping people away from whole tables or by making filtered copies.In some engines or BI tools but not others.Centrally for most sensitive data, with known gaps.Centrally, for every engine and interface that can read the data.

6. If someone asked who read a sensitive table last month, and through which tool, how long would the answer take?

We could not answer it.Days, stitching logs together from several systems.Hours, from one audit log that records the service account but not always the person.Minutes, from an audit trail that ties each query to a person and the tool acting for them.

Area 3 of 5

Data quality and freshness

An agent does not know a table is three days stale unless something tells it. A confident answer from stale data is worse than no answer.

7. How do you learn that a key table is stale or wrong?

A business user notices.Pipeline failure alerts, but no checks on the data itself.Automated tests on important tables (nulls, row counts, freshness) with alerts.Tests, plus a published freshness and quality status that downstream tools can read before they query.

8. For the datasets an agent would use first, is there a named owner and an agreed freshness target?

No.Owners are known informally.Owners are named; freshness expectations are informal.Named owners, written freshness targets, and a clear way to report a problem to them.

9. How many copies of your most-used datasets exist across warehouses, marts, extracts, and BI caches?

Many, and it is not clear which one is authoritative.Several, with one understood to be authoritative.A few, with the authoritative copy recorded in a catalog.One authoritative copy in open formats that every engine reads, with derived copies tracked.

Area 4 of 5

Open interfaces (MCP, SQL, APIs)

Agents reach data through interfaces. Standard ones let you change the model or the agent framework without rebuilding access and definitions.

10. How can software (not a person clicking) reach your governed data?

Only through the BI tool's interface or exported files.Direct database connections for a few engineering teams.Standard SQL interfaces (JDBC, ODBC, Arrow Flight) and documented APIs, governed like any other client.Standard SQL and APIs, plus an agent-facing interface such as an MCP server that exposes the semantic layer and respects the caller’s permissions.

11. Can an agent discover which datasets and metrics exist, with descriptions, without someone pasting a schema into a prompt?

No.Table and column names only, from the database catalog.A searchable catalog with descriptions for the important datasets.A catalog and semantic layer reachable through an API or MCP, with descriptions, owners, and definitions.

12. If you changed the model or agent framework next year, what would you have to rebuild?

Most of it. Definitions and access rules are written into prompts and one vendor's tooling.The integrations, plus definitions that live in the agent's configuration.Only the agent integration. Definitions and access live in the data platform.A configuration change. The agent reaches data through open interfaces the next framework also speaks.

Area 5 of 5

Team skills and operating model

A production agent is a data product. Someone has to own it, measure it, and maintain the definitions it depends on.

13. Who would own an analytics agent in production?

No one has been named.An innovation or AI team, separate from the data platform and analytics teams.A named owner, with data platform and analytics engineering involved part time.A named owner with a standing group: data platform, analytics engineering, security, and a business domain owner.

14. How would you know whether an agent's answers are right?

We would rely on users to spot problems.Analysts would spot-check answers during a pilot.A set of questions with known correct answers, run before changes.A maintained question set run on every change, plus logged corrections and refusals in production that feed the definition backlog.

15. How ready are your analytics and data engineering teams to maintain definitions and access policies as code?

Definitions and policies live in tools' settings screens, with little version control.Some transformation code is in version control; definitions and policies are not.Transformations and metric definitions are in version control, with review.Definitions, policies, and tests are reviewed and deployed like application code, and time is set aside to maintain them.

0 of 15 answered

Your result

By area

Then

The 90-day rollout playbook: phases, owners, and exit criteria for a first domain.
The use-case library: pick the domain and the first questions.
The evaluation checklist: what to ask vendors, once the foundations are in place.

Link to this result

About this assessment

Do my answers leave the browser?

No. The scoring runs in JavaScript on this page and nothing is sent to a server. The share link keeps your answers in the part of the address after the # sign, which browsers do not send to the site. Google Analytics measures visits to this site and records that the assessment was started and finished, and the level reached, but not your answers.

How is the score calculated?

Each answer scores 0 to 3, from the first option to the last. Each of the five areas is scored out of 9, and the overall score is out of 45. The level comes from the overall percentage: Exploring below 40%, Building from 40%, Ready from 65%, Scaling from 85%. One rule overrides it: if any area scores 3 or less out of 9, the level cannot be higher than Building, because an agent fails at the weakest point, not the average.

Who should fill it in?

The person accountable for analytics, ideally with the data platform lead and an analytics engineer in the room. The disagreements between them are often more useful than the score.

What should I do with the result?

Work on the areas listed under next steps before scaling anything. If the level is Ready or Scaling, the 90-day rollout playbook is the next page to read. Save the share link and take the assessment again after the pilot to compare.

The other leader tools

Use-case libraryFinance, sales and revenue operations, supply chain, customer support, and marketing: the questions people ask, the definitions an agent needs, and the governance risks.
Evaluation checklistVendor-neutral RFP questions on semantics, governance, lock-in, accuracy, cost, and operations, with what a good answer contains. Print it or download the Markdown.
90-day rollout playbookFive phases across thirteen weeks, from picking a domain to deciding whether to expand, with owners, exit criteria, and the risks that stall each phase.