Answer six questions. You get the rung of the causal ladder you are standing on, the requirement between you and the rung you came for, and how many days it takes to meet it. No email, no sales call.
Answer for the messiest case you can think of.
No email required. Nothing leaves your browser until you choose to book.
Maturity models normally exist to tell you that you are not ready yet. This one names the single requirement that is currently binding. In practice there is almost always only one, and it is almost always smaller than the fear of it.
The five rungs are ordered because each one depends on the rung below it. External weather data cannot resolve onto sites that have not been resolved yet. A causal graph learned over duplicated customers encodes the duplication as a business relationship. Every rung is also useful on its own, so nothing here asks you to climb all five before anything works.
Most teams land higher than they expect, because the parts that used to take quarters are the parts we automated.
This is the model the six questions read against, written out in full. You can skip the assessment and place yourself.
Days. Connection to a working ontology usually lands inside the first week.
Read the pageRead access to the systems the decision touches. No glossary, catalog or dbt project has to exist first.
Every question restarts from joins and mappings. Two analyses of the same metric disagree and nobody can settle which one is right.
An ontology with entity, time and location already resolved, so external data has something to join onto.
External data lands at the wrong grain and the wrong geography. The enrichment project outlives the question that started it.
Minutes to connect a client. Access rules follow the user who asked.
Read the pageGoverned definitions to query against, so the agent asks for a named metric rather than guessing at a schema.
The agent writes fluent SQL against the wrong table and presents the result with confidence. The failure is silent, which costs more than an error.
Resolved entities and a settled grain, plus observed variation in the levers you care about.
Structure gets learned over duplicated entities and mismatched grain. The graph then encodes your data problems as business relationships.
Immediate once a model exists. Scenarios run live in the meeting.
Read the pageA validated causal model, and a lever that has moved at least once.
You get a confident number with no way to check it. Intervention estimates on an unvalidated graph are the most expensive kind of wrong.
This assessment cannot see your data. It is reliable about the ordering, because the dependencies between the rungs are structural. It stays quiet on whether one specific effect is estimable.
That answer depends on variation you have not described here, on the grain your events land at, and on whether something unmeasured is driving both of the things you care about. We check it against real systems in the first session. Sometimes the finding is that a lever needs instrumenting before anything can be modeled from it, and we say so on the call.
One real question and the systems behind it. On the call we tell you whether it is estimable, including when the answer is no.