Define a measure once and every AI will agree with your dashboard
Why the layer of meaning over a company's data has become a feature of the platforms it already owns and what the connectors still cannot carry and why the real work is writing the definitions down.
What went wrong when teams first gave AI their data
Two years ago a team would export a spreadsheet and paste it into a model and ask for analysis. The model read the rows and estimated an answer and the answer contradicted the dashboard because the dashboard was built on defined measures and the model was guessing at them. Every definition had to be typed into the prompt again and again. Then teams moved to AI writing code over the file and the sums became exact but the model still did not know how the tables related or what a measure meant.
That has changed. The main data platforms now let an AI read a semantic model. A semantic model is the layer that says what each table is and how the tables relate and how each measure is calculated and who may see what. When an AI answers through that layer the sums are done by the data engine and the AI only explains them. The numbers match the dashboard because they come from the same place.
The layer is now a feature and not a product
Microsoft has it in Power BI and Fabric with Copilot on top. Google has it in Looker and BigQuery. Snowflake and Databricks ship their own. The companies that sell marketing data connectors have each added one as well. For a business that already pays for one of these platforms the layer is there and the cost of using it is marginal. Nobody should build it again from scratch and nobody needs a new subscription to get it.
What the connectors still cannot carry
A connector moves data that already exists in a platform with a known shape. Media delivery and web analytics and sales come through cleanly because the industry settled their definitions years ago. What AI models say about a brand does not exist in any platform until somebody asks the questions and keeps the answers. What people believe about a brand lives in a survey. What search shows at each step of a journey has to be collected. None of that arrives through a connector and none of it has agreed definitions yet.
The work is the definitions
So the valuable work is not the pipes. It is deciding once what each measure means and writing it down where every AI the business uses can read it. What counts as the brand being named in an answer. What counts as an incorrect fact. Which step of the journey a question belongs to. How a reading taken this month is kept comparable with the one taken next month. A business that writes those down once gets the same answer from its dashboard and its chatbot and the assistant in its spreadsheet. A business that does not gets four numbers for one measure and a meeting about which is right.
One measure as four tools report it
The chart shows the same measure for the same month as four tools in one marketing team reported it. The spread is not an error in any of the tools. Each one is right by its own definition. The team had never agreed one.
Questions people ask me
What is a semantic layer in plain words?
It is the written agreement between a company's data and the people and the AI that read it. It says what each table holds and how the tables join and how each measure is worked out. A dashboard reads it and so can an AI assistant. When both read the same layer they give the same number.
Does a business need a new tool to get a semantic layer?
Almost never. If the business already uses Power BI or Looker or Snowflake or Databricks the layer is a feature it already pays for. The work is defining the measures and that is a job for the people who own the numbers and not for a new subscription.
Why does an AI give a different number from the dashboard?
Because it read the raw data and guessed at the definition. Point the AI at the same layer the dashboard uses and the sums are done by the data engine and the AI only explains them. The difference disappears.
Can what AI models say about a brand go into a semantic layer?
Yes once it has been collected and defined. It does not arrive through a connector because it does not exist until the questions are asked and the answers kept with their dates. Once it is defined like any other measure it sits beside sales and traffic and the business can ask about all three in one question.
My details are on the contact page if your team has settled its definitions and found something different.