Latest / MBA Training Data / Building a semantic layer for consistent metrics across business units
Transcript
- 0:00Welcome to the Leaders Insights Podcast. Today's episode, Building a Semantic Layer
- 0:09for Consistent Metrics Across Business Units. Two people walk into a board meeting with
- 0:14the same company's revenue number, and they disagree. How is that even possible in
- 0:192026?
- 0:20Because they're not reading the same number. They're reading the same word. Finance says
- 0:25revenue. Sales says revenue. And each of them has quietly coded their own
- 0:30definition into their own report. One counts a deal when the contract signed,
- 0:34the other when the cash lands. Same label, different math. So it's not a data
- 0:40problem. It's a definitions problem. Exactly. The data is usually fine. The
- 0:45warehouse in Dors, the central place where all the raw numbers live,
- 0:49MOS, is fine. What's broken is that every business unit builds its own logic on
- 0:53top of that data in a spreadsheet or a dashboard. And nobody wrote down the
- 0:58rules. That's where the semantic layer comes in. Define that for me because it
- 1:03sounds like consultant vapor. Fair. A semantic layer is a single translation
- 1:07layer that sits between your raw data and every report anyone builds. It's one
- 1:13place where you define, encode, what revenue means, what active customer means,
- 1:18what churn means, the share of customers who walk in a given period.
- 1:23Every dashboard then has to ask that layer for the number instead of
- 1:27computing its own. So it's a dictionary that everyone's forced to use. It's a
- 1:31dictionary with a bouncer. You can't sneak your own definition of revenue past
- 1:36the door. And that's the whole point. Consistency isn't a policy you email
- 1:40around. It's something you enforce in the plumbing. Give me the concrete
- 1:44version. Where does this actually live? Tools like DBT. That's a data
- 1:49transformation tool. It lets you define metrics as reusable code or cube or
- 1:55looker's modeling layer at Google. You write the definition once. Monthly
- 2:00recurring revenue equals the sum of active subscription values recognized on
- 2:04the first of the month. Then Tableau, your finance team, the CEO's phone app.
- 2:10They all pull from that one definition. Change it once. It changes everywhere.
- 2:14Sounds great. But every CDO listening is thinking, that means ripping out my
- 2:19warehouse and rebuilding for a year. No. And that's the misconception I want
- 2:23to kill. The semantic layer sits on top of what you already have. You don't
- 2:28touch the warehouse. When I was advising a mid-market software firm last year,
- 2:33they had three business units reporting revenue that differed by 11 percent.
- 2:37About $4 million of phantom disagreement every quarter. We didn't migrate
- 2:41anything. We spent six weeks defining 40 core metrics and put a semantic
- 2:46layer over their existing snowflake setup. The gap closed to under half a
- 2:51percent. 40 metrics. That's the part that scares me. Who decides what revenue
- 2:56means when finance and sales both think they're right? That's the actual hard
- 3:00work. And anyone who tells you the technology is the hard part is selling
- 3:03you something. You need a metric owner for each definition. One named human,
- 3:08not a committee. Finance owns revenue. Full stop. Sales can have its own
- 3:14bookings metric, defined separately and labeled honestly. So nobody confuses the
- 3:19two. The trick isn't forcing everyone into one number. It's making sure every
- 3:24number has one clear definition and one owner's name on it. What happens if you
- 3:29skip the owner and just let the tool enforce whatever's already there? Then
- 3:33you've built a very expensive way to make the same argument faster. The tool
- 3:37enforces the definition you give it. Garbage definition in. Consistent
- 3:42garbage out. It's like buying a Swiss watch and setting it to the wrong time.
- 3:46Beautifully precise. Still makes you late. How do you know it's working? What's
- 3:51the tell? The meetings get shorter. When people stop the first 15 minutes of every
- 3:56review arguing about whose number is real, that's the signal. One retail
- 4:01client of mine clocked it. Their quarterly business review used to run
- 4:0440 minutes of number reconciliation before anyone made a decision. After the
- 4:09semantic layer, zero. Give me the one thing to do Monday morning. Pick your
- 4:14single most argued about metric, usually revenue or active users and write down
- 4:19its exact definition in one sentence with one owner's name attached. If you
- 4:24can't, you just found your problem. That's a wrap. Fresh CDO briefings drop
- 4:29daily at MBA-training.com.