The semantic layer and the self-service fallacy
Why self-service won't solve all your problems
👋 Hello! I’m Robert, CPO of Hyperquery and former data scientist + analyst. Welcome to Win With Data, where we talk weekly about maximizing the impact of data. As always, find me on LinkedIn or Twitter — I’m always happy to chat. 🙂
At Airbnb, we had a robust semantic layer. And it was transformative. Everyone in the company had ready access to a glut of metrics and dimensions by which to cut them, and nearly everyone in the company leveraged this power. Having been at companies where business logic is not so codified1, I can affirm that this is a world that is undeniably better. As a company grows, scalable access to data is the holy grail of analytics. Everyone wants food, and the self-service metrics layer is like a salad bar2. It’s the most efficient way to ensure everyone is, at some base level, able to be fed, able to get data. Base needs met. Hunger sated.
That said, while this is a panacea for some companies, for most, it isn’t.
"The poor eat to end their hunger. But when you have more than enough to eat your hunger doesn't end".
The fallacy of self-service is that analytics is about data exposure. It’s a myth perpetuated by the decade-long crusade [lie] that Business Intelligence is tantamount to dashboarding, when real insight, real action, real strategy-influencing, world-changing analytics comes from insights, not dashboards. Recommendations, not reporting. Knowledge, not data. Interpretation, not just raw exposure.
I’m still confident that the promises of the semantic layer will spur step-function change, certainly, but I want to stem the buzz and help you understand what the world will look like in a post-semantic world. All your problems won’t be solved, and I imagine things won’t change as much as you’d think.
In terms of impact to the business, surely the semantic layer will be uniquely enabling, revolutionary even. We’ll have to worry less about ad hoc requests for data that others can just pull themselves. If you’re in a position where you’re completely overwhelmed by metrics-level reporting inquiries, this might just give your team enough purchase to proactively build a self-guided roadmap.
But in general, the questions won’t go away get immediately better — the flavor of typical question will simply change. “Do you have metric X cut by dimension Y?” will become “why did this metric drop?” or “how do we define this metric?”. Self-service certainly isn’t a lie, but induced demand is a reality.
So why is this so unsatisfying? The business is getting more answers to more questions, after all. We should be happy. But fundamentally, the problem we as analysts want self-service to solve is not actually solved by self-service, and that is that analytics often sucks. The semantic layer may answer more questions, but it won’t fix the bad loops that shape our perception as SQL monkeys. No tool can fix people, behaviors, process, and the semantic layer, however conceptually elegant or impactful, is no exception.
That said, not all is so dire. The semantic layer will open up opportunities for you to change how you operate. A metric is a metric is a metric, and when it’s needed, there’s not much you can do but deliver it. But with more nuance comes more opportunity for investigation, thought, root cause analysis, Asking Why. And in this world, business acumen, communication, contextual immersion will be more highly valued, for you can’t fake insights and thought partnership as readily as you can fake a dashboard with plausible utility.
The semantic layer will change things, and while it won’t directly solve the quality of life problems you want it to solve, it will present new opportunities to do so.
A final comment: if this resonated with you, I’d recommend checking out my article “The bad loop ruining analytics”, where I dive into how we can think about elevating the status and value-add of analysts. And check out Hyperquery if you are curious how tooling might be able to help.
Including internally at Hyperquery, honestly. Our data stack is toothpicks.
Yes, yes, I know I’ve compared it to Blue Apron before, but it really is more like a buffet.