Mapping the World of Work

Speaker Notes

Here's the fundamental gap. Most talent databases — even large ones — treat profiles as flat documents. They have no concept of role equivalence, no skills taxonomy, no career trajectory data. When you search for "DevOps Engineer," they can only find that exact string. Censia is different. Because we've built a structured model of the entire worker universe, our search understands that an SRE is the same role, and that a Backend Engineer with Kubernetes and Terraform experience who's been in role for 42 months has the right skillset to fulfill DevOps duties. That Backend Engineer is invisible to every other platform. We surface them — with an explanation of why they're qualified.

Speaker Notes

This is how we model the job universe. We've mapped roughly 500 role archetypes that cover the entire job market. Every title in our database — and there are millions of variations — resolves to one of these archetypes. "DevOps Engineer," "SRE," "Platform Engineer," "Cloud Ops" — they all map to the same archetype. But it goes deeper. For each archetype, we know which skills define it — Kubernetes is 8.5 times more common among DevOps people than the general population. That's not from job postings — it's from analyzing 300 million real profiles. This lets us find people with the right skills who've never held the title. That's a talent pool no one else can see.

Speaker Notes

This is where our understanding of the worker universe becomes truly powerful. We've analyzed over 300 million career histories to build a transition graph — we know which roles feed into other roles, and at what rate. Cloud Engineers, Systems Admins, Backend Engineers — they all flow into DevOps at different rates. We also know the median tenure before each transition. This lets us compute a proximity score for every person, for every target role — not a percentage, but a composite signal of how close someone's skillset, experience, and trajectory align with the duties of a given role. John Smith is a Backend Engineer who's never held a DevOps title. But his capabilities place him in close proximity to fulfilling that role's core responsibilities. No other platform can surface that insight.

Speaker Notes

This is where we should talk about what's under the hood — at a strategic level. We don't use one search algorithm. We use four fundamentally different approaches, each designed to catch a different type of match. Precise matching catches exact alignments. Relevance scoring finds strong keyword signals. Neural embeddings understand conceptual similarity — that an SRE and a DevOps Engineer are the same thing even though they share no words. And learned sparse expansion infers terms a candidate would use, even if they haven't listed them. These four signals are then fused together using a weighted rank fusion engine — not a simple average, but a context-sensitive weighting that adjusts based on each query. The result is one unified ranking that combines precision with discovery. And critically — these vectors aren't searching raw text. They're searching through our enriched data layers: the skills taxonomy, the role archetypes, the career graph. That combination is what makes this proprietary.

Speaker Notes

Let me make this concrete. When someone searches for "Machine Learning Engineer," here's what happens across our four search vectors. Precise matching finds 340 exact ML Engineers. Relevance scoring adds another 1,200 profiles rich in ML terminology. Neural embeddings recognize that Applied Scientists and AI Researchers are conceptually the same — adding 2,800 more. And our expansion layer infers that certain Data Scientists with deployment experience are functionally equivalent — adding 4,100 more. That's a talent pool that went from 340 to over 8,000 qualified candidates. But the real magic is in the fusion. These aren't just unioned together — they're ranked by a weighted fusion model that considers which signals are most meaningful for each query. And crucially, these vectors are searching through our enriched data — canonicalized skills, resolved archetypes, embedded career context. That's why this is proprietary and defensible. You'd need our data layers AND our search architecture to replicate this.

Speaker Notes

Let me show you the four major outcomes this intelligence enables. First, talent sourcing — our search expands candidate pools by 3-5x because we surface equivalent roles and career-ready candidates, not just keyword matches. Second, workforce planning — we can map any organization's actual skill landscape against what's needed and flag gaps before they're critical. Third, internal mobility and succession — proximity analysis is pre-computed, so you can query "who's closest to Engineering Manager" and get a ranked list instantly. And fourth, competitive intelligence — we can show talent composition, skill density, and migration patterns for any company or industry. All of this is only possible because we've modeled the structure of the worker universe.

Speaker Notes

This is the moat slide — why this is hard to replicate. We've built four interconnected intelligence layers, and the key word is interconnected. The skills taxonomy feeds into role archetypes — you need to understand skills to cluster roles correctly. Role archetypes define the career transition graph — you need to know which roles are equivalent to track transitions accurately. And the transition graph drives readiness scores. You can't build layer 4 without layers 1 through 3. And you can't build layer 1 without years of curation — 29,640 canonical skills with human-reviewed mappings, disambiguation rules, industry-specific context. This isn't something you can bootstrap in a quarter. Add in 300 million real career histories as the training data, and our multi-algorithm search that fuses four different approaches into one ranking — this is a compounding advantage that gets stronger over time.