Data modeller: which parts can AI do?
A data modeller needs Data modelling: 8 activities and 307 knobits, the smallest units of knowledge and skill worth learning and testing on their own.
How much of this job can AI do?
130knobits most AI models master
35that some models master
34where people still lead
108not tested yet
Of the 199 knobits the benchmarks cover, 65% are mastered by most of the 10 AI models tested. Mastering the knobits in a test is not the same as doing the job: someone still decides what to build, talks to the business, and answers for the result. But it shows where the work is shifting.
Activities, from most to least covered by AI
| Activity | Best AI model | Typical AI model (median) |
|---|---|---|
| Build the conceptual model and taxonomies | 100% | 100% |
| Derive data quality rules | 98% | 94% |
| Derive the logical data model | 95% | 94% |
| Document data rules as SBVR fact types and rule statements | 100% | 93% |
| Design the physical data model | 98% | 92% |
| Define business terms | 99% | 91% |
| Capture data requirements | 88% | 72% |
| Generate and review DDL | 82% | 70% |
For your career: where people still lead
The knobits the fewest AI models showed they master. Being good at these is what a person adds next to AI tools.
- K-103 Recognises a statement outside the scope of data requirements
- K-821 Recognises a restricted column without access design
- K-810 Recognises a truncated identifier taken as the real name
- K-117 Recognises a missing or wrong requirement type
- K-816 Recognises one decimal scale for every currency
- K-309 Recognises a gap in a partition of cases
- K-808 Recognises a CHECK constraint parsed and ignored
- K-820 Recognises a default inventing a business fact
- K-122 Recognises paraphrased data requirements
- K-116 Recognises a missing or meaningless requirement priority
- K-121 Recognises duplicate data requirements
- K-718 Recognises a misfiled data quality dimension
Where AI already helps the most
Knobits that almost every model tested masters: good candidates to hand to an AI assistant and review.
- K-160 Explains persistent identifiers
- K-135 Retires a superseded register entry instead of deleting it
- K-251 Explains intensional definition by superordinate concept and delimiting characteristics
- K-252 Explains the substitution principle for definitions
- K-253 Explains generic and partitive concept relations
- K-256 Explains essential and delimiting characteristics
- K-355 Explains Witt's classification of business rules
- K-356 Explains rule enforcement mechanisms and platform enforcement
Choosing an AI model for data modeller work
Best result for data modeller work: gemini-4-argon (89% test score).
Per-activity results are on each activity page; all models are compared on the benchmarks page.
Test yourself
Skilirium's adaptive test measures you on the same knobits as the AI models. How a test works.