Three fields, and the routing model pointed at all of them
Routing and reporting were both built on the field naming the account executive, and I measured it at 3.9 percent fill.
The SDR field came out at 0.7 percent and the lead status field at 1.3 percent. All three rates are from one sample of 1,000 of the 39,731 companies in the portal, taken oldest first by record id. Those three fields are what our routing model pointed at, so every rule we had written against them could only ever apply to a small fraction of the records.
Nothing about that fails loudly. A rule whose field is empty evaluates to nothing and reports nothing, so the rule stays in the documentation and the process it describes never runs.
3.9 percent on a sample of 1,000 companies. That is how often the field naming the account executive actually had a value in it.
Our routing, our reporting and several of our written rules of engagement all pointed at it.
From six months inside a B2B GTM organization with 36,000 accounts in its CRM.
What the whole schema looked like
I checked all 437 of our company fields for how often they carry a value, against the same sample of 1,000 of the 39,731 companies.
| Fill rate on the sample | Fields |
|---|---|
| Above 50 percent, genuinely in use | 57 |
| Between 5 and 50 percent, used sometimes | 28 |
| Under 5 percent, barely used | 186 |
| At zero, never used once | 166 |
352 of the 437 sit in the bottom two rows, which is over 80 percent of the schema, and that puts the three routing fields in the majority. Around them, in our own schema: marketplace fields between zero and four percent, a qualification framework under one percent, a customer health score on a fraction of a percent of records, 21 empty lifecycle timers for pipeline stages nobody ever used, and test fields still sitting in the live schema.
The caveat on that number, and it is mine
The sample was the 1,000 oldest companies by record id, and that changes what the number means.
Fields that arrived recently, which is most of the enrichment and AI fields, are systematically understated by a sample of the oldest records, so their real fill rates are higher than this reads. I marked that rather than letting the number stand on its own, and it is the first thing to fix if anybody repeats the exercise: sort the sample randomly, or take one sample of oldest and one of newest and compare them.
What the caveat does not change is which fields the routing was built to read. A rate this low on any cohort is the finding that matters for routing, because the rule either fires or it does not.
Why over 80 percent empty is a schema problem rather than a data problem
166 of those fields held no value on any record in the sample. They were created and never used once.
That distinction decides what you do next. A data quality problem is fixed by filling fields, by enrichment, by process, by asking reps to maintain something. A schema problem is fixed by deleting fields, and no amount of enrichment touches it.
The empty ones cost you in the field picker, where they sit next to the fields that work and look identical. Somebody building a list picks one because its name matches what they need, gets a small result or an empty one, and has no way to tell whether that is the market or the field.
What to do instead
- Read the fill rate of a field before anything gets built on it. One query, and it prevents a rule that quietly never fires.
- Sample deliberately, and say how. Oldest records measure legacy data. Newest records measure the current model. Either is fine as long as the report says which one it is.
- Publish the distribution, not the headline. Over 80 percent empty is a number people agree with. 57 usable fields, 166 that have never held a value, and three routing fields under four percent is a number somebody has to decide about.
- Delete the fields that have never held a value. They are not free, because they sit in the picker beside the ones that work.
- Check every rule you have written against the coverage of the field it names. A rule pointing at an empty field reads exactly like a rule that works.
The consequence of building on a field nobody validated shows up one layer up, where a market coverage claim rested on a substitute field. And the contact side of the same CRM produced its own version, where companies that scored as a good fit had nobody on the record.