Key-Person Risk
How much of your team's working output comes from a single analyst.
In one estate we assessed, a single analyst authored 34 percent of all successful queries over 365 days. Usually the strongest one, the person everyone routes questions to. Above 25 percent, that is a single point of failure.
You already know which person on your team this is. What you do not know is how much of the estate runs through them. Key-Person Risk puts a share on it: the proportion of clean, used queries written by one analyst, and where the reusable logic everyone else runs is densest.
You have the name. You do not have the number. You can, by the end of the day, off your own query history, for free.
What it tells us
A third of the output riding on one person does not mean the team is weak. It means the reusable part of the work, the patterns and definitions everyone else runs, has clustered around one node. The person carrying that share is your strongest asset.
The exposure is that the reasoning behind that work is reachable only through them. Two weeks of PTO can freeze a third of your definitions. The answer is not to slow that person down. You capture what they have already worked out so the rest of the team can build on it, with their name attached.
What we found
The raw percentage is the least of it. Add analysts and the top share dilutes on its own, until the number looks like a solved problem. It isn't. The unit just moves up: from one analyst to one pod, or to one departed principal whose logic dozens of people still copy without re-deriving it. When that principal walks out, the reasoning evaporates with them, and what everyone keeps copying is a concentrated definition no one left can explain.
The clearest case we see looks like a contradiction at first. One definition, authored by one person, that the entire team depends on. Both halves are readable straight off the history. The dependence is visible because the same block of logic appears, copied and unchanged, in dozens of other analysts' queries. The ownership is visible because only one person has ever written or modified it.
SELECT patient_id
FROM fills
GROUP BY patient_id
HAVING SUM(days_supply)
/ 365.0 >= 0.80
How the team calculates adherence. A fixed 365 day denominator rather than the observed enrollment window, and a cutoff at 80 percent. One analyst wrote both choices. The history shows no one else has ever touched them.
WITH adherent AS (
/* pasted, character for character */
SELECT patient_id FROM fills
GROUP BY patient_id
HAVING SUM(days_supply)/365.0 >= 0.80
)
SELECT * FROM adherent JOIN new_cohort ...
Someone else's study, with the same block pasted in front of it. The cohort around it changes every time. The definition itself never does, because nobody else knows which parts are safe to change.
So the two facts sit together without conflict. The team inherited a copy and can run it. The author is the only one who can correctly change it, because the reasoning behind the 365 days and the 80 percent was never written down. Ask for the same measure on a six month study and everyone else is guessing at which number moves. That gap, between running a definition and reasoning about one, is what the 34 percent is really measuring. The reasoning does exist, in a Slack reply from 2021. Findable, in the way that anything is findable.
Why it matters to you
A third of your definitions depend on one person. Two weeks of PTO freezes them, and no amount of documentation you have today changes that.
Key-person risk you can finally put a number on. Not a worry in the abstract, a share of authored output you can watch and act on.
Your best analyst becomes a bottleneck, because their method travels as a copy and not as a method. Every new study routes back through them instead of building on what they already settled.
Continuity risk that no succession plan on paper actually covers. The seat is backfilled; the reasoning that made the seat valuable is not.
What good looks like
The top contributor's reasoning is written down beside their SQL, so the team and the agent can adapt it instead of pasting it forward and hoping. The next analyst can see why the denominator is a year, and change it on purpose when the study calls for it. What was in one person's head becomes an institutional knowledge asset, a set of files that people and machines can both read, with the author credited on every definition they settled. The person stays exactly as valuable, without being a single point of failure. The settled part runs without them. The judgment for a question nobody has asked yet still needs the human, and that is what you want them spending their time on.
What to do about it
- Run Neuron on your query history.
- Let it surface the definitions your top contributor authored, with a count of how many other people copy each one.
- Review the top few with that person, so the reasoning behind each threshold gets captured next to the SQL.
- Publish those definitions and their owners into the semantic model, so your BI tools and your agent read the same thing.
- Pull 365 days of query authorship.
- Compute the top contributor's share of clean, used queries.
- Find the blocks of their logic that reappear, copied, in other people's queries. Those are your dependencies.
- Sit with the author, write down why each threshold is what it is, and save it where the team and an AI tool can both read it.