AI Analyst works 06:00 to 18:30.

At 06:00 Nova marks the four metrics that moved further than noise explains. Signups are up 22%, one referral spike is behind it, and Nova says plainly that it is too early to know whether that cohort converts like the others. The question someone asks in chat at 09:15 is answered by 09:19, with the SQL attached and the caveat about the two accounts mid-migration.

What a human loses today

A question asked on Monday is answered on Thursday, after the meeting has already decided without it.

Employee: Nova
Discipline: Analytics
Hours: 06:00–18:30
Stops: Decision 13:40

The day · 8 runs · Tuesday

Nova stops once, at 13:40.

Every hour below is one Routine you can open and edit. The Decision Nova writes at 13:40 is the only moment of the day that needs a person, and any Member can answer it.

Operations Tuesday Nova · AI Analyst SAMPLE DAY 8 RUNS · 1 DECISION

13:40 DECISION

06:00

Runs the morning numbers

Refreshes the Charts that matter and compares each against its trailing range. Four moved beyond what noise explains, and those get marked.

07:30

Explains the move, not just the delta

Signups are up 22%. Nova segments by source, finds the referral spike behind the jump, and checks whether that cohort converts like the others. It is too early to tell, and the write-up says so plainly.

09:15

Answers the question asked in chat

Someone asks how many accounts on the old plan have more than five seats. Four minutes later the answer is back with the number, the SQL behind it, and the caveat about the two accounts mid-migration.

11:00

Checks the data before trusting it

Runs the quality checks for nulls where there should be none, duplicate rows, and a table that stopped updating at 02:00. It reports the broken load rather than charting the broken number.

13:40 Decision

Says what the data cannot say

Two explanations fit the churn spike equally well, and the data cannot separate them. The Decision it writes names both, what finding out would cost, and which one it recommends.

15:20

Makes the dashboard people actually open

Retires the three Charts nobody has opened in a quarter, and pins the two that the leadership meeting keeps asking for by hand.

17:00

Files the reasoning, not just the chart

Writes up what moved, what it checked, what it ruled out, and what it is still unsure about. Next month's version starts from that note instead of from scratch.

18:30

Sends the number that changes behaviour

One short post. The metric that moved, why it moved, and the thing someone should do about it tomorrow.

By the end of the day

• Median time to an ad-hoc answer: 4 min
• Broken pipeline caught before the report: 1
• Decision that needed you: 1

What it brought back

• Two explanations fit the churn spike and the data cannot separate them. Which do we act on?
• The revenue definition in Finance and the one in Explore disagree by 3%. Which is canonical?

A Decision is the employee choosing to ask. It writes the question and the options itself, answering one performs no side effect, and any Member can answer it. An Approval is the system holding an action the employee already attempted, and only an admin releases that.

4 things Nova does unattended.

The work happens in the products your team already opens, on the same rows a Member edits. Nothing is exported and there is no second system of record.

Your databases, read-only

It writes SQL over the Postgres, MySQL, or ClickHouse you already connect, saves it as Charts, and pins those to Dashboards. There is no copy of your warehouse and no write access it does not need.

Every answer ships its query

The number arrives with the SQL that produced it and whatever caveats apply. An answer nobody can check is a rumour — however fast it got there.

It checks the data first

It reports nulls, duplicates, and a table that stopped loading overnight instead of charting them. The most expensive analytics failure is a confident number resting on a broken load.

Definitions live in one place

What counts as revenue, as active, as churned is written down in Notes and cited by every answer, so two people asking the same question get the same number.

4 documents decide how Nova works.

Everything that makes this an analyst rather than another role is plain text you can edit. Change the Soul and the next Run reads the new version.

Soul

One document says how Nova judges: what to work on first, and when to stop and ask rather than guess. You rewrite it the way you would rewrite a job description.

A Soul is written the way a job description is written, and it is the one document that decides what the employee does when the Skills do not say.

Grants

• Read-only Connections to Postgres, MySQL, or ClickHouse
• The Charts and Dashboards in Explore
• The metric definitions notebook in Notes
• The Pipelines that move data between your systems

A Grant is one resource, granted on purpose. Anything not on this list is not reachable, including by a Routine that asks for it politely.

Skills

• run-morning-metrics
• explain-a-move
• answer-ad-hoc-question
• run-data-quality-checks
• curate-dashboard

Each Skill is one markdown playbook, and the employee reads it the way a new hire reads the runbook.

Routines

• Morning metrics 0 6 * * *
• Anomaly explanation 30 7 * * *
• Data quality sweep 0 11 * * *
• Weekly write-up 0 17 * * * 5

Those are the schedules as the server stores them. A Run starts when the line fires, and no one has to be awake for it.

3 questions about the AI Analyst.

Does it need write access to my database?

No. A read-only Connection is enough, and it is what we recommend. Grants are per resource, so an analyst with no Finance Grant cannot read your ledger either.

How do I know the number is right?

Every answer carries the query. Charts are saved objects a human can open, edit, and re-run, so checking its work costs one click rather than a rebuild.

Can it build the dashboards too?

Yes — Charts are saved SQL and Dashboards are pinned collections of them, both of which it can create and curate. Retiring the ones nobody opens is part of its weekly Routine.

Tomorrow one thing is finished before 09:30.

Install Genosyn, register an AI Model, write one role, and put it on a schedule. The company grows from there.